Tan, George S. Q.; Miller, Lynn; Wade, Sam; Ilomäki, Jenni; Lukose, Dickson; Rong, Jia; Webb, Geoffrey I.
Clinical Pharmacology and Therapeutics, vol. 119, no. 1, pp. 228-240, 2026, ISSN: 1532-6535.
@article{Tan2025a,
title = {Association Discovery Approach in Healthcare Big Data to Identify Drug Safety and Drug Repurposing Signals},
author = {George S. Q. Tan and Lynn Miller and Sam Wade and Jenni Ilom\"{a}ki and Dickson Lukose and Jia Rong and Geoffrey I. Webb},
doi = {10.1002/cpt.70080},
issn = {1532-6535},
year = {2026},
date = {2026-10-01},
journal = {Clinical Pharmacology and Therapeutics},
volume = {119},
number = {1},
pages = {228-240},
publisher = {Wiley},
abstract = {Data science approaches have been increasingly implemented in healthcare big data to evaluate the safety and effectiveness of drugs. Association discovery is a data mining approach that finds potentially associated elements in high-dimensional data. We present a novel implementation of the association discovery approach in longitudinal healthcare data to identify drug safety and drug repurposing signals from positive and inverse associations between drug use and clinical outcomes, respectively. We used the 10% sample data from the Australian Pharmaceutical Benefits Scheme (2014\textendash2024), which comprises prescription claims records. Using the Magnum Opus association discovery tool, we identified associations between a wide range of drugs and three common chronic medical conditions (i.e., coronary artery disease, type 2 diabetes, epilepsy). Cases with the conditions were identified using the supply of indicator drug(s) as a proxy for the conditions and matched to controls not supplied the indicator drug(s). Drug use was defined using Anatomical Therapeutic Chemical (ATC) codes supplied during a one-year lookback period before the supply of the indicator drug(s). We also evaluated combinations of up to four drugs and identified associations at the ATC level of drug class(es). In this study, we reproduced several known adverse drug events and protective drug effects, while some associations were attributable to confounding, mutual indications, or reverse causation. The remaining associations may represent previously uncharacterized drug safety and drug repurposing signals, necessitating further validation. We also discussed methodological differences between this association discovery approach and other similar data science approaches.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Data science approaches have been increasingly implemented in healthcare big data to evaluate the safety and effectiveness of drugs. Association discovery is a data mining approach that finds potentially associated elements in high-dimensional data. We present a novel implementation of the association discovery approach in longitudinal healthcare data to identify drug safety and drug repurposing signals from positive and inverse associations between drug use and clinical outcomes, respectively. We used the 10% sample data from the Australian Pharmaceutical Benefits Scheme (2014–2024), which comprises prescription claims records. Using the Magnum Opus association discovery tool, we identified associations between a wide range of drugs and three common chronic medical conditions (i.e., coronary artery disease, type 2 diabetes, epilepsy). Cases with the conditions were identified using the supply of indicator drug(s) as a proxy for the conditions and matched to controls not supplied the indicator drug(s). Drug use was defined using Anatomical Therapeutic Chemical (ATC) codes supplied during a one-year lookback period before the supply of the indicator drug(s). We also evaluated combinations of up to four drugs and identified associations at the ATC level of drug class(es). In this study, we reproduced several known adverse drug events and protective drug effects, while some associations were attributable to confounding, mutual indications, or reverse causation. The remaining associations may represent previously uncharacterized drug safety and drug repurposing signals, necessitating further validation. We also discussed methodological differences between this association discovery approach and other similar data science approaches.
Dempster, Angus; Webb, Geoffrey I.; Schmidt, Daniel F.
Machine Learning, vol. 115, no. 7, 2026, ISSN: 1573-0565.
@article{Dempster2026,
title = {Prevalidated Ridge Regression is a Highly-Efficient Drop-In Replacement for Logistic Regression for High-dimensional Data},
author = {Angus Dempster and Geoffrey I. Webb and Daniel F. Schmidt},
doi = {10.1007/s10994-026-07059-1},
issn = {1573-0565},
year = {2026},
date = {2026-06-01},
journal = {Machine Learning},
volume = {115},
number = {7},
publisher = {Springer Science and Business Media LLC},
abstract = {Logistic regression is a ubiquitous method for probabilistic classification. However, the effectiveness of logistic regression depends upon careful and relatively computationally expensive tuning, especially for the regularisation hyperparameter, and especially in the context of high-dimensional data. We present a prevalidated ridge regression model that in practice closely matches logistic regression in terms of 0\textendash1 loss and log-loss, particularly for high-dimensional data, while being significantly more computationally efficient and having no user-tuned hyperparameters (the regularisation hyperparameter is learned automatically as part of the fitting process). We scale the coefficients of the model so as to minimise log-loss for a set of prevalidated predictions derived from the estimated leave-one-out cross-validation error. This exploits quantities already computed in the course of fitting the ridge regression model in order to find the scaling parameter with nominal additional computation.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Logistic regression is a ubiquitous method for probabilistic classification. However, the effectiveness of logistic regression depends upon careful and relatively computationally expensive tuning, especially for the regularisation hyperparameter, and especially in the context of high-dimensional data. We present a prevalidated ridge regression model that in practice closely matches logistic regression in terms of 0–1 loss and log-loss, particularly for high-dimensional data, while being significantly more computationally efficient and having no user-tuned hyperparameters (the regularisation hyperparameter is learned automatically as part of the fitting process). We scale the coefficients of the model so as to minimise log-loss for a set of prevalidated predictions derived from the estimated leave-one-out cross-validation error. This exploits quantities already computed in the course of fitting the ridge regression model in order to find the scaling parameter with nominal additional computation.
Qu, Yanda; Currie, Graham; Webb, Geoffrey I.
“Accessibility misrepresentation” in public transport
Journal of Transport Geography, vol. 133, 2026, ISSN: 0966-6923.
@article{Qu2026,
title = {“Accessibility misrepresentation” in public transport},
author = {Yanda Qu and Graham Currie and Geoffrey I. Webb},
doi = {10.1016/j.jtrangeo.2026.104632},
issn = {0966-6923},
year = {2026},
date = {2026-05-01},
journal = {Journal of Transport Geography},
volume = {133},
publisher = {Elsevier BV},
abstract = {Location-based accessibility is widely assessed using zonal frameworks that aggregate high-resolution results into summary statistics. This aggregation-induced error is often assumed to be negligible, yet this assumption is rarely verified and may obscure critical disparities. This study introduces “Accessibility Misrepresentation” as the deviation between high-resolution accessibility and zonal proxies. Using building-level public transport accessibility in Metropolitan Melbourne, the research evaluates the magnitude, spatial patterns, and equity implications of misrepresentation. Results show systematic variation across travel time thresholds, built environment characteristics, aggregation methods, and service levels. Missed detection of populations without accessibility peaks at 15-20 min thresholds commonly used in neighbourhood accessibility policies, predominantly affecting densely populated middle suburbs where public transport catchments misalign with administrative boundaries. Continuous errors in accessibility values show no systematic directional bias; however, minorities can experience substantial misstatements sufficient to alter normative assessments. Equity analysis reveals disproportionate impacts on middle-status populations often overlooked by conventional frameworks. Population-weighted medians substantially outperform means in reducing misrepresentation. However, no single statistic adequately captures internal variation, particularly with multimodal distributions. A dual-scale approach is recommended: high-resolution computation to characterise within-zone distributions, combined with zonal summaries accompanied by distributional diagnostics, including share of residents without accessibility and upper-tail error indicators. This approach enhances transparency, supports equity-focused and defensible planning decisions without costing additional resources.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Location-based accessibility is widely assessed using zonal frameworks that aggregate high-resolution results into summary statistics. This aggregation-induced error is often assumed to be negligible, yet this assumption is rarely verified and may obscure critical disparities. This study introduces “Accessibility Misrepresentation” as the deviation between high-resolution accessibility and zonal proxies. Using building-level public transport accessibility in Metropolitan Melbourne, the research evaluates the magnitude, spatial patterns, and equity implications of misrepresentation. Results show systematic variation across travel time thresholds, built environment characteristics, aggregation methods, and service levels. Missed detection of populations without accessibility peaks at 15-20 min thresholds commonly used in neighbourhood accessibility policies, predominantly affecting densely populated middle suburbs where public transport catchments misalign with administrative boundaries. Continuous errors in accessibility values show no systematic directional bias; however, minorities can experience substantial misstatements sufficient to alter normative assessments. Equity analysis reveals disproportionate impacts on middle-status populations often overlooked by conventional frameworks. Population-weighted medians substantially outperform means in reducing misrepresentation. However, no single statistic adequately captures internal variation, particularly with multimodal distributions. A dual-scale approach is recommended: high-resolution computation to characterise within-zone distributions, combined with zonal summaries accompanied by distributional diagnostics, including share of residents without accessibility and upper-tail error indicators. This approach enhances transparency, supports equity-focused and defensible planning decisions without costing additional resources.
Wang, Xiaoyu; Pan, Tong; Chen, Sihan; Webb, Geoffrey I.; Jiang, Yunzhe; Rozowsky, Joel; Gerstein, Mark; Song, Jiangning
Genome Biology, vol. 27, 2026, ISSN: 1474-760X.
@article{Wang2026,
title = {Predicting disease-specific histone modifications and functional effects of non-coding variants by leveraging DNA language models},
author = {Xiaoyu Wang and Tong Pan and Sihan Chen and Geoffrey I. Webb and Yunzhe Jiang and Joel Rozowsky and Mark Gerstein and Jiangning Song},
doi = {10.1186/s13059-026-04003-3},
issn = {1474-760X},
year = {2026},
date = {2026-02-01},
journal = {Genome Biology},
volume = {27},
publisher = {Springer Science and Business Media LLC},
abstract = {Background
Epigenetic modifications play a vital role in the pathogenesis of human diseases, particularly neurodegenerative disorders such as Alzheimer's disease, where dysregulated histone modifications are strongly implicated in disease mechanisms. While recent advances underscore the importance of accurately identifying these modifications to elucidate their contribution to Alzheimer's disease pathology, existing computational methods remain limited by their generic approaches that overlook disease-specific epigenetic signatures.
Results
To bridge this gap, we develop a novel large language model-based deep learning framework tailored for disease-contextual prediction of histone modifications and variant effects. Focusing on Alzheimer's disease as a case study, we integrate epigenomic data from multiple patient samples to construct a comprehensive, disease-specific histone modification dataset, enabling our model to learn Alzheimer's disease -associated molecular signatures. A key innovation of our approach is the incorporation of a Mixture of Experts architecture, which effectively distinguishes between disease and healthy epigenetic states, allowing for precise identification of Alzheimer's disease -relevant epigenetic modification patterns. Our model demonstrates robust performance in disease-specific histone modification prediction, significantly outperforming existing state-of-the-art methods that lack disease context. Beyond accurate modification site prediction, our framework provides important biological insights by successfully prioritizing Alzheimer's disease-associated genetic variants, which show significant enrichment in disease-relevant pathways.
Conclusions
Our framework establishes a powerful new paradigm for epigenetic research that can be extended to other complex diseases, offering both a valuable tool for variant effect interpretation and a promising strategy for uncovering novel disease mechanisms through epigenetic profiling.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Background
Epigenetic modifications play a vital role in the pathogenesis of human diseases, particularly neurodegenerative disorders such as Alzheimer's disease, where dysregulated histone modifications are strongly implicated in disease mechanisms. While recent advances underscore the importance of accurately identifying these modifications to elucidate their contribution to Alzheimer's disease pathology, existing computational methods remain limited by their generic approaches that overlook disease-specific epigenetic signatures.
Results
To bridge this gap, we develop a novel large language model-based deep learning framework tailored for disease-contextual prediction of histone modifications and variant effects. Focusing on Alzheimer's disease as a case study, we integrate epigenomic data from multiple patient samples to construct a comprehensive, disease-specific histone modification dataset, enabling our model to learn Alzheimer's disease -associated molecular signatures. A key innovation of our approach is the incorporation of a Mixture of Experts architecture, which effectively distinguishes between disease and healthy epigenetic states, allowing for precise identification of Alzheimer's disease -relevant epigenetic modification patterns. Our model demonstrates robust performance in disease-specific histone modification prediction, significantly outperforming existing state-of-the-art methods that lack disease context. Beyond accurate modification site prediction, our framework provides important biological insights by successfully prioritizing Alzheimer's disease-associated genetic variants, which show significant enrichment in disease-relevant pathways.
Conclusions
Our framework establishes a powerful new paradigm for epigenetic research that can be extended to other complex diseases, offering both a valuable tool for variant effect interpretation and a promising strategy for uncovering novel disease mechanisms through epigenetic profiling.
Epigenetic modifications play a vital role in the pathogenesis of human diseases, particularly neurodegenerative disorders such as Alzheimer's disease, where dysregulated histone modifications are strongly implicated in disease mechanisms. While recent advances underscore the importance of accurately identifying these modifications to elucidate their contribution to Alzheimer's disease pathology, existing computational methods remain limited by their generic approaches that overlook disease-specific epigenetic signatures.
Results
To bridge this gap, we develop a novel large language model-based deep learning framework tailored for disease-contextual prediction of histone modifications and variant effects. Focusing on Alzheimer's disease as a case study, we integrate epigenomic data from multiple patient samples to construct a comprehensive, disease-specific histone modification dataset, enabling our model to learn Alzheimer's disease -associated molecular signatures. A key innovation of our approach is the incorporation of a Mixture of Experts architecture, which effectively distinguishes between disease and healthy epigenetic states, allowing for precise identification of Alzheimer's disease -relevant epigenetic modification patterns. Our model demonstrates robust performance in disease-specific histone modification prediction, significantly outperforming existing state-of-the-art methods that lack disease context. Beyond accurate modification site prediction, our framework provides important biological insights by successfully prioritizing Alzheimer's disease-associated genetic variants, which show significant enrichment in disease-relevant pathways.
Conclusions
Our framework establishes a powerful new paradigm for epigenetic research that can be extended to other complex diseases, offering both a valuable tool for variant effect interpretation and a promising strategy for uncovering novel disease mechanisms through epigenetic profiling.
Pan, Tong; Cui, Xin; Koh, Huan Yee; Bi, Yue; Wang, Xiaoyu; Zhang, Yumeng; Hu, Shantong; Webb, Geoffrey I.; Kurgan, Lukasz; Zhang, Guimin; Song, Jiangning
A geometric deep learning framework for genome-wide prediction of enzyme turnover number
Genome Biology, vol. 27, no. 1, 2026, ISSN: 1474-760X.
@article{Pan2026,
title = {A geometric deep learning framework for genome-wide prediction of enzyme turnover number},
author = {Tong Pan and Xin Cui and Huan Yee Koh and Yue Bi and Xiaoyu Wang and Yumeng Zhang and Shantong Hu and Geoffrey I. Webb and Lukasz Kurgan and Guimin Zhang and Jiangning Song},
doi = {10.1186/s13059-026-03986-3},
issn = {1474-760X},
year = {2026},
date = {2026-01-01},
journal = {Genome Biology},
volume = {27},
number = {1},
publisher = {Springer Science and Business Media LLC},
abstract = {Background
Enzyme turnover numbers are fundamental kinetic constants that quantify enzymatic efficiency. Systematic studies of
are essential for characterizing the mechanisms underlying proteomic composition and cellular metabolism. However, experimental measurements of
remain limited and prone to noise.
Results
To address this, we present KcatNet, a geometric deep learning model designed for high-throughput prediction of
in metabolic enzymes across all organisms, leveraging paired enzyme sequence and substrate representations. KcatNet consistently outperforms existing predictors, particularly for enzymes with high catalytic efficiency, and demonstrates strong generalization to enzymes that are dissimilar to those in the training set. Furthermore, KcatNet uncovers structural mechanisms and interaction patterns within enzyme\textendashsubstrate complexes, providing insights into architectural principles that are inaccessible with existing methods by harnessing the representational power of large-scale protein language models. We apply KcatNet to genome-scale
prediction across diverse yeast species, improving proteome allocation predictions by integrating its outputs into metabolic models. Experimental validation confirms the model's ability to identify enzyme mutants with enhanced activity.
Conclusion
By bridging the gap between sequence, structure, and function, KcatNet establishes a robust foundation for advancing understanding of molecular-level mechanisms and accelerating enzyme engineering efforts.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Background
Enzyme turnover numbers are fundamental kinetic constants that quantify enzymatic efficiency. Systematic studies of
are essential for characterizing the mechanisms underlying proteomic composition and cellular metabolism. However, experimental measurements of
remain limited and prone to noise.
Results
To address this, we present KcatNet, a geometric deep learning model designed for high-throughput prediction of
in metabolic enzymes across all organisms, leveraging paired enzyme sequence and substrate representations. KcatNet consistently outperforms existing predictors, particularly for enzymes with high catalytic efficiency, and demonstrates strong generalization to enzymes that are dissimilar to those in the training set. Furthermore, KcatNet uncovers structural mechanisms and interaction patterns within enzyme–substrate complexes, providing insights into architectural principles that are inaccessible with existing methods by harnessing the representational power of large-scale protein language models. We apply KcatNet to genome-scale
prediction across diverse yeast species, improving proteome allocation predictions by integrating its outputs into metabolic models. Experimental validation confirms the model's ability to identify enzyme mutants with enhanced activity.
Conclusion
By bridging the gap between sequence, structure, and function, KcatNet establishes a robust foundation for advancing understanding of molecular-level mechanisms and accelerating enzyme engineering efforts.
Enzyme turnover numbers are fundamental kinetic constants that quantify enzymatic efficiency. Systematic studies of
are essential for characterizing the mechanisms underlying proteomic composition and cellular metabolism. However, experimental measurements of
remain limited and prone to noise.
Results
To address this, we present KcatNet, a geometric deep learning model designed for high-throughput prediction of
in metabolic enzymes across all organisms, leveraging paired enzyme sequence and substrate representations. KcatNet consistently outperforms existing predictors, particularly for enzymes with high catalytic efficiency, and demonstrates strong generalization to enzymes that are dissimilar to those in the training set. Furthermore, KcatNet uncovers structural mechanisms and interaction patterns within enzyme–substrate complexes, providing insights into architectural principles that are inaccessible with existing methods by harnessing the representational power of large-scale protein language models. We apply KcatNet to genome-scale
prediction across diverse yeast species, improving proteome allocation predictions by integrating its outputs into metabolic models. Experimental validation confirms the model's ability to identify enzyme mutants with enhanced activity.
Conclusion
By bridging the gap between sequence, structure, and function, KcatNet establishes a robust foundation for advancing understanding of molecular-level mechanisms and accelerating enzyme engineering efforts.
Qu, Yanda; Currie, Graham; Webb, Geoffrey I.
Spatial resolution impact on public transport accessibility measurement error
Journal of Transport Geography, vol. 129, 2025, ISSN: 0966-6923.
@article{Qu2025,
title = {Spatial resolution impact on public transport accessibility measurement error},
author = {Yanda Qu and Graham Currie and Geoffrey I. Webb},
doi = {10.1016/j.jtrangeo.2025.104408},
issn = {0966-6923},
year = {2025},
date = {2025-12-01},
journal = {Journal of Transport Geography},
volume = {129},
publisher = {Elsevier BV},
abstract = {Place-based accessibility measures are critical to transport planning. However, their accuracy can vary greatly depending on the spatial resolution of representative points used in calculations. This study refers to this precision issue as measurement resolution error, and its impacts have not yet been comprehensively explored in public transport accessibility analysis. Using Melbourne, Australia, as the study area, this study examined measurement resolution error across various spatial resolutions, public transport service levels, opportunity densities, types of accessibility measures and time thresholds. Results reveal that while the errors are generally small, their magnitude is highly context dependent. Smaller time thresholds, lower service levels, opportunity density that leads to lower accessibility values, and more complex accessibility measures all increase sensitivity to this error. Suburban areas with sparsely populated mixed land use that are served with modest public transport services are the most vulnerable to measurement resolution error, creating significant transport equity implications. The study recommends the use of data at resolutions that can reflect land use information for general public transport accessibility analysis, while building-level resolution data should be selectively applied if analysis objectives warrant such granularity. These findings highlight the importance of being aware of measurement resolution errors and provide evidence-based guidance for balancing analytical precision with computational efficiency in transport planning.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Place-based accessibility measures are critical to transport planning. However, their accuracy can vary greatly depending on the spatial resolution of representative points used in calculations. This study refers to this precision issue as measurement resolution error, and its impacts have not yet been comprehensively explored in public transport accessibility analysis. Using Melbourne, Australia, as the study area, this study examined measurement resolution error across various spatial resolutions, public transport service levels, opportunity densities, types of accessibility measures and time thresholds. Results reveal that while the errors are generally small, their magnitude is highly context dependent. Smaller time thresholds, lower service levels, opportunity density that leads to lower accessibility values, and more complex accessibility measures all increase sensitivity to this error. Suburban areas with sparsely populated mixed land use that are served with modest public transport services are the most vulnerable to measurement resolution error, creating significant transport equity implications. The study recommends the use of data at resolutions that can reflect land use information for general public transport accessibility analysis, while building-level resolution data should be selectively applied if analysis objectives warrant such granularity. These findings highlight the importance of being aware of measurement resolution errors and provide evidence-based guidance for balancing analytical precision with computational efficiency in transport planning.
Zheng, Yizhen; Koh, Huan Yee; Ju, Jiaxin; Yang, Madeleine; May, Lauren T.; Webb, Geoffrey I.; Li, Li; Pan, Shirui; Church, George
Large language models for drug discovery and development
Patterns, vol. 6, no. 10, 2025, ISSN: 2666-3899.
@article{Zheng2025a,
title = {Large language models for drug discovery and development},
author = {Yizhen Zheng and Huan Yee Koh and Jiaxin Ju and Madeleine Yang and Lauren T. May and Geoffrey I. Webb and Li Li and Shirui Pan and George Church},
doi = {10.1016/j.patter.2025.101346},
issn = {2666-3899},
year = {2025},
date = {2025-10-01},
journal = {Patterns},
volume = {6},
number = {10},
publisher = {Elsevier BV},
abstract = {The integration of large language models (LLMs) into the drug discovery and development field marks a sig
nificant paradigm shift, offering novel methodologies for understanding disease mechanisms, facilitating de
novo drug discovery, and optimizing clinical trial processes. This review highlights the expanding role of
LLMs in revolutionizing various stages of the drug development pipeline. We investigate how these advanced
computational models can uncover target-disease linkage, interpret complex biomedical data, enhance drug
molecule design, predict drug efficacy and safety profiles, and facilitate clinical trial processes. In this paper,
we aim to provide a comprehensive overview for researchers and practitioners in computational biology,
pharmacology, and AI4Science by offering insights into the potential transformative impact of LLMs on
drug discovery and development},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The integration of large language models (LLMs) into the drug discovery and development field marks a sig
nificant paradigm shift, offering novel methodologies for understanding disease mechanisms, facilitating de
novo drug discovery, and optimizing clinical trial processes. This review highlights the expanding role of
LLMs in revolutionizing various stages of the drug development pipeline. We investigate how these advanced
computational models can uncover target-disease linkage, interpret complex biomedical data, enhance drug
molecule design, predict drug efficacy and safety profiles, and facilitate clinical trial processes. In this paper,
we aim to provide a comprehensive overview for researchers and practitioners in computational biology,
pharmacology, and AI4Science by offering insights into the potential transformative impact of LLMs on
drug discovery and development
nificant paradigm shift, offering novel methodologies for understanding disease mechanisms, facilitating de
novo drug discovery, and optimizing clinical trial processes. This review highlights the expanding role of
LLMs in revolutionizing various stages of the drug development pipeline. We investigate how these advanced
computational models can uncover target-disease linkage, interpret complex biomedical data, enhance drug
molecule design, predict drug efficacy and safety profiles, and facilitate clinical trial processes. In this paper,
we aim to provide a comprehensive overview for researchers and practitioners in computational biology,
pharmacology, and AI4Science by offering insights into the potential transformative impact of LLMs on
drug discovery and development
Abourayya, Amr; Kleesiek, Jens; Rao, Kanishka; Ayday, Erman; Rao, Bharat; Webb, Geoffrey I.; Kamp, Michael
Little Is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning
Proceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 15, pp. 15293-15301, 2025, ISSN: 2159-5399.
@article{Abourayya2025,
title = {Little Is Enough: Boosting Privacy by Sharing Only Hard Labels in Federated Semi-Supervised Learning},
author = {Amr Abourayya and Jens Kleesiek and Kanishka Rao and Erman Ayday and Bharat Rao and Geoffrey I. Webb and Michael Kamp},
doi = {10.1609/aaai.v39i15.33678},
issn = {2159-5399},
year = {2025},
date = {2025-04-01},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {39},
number = {15},
pages = {15293-15301},
publisher = {Association for the Advancement of Artificial Intelligence (AAAI)},
abstract = {In many critical applications, sensitive data is inherently distributed and cannot be centralized due to privacy concerns. A wide range of federated learning approaches have been proposed to train models locally at each client without sharing their sensitive data, typically by exchanging model parameters, or probabilistic predictions (soft labels) on a public dataset or a combination of both. However, these methods still disclose private information and restrict local models to those that can be trained using gradient-based methods. We propose a federated co-training (FEDCT) approach that improves privacy by sharing only definitive (hard) labels on a public unlabeled dataset. Clients use a consensus of these shared labels as pseudo-labels for local training. This federated co-training approach empirically enhances privacy without compromising model quality. In addition, it allows the use of local models that are not suitable for parameter aggregation in traditional federated learning, such as gradient-boosted decision trees, rule ensembles, and random forests. Furthermore, we observe that FEDCT performs effectively in federated fine-tuning of large language models, where its pseudo-labeling mechanism is particularly beneficial. Empirical evaluations and theoretical analyses suggest its applicability across a range of federated learning scenarios.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In many critical applications, sensitive data is inherently distributed and cannot be centralized due to privacy concerns. A wide range of federated learning approaches have been proposed to train models locally at each client without sharing their sensitive data, typically by exchanging model parameters, or probabilistic predictions (soft labels) on a public dataset or a combination of both. However, these methods still disclose private information and restrict local models to those that can be trained using gradient-based methods. We propose a federated co-training (FEDCT) approach that improves privacy by sharing only definitive (hard) labels on a public unlabeled dataset. Clients use a consensus of these shared labels as pseudo-labels for local training. This federated co-training approach empirically enhances privacy without compromising model quality. In addition, it allows the use of local models that are not suitable for parameter aggregation in traditional federated learning, such as gradient-boosted decision trees, rule ensembles, and random forests. Furthermore, we observe that FEDCT performs effectively in federated fine-tuning of large language models, where its pseudo-labeling mechanism is particularly beneficial. Empirical evaluations and theoretical analyses suggest its applicability across a range of federated learning scenarios.
Zheng, Yizhen; Koh, Huan Yee; Ju, Jiaxin; Nguyen, Anh T. N.; May, Lauren T.; Webb, Geoffrey I.; Pan, Shirui
Large language models for scientific discovery in molecular property prediction
Nature Machine Intelligence, vol. 7, pp. 437–447, 2025, ISSN: 2522-5839.
@article{Zheng2025,
title = {Large language models for scientific discovery in molecular property prediction},
author = {Yizhen Zheng and Huan Yee Koh and Jiaxin Ju and Anh T. N. Nguyen and Lauren T. May and Geoffrey I. Webb and Shirui Pan},
doi = {10.1038/s42256-025-00994-z},
issn = {2522-5839},
year = {2025},
date = {2025-02-01},
journal = {Nature Machine Intelligence},
volume = {7},
pages = {437\textendash447},
publisher = {Springer Science and Business Media LLC},
abstract = {Large language models (LLMs) are a form of artificial intelligence system encapsulating vast knowledge in the form of natural language. These systems are adept at numerous complex tasks including creative writing, storytelling, translation, question-answering, summarization and computer code generation. Although LLMs have seen initial applications in natural sciences, their potential for driving scientific discovery remains largely unexplored. In this work, we introduce LLM4SD, a framework designed to harness LLMs for driving scientific discovery in molecular property prediction by synthesizing knowledge from literature and inferring knowledge from scientific data. LLMs synthesize knowledge by extracting established information from scientific literature, such as molecular weight being key to predicting solubility. For inference, LLMs identify patterns in molecular data, particularly in Simplified Molecular Input Line Entry System-encoded structures, such as halogen-containing molecules being more likely to cross the blood\textendashbrain barrier. This information is presented as interpretable knowledge, enabling the transformation of molecules into feature vectors. By using these features with interpretable models such as random forest, LLM4SD can outperform the current state of the art across a range of benchmark tasks for predicting molecular properties. We foresee it providing interpretable and potentially new insights, aiding scientific discovery in molecular property prediction.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Large language models (LLMs) are a form of artificial intelligence system encapsulating vast knowledge in the form of natural language. These systems are adept at numerous complex tasks including creative writing, storytelling, translation, question-answering, summarization and computer code generation. Although LLMs have seen initial applications in natural sciences, their potential for driving scientific discovery remains largely unexplored. In this work, we introduce LLM4SD, a framework designed to harness LLMs for driving scientific discovery in molecular property prediction by synthesizing knowledge from literature and inferring knowledge from scientific data. LLMs synthesize knowledge by extracting established information from scientific literature, such as molecular weight being key to predicting solubility. For inference, LLMs identify patterns in molecular data, particularly in Simplified Molecular Input Line Entry System-encoded structures, such as halogen-containing molecules being more likely to cross the blood–brain barrier. This information is presented as interpretable knowledge, enabling the transformation of molecules into feature vectors. By using these features with interpretable models such as random forest, LLM4SD can outperform the current state of the art across a range of benchmark tasks for predicting molecular properties. We foresee it providing interpretable and potentially new insights, aiding scientific discovery in molecular property prediction.
Zhang, Huan; Jiang, Liangxiao; Zhang, Wenjun; Webb, Geoffrey I.
Dual-View Learning from Crowds
ACM Transactions on Knowledge Discovery from Data, vol. 19, no. 3, 2025, ISSN: 1556-472X.
@article{Zhang2025,
title = {Dual-View Learning from Crowds},
author = {Huan Zhang and Liangxiao Jiang and Wenjun Zhang and Geoffrey I. Webb},
doi = {10.1145/3712605},
issn = {1556-472X},
year = {2025},
date = {2025-02-01},
journal = {ACM Transactions on Knowledge Discovery from Data},
volume = {19},
number = {3},
publisher = {Association for Computing Machinery (ACM)},
abstract = {Crowdsourcing services provide a fast and cheap way to obtain substantial labeled data by employing crowd workers on the Internet. In crowdsourcing learning, two-stage methods have been widely used, which first infer the integrated label for each instance and then build a learning model using instances with their integrated labels. However, existing two-stage methods mainly focus on how to infer more accurate integrated labels, after that, most of them directly regard the integrated labels as class labels to build a learning model, which loses the detailed worker labeling information in multiple noisy labels and thus results in sub-optimal model accuracy. To solve this problem, in this study, we take the multiple noisy labels of each instance as its attribute value vector to construct another view in addition to the original attribute view, and propose a novel two-stage method called dual-view learning from crowds (DVLFC). In DVLFC, we first pick out workers with sufficient number of labels and augment the multiple noisy label set for each instance, then we build a supervised learning model in each view and at last we fuse their class-membership probabilities to get the final classification result. Extensive experiments on both real-world and artificial crowdsourced datasets prove the effectiveness of DVLFC.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Crowdsourcing services provide a fast and cheap way to obtain substantial labeled data by employing crowd workers on the Internet. In crowdsourcing learning, two-stage methods have been widely used, which first infer the integrated label for each instance and then build a learning model using instances with their integrated labels. However, existing two-stage methods mainly focus on how to infer more accurate integrated labels, after that, most of them directly regard the integrated labels as class labels to build a learning model, which loses the detailed worker labeling information in multiple noisy labels and thus results in sub-optimal model accuracy. To solve this problem, in this study, we take the multiple noisy labels of each instance as its attribute value vector to construct another view in addition to the original attribute view, and propose a novel two-stage method called dual-view learning from crowds (DVLFC). In DVLFC, we first pick out workers with sufficient number of labels and augment the multiple noisy label set for each instance, then we build a supervised learning model in each view and at last we fuse their class-membership probabilities to get the final classification result. Extensive experiments on both real-world and artificial crowdsourced datasets prove the effectiveness of DVLFC.
Darban, Zahra Zamanzadeh; Webb, Geoffrey I.; Pan, Shirui; Aggarwal, Charu C.; Salehi, Mahsa
CARLA: Self-supervised contrastive representation learning for time series anomaly detection
Pattern Recognition, vol. 157, 2025, ISSN: 0031-3203.
@article{Darban2025,
title = {CARLA: Self-supervised contrastive representation learning for time series anomaly detection},
author = {Zahra Zamanzadeh Darban and Geoffrey I. Webb and Shirui Pan and Charu C. Aggarwal and Mahsa Salehi},
doi = {10.1016/j.patcog.2024.110874},
issn = {0031-3203},
year = {2025},
date = {2025-01-01},
journal = {Pattern Recognition},
volume = {157},
abstract = {One main challenge in time series anomaly detection (TSAD) is the lack of labelled data in many real-life scenarios. Most of the existing anomaly detection methods focus on learning the normal behaviour of unlabelled time series in an unsupervised manner. The normal boundary is often defined tightly, resulting in slight deviations being classified as anomalies, consequently leading to a high false positive rate and a limited ability to generalise normal patterns. To address this, we introduce a novel end-to-end self-supervised ContrAstive Representation Learning approach for time series Anomaly detection (CARLA). While existing contrastive learning methods assume that augmented time series windows are positive samples and temporally distant windows are negative samples, we argue that these assumptions are limited as augmentation of time series can transform them to negative samples, and a temporally distant window can represent a positive sample. Existing approaches to contrastive learning for time series have directly copied methods developed for image analysis. We argue that these methods do not transfer well. Instead, our contrastive approach leverages existing generic knowledge about time series anomalies and injects various types of anomalies as negative samples. Therefore, CARLA not only learns normal behaviour but also learns deviations indicating anomalies. It creates similar representations for temporally close windows and distinct ones for anomalies. Additionally, it leverages the information about representations’ neighbours through a self-supervised approach to classify windows based on their nearest/furthest neighbours to further enhance the performance of anomaly detection. In extensive tests on seven major real-world TSAD datasets, CARLA shows superior performance (F1 and AU-PR) over state-of-the-art self-supervised, semi-supervised, and unsupervised TSAD methods for univariate time series and multivariate time series. Our research highlights the immense potential of contrastive representation learning in advancing the TSAD field, thus paving the way for novel applications and in-depth exploration.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
One main challenge in time series anomaly detection (TSAD) is the lack of labelled data in many real-life scenarios. Most of the existing anomaly detection methods focus on learning the normal behaviour of unlabelled time series in an unsupervised manner. The normal boundary is often defined tightly, resulting in slight deviations being classified as anomalies, consequently leading to a high false positive rate and a limited ability to generalise normal patterns. To address this, we introduce a novel end-to-end self-supervised ContrAstive Representation Learning approach for time series Anomaly detection (CARLA). While existing contrastive learning methods assume that augmented time series windows are positive samples and temporally distant windows are negative samples, we argue that these assumptions are limited as augmentation of time series can transform them to negative samples, and a temporally distant window can represent a positive sample. Existing approaches to contrastive learning for time series have directly copied methods developed for image analysis. We argue that these methods do not transfer well. Instead, our contrastive approach leverages existing generic knowledge about time series anomalies and injects various types of anomalies as negative samples. Therefore, CARLA not only learns normal behaviour but also learns deviations indicating anomalies. It creates similar representations for temporally close windows and distinct ones for anomalies. Additionally, it leverages the information about representations’ neighbours through a self-supervised approach to classify windows based on their nearest/furthest neighbours to further enhance the performance of anomaly detection. In extensive tests on seven major real-world TSAD datasets, CARLA shows superior performance (F1 and AU-PR) over state-of-the-art self-supervised, semi-supervised, and unsupervised TSAD methods for univariate time series and multivariate time series. Our research highlights the immense potential of contrastive representation learning in advancing the TSAD field, thus paving the way for novel applications and in-depth exploration.
Yao, Shunhan; Huang, Yuanxiang; Wang, Xiaoyu; Zhang, Yiwen; Paixao, Ian Costa; Wang, Zhikang; Chai, Charla Lu; Wang, Hongtao; Lu, Dinggui; Webb, Geoffrey I; Li, Shanshan; Guo, Yuming; Chen, Qingfeng; Song, Jiangning
A Radiograph Dataset for the Classification, Localization, and Segmentation of Primary Bone Tumors
Scientific Data, vol. 12, 2025, ISSN: 2052-4463.
@article{Yao2025,
title = {A Radiograph Dataset for the Classification, Localization, and Segmentation of Primary Bone Tumors},
author = {Shunhan Yao and Yuanxiang Huang and Xiaoyu Wang and Yiwen Zhang and Ian Costa Paixao and Zhikang Wang and Charla Lu Chai and Hongtao Wang and Dinggui Lu and Geoffrey I Webb and Shanshan Li and Yuming Guo and Qingfeng Chen and Jiangning Song},
doi = {10.1038/s41597-024-04311-y},
issn = {2052-4463},
year = {2025},
date = {2025-01-01},
journal = {Scientific Data},
volume = {12},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ramakrishnaiah, Yashpal; Macesic, Nenad; Webb, Geoffrey I.; Peleg, Anton Y.; Tyagi, Sonika
International Journal of Medical Informatics, pp. 105816, 2025, ISSN: 1386-5056.
@article{Ramakrishnaiah2025,
title = {EHR-ML: A Data-Driven Framework for Designing Machine Learning Applications with Electronic Health Records},
author = {Yashpal Ramakrishnaiah and Nenad Macesic and Geoffrey I. Webb and Anton Y. Peleg and Sonika Tyagi},
url = {https://www.sciencedirect.com/science/article/pii/S1386505625000334},
doi = {https://doi.org/10.1016/j.ijmedinf.2025.105816},
issn = {1386-5056},
year = {2025},
date = {2025-01-01},
journal = {International Journal of Medical Informatics},
pages = {105816},
abstract = {Objective
The healthcare landscape is experiencing a transformation with the integration of Artificial Intelligence (AI) into traditional analytic workflows. However, its integration faces challenges resulting in a crisis of generalisability. Key obstacles include; 1) Insufficient consideration of local contextual factors, such as institution-specific data formats, practices, and protocols, which can lead to variability in clinical practices across different institutions. 2) ad-hoc data preparation and design of machine learning strategies. 3) manual subjective adjustment of design parameters resulting in sub-optimal performance. 4) EHR specific challenges regarding data biases affecting the model outcomes and unique intermittent temporal nature of the data necessitating specialised handling 5) lack of cross-institutional data validations.
Methods
To address these challenges, EHR-ML, provides an easy to use structured framework for designing optimum machine learning applications in a data-driven manner. The framework supports ingestion of local institutional electronic health records (EHRs) and process standardisation. The study design and parameter optimisisation is done in a fully data-driven evidence-based approach. It seamlessly integrating with existing quality control tools. To handle the unique characteristics of the EHR data, it offers customisable ensemble models. It enables the acquisition of EHR data from diverse systems and harmonise them into common formats following international standards.
Results
The effectiveness of the EHR-ML is demonstrated through a series of case studies. These studies highlight its capability to develop high-performance models in a fully automated manner, consistently surpassing the performance of traditional methodologies. Furthermore, they exhibited strong generalisability across diverse healthcare settings.
Discussion and Conclusion
EHR-ML enhances the clinical relevance and accuracy of predictive models by incorporating local context into machine learning applications. Additionally, by providing an user-friendly fully-automated framework, it facilitates rapid hypothesis testing aimed to generate localised biomedical knowledge.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Objective
The healthcare landscape is experiencing a transformation with the integration of Artificial Intelligence (AI) into traditional analytic workflows. However, its integration faces challenges resulting in a crisis of generalisability. Key obstacles include; 1) Insufficient consideration of local contextual factors, such as institution-specific data formats, practices, and protocols, which can lead to variability in clinical practices across different institutions. 2) ad-hoc data preparation and design of machine learning strategies. 3) manual subjective adjustment of design parameters resulting in sub-optimal performance. 4) EHR specific challenges regarding data biases affecting the model outcomes and unique intermittent temporal nature of the data necessitating specialised handling 5) lack of cross-institutional data validations.
Methods
To address these challenges, EHR-ML, provides an easy to use structured framework for designing optimum machine learning applications in a data-driven manner. The framework supports ingestion of local institutional electronic health records (EHRs) and process standardisation. The study design and parameter optimisisation is done in a fully data-driven evidence-based approach. It seamlessly integrating with existing quality control tools. To handle the unique characteristics of the EHR data, it offers customisable ensemble models. It enables the acquisition of EHR data from diverse systems and harmonise them into common formats following international standards.
Results
The effectiveness of the EHR-ML is demonstrated through a series of case studies. These studies highlight its capability to develop high-performance models in a fully automated manner, consistently surpassing the performance of traditional methodologies. Furthermore, they exhibited strong generalisability across diverse healthcare settings.
Discussion and Conclusion
EHR-ML enhances the clinical relevance and accuracy of predictive models by incorporating local context into machine learning applications. Additionally, by providing an user-friendly fully-automated framework, it facilitates rapid hypothesis testing aimed to generate localised biomedical knowledge.
The healthcare landscape is experiencing a transformation with the integration of Artificial Intelligence (AI) into traditional analytic workflows. However, its integration faces challenges resulting in a crisis of generalisability. Key obstacles include; 1) Insufficient consideration of local contextual factors, such as institution-specific data formats, practices, and protocols, which can lead to variability in clinical practices across different institutions. 2) ad-hoc data preparation and design of machine learning strategies. 3) manual subjective adjustment of design parameters resulting in sub-optimal performance. 4) EHR specific challenges regarding data biases affecting the model outcomes and unique intermittent temporal nature of the data necessitating specialised handling 5) lack of cross-institutional data validations.
Methods
To address these challenges, EHR-ML, provides an easy to use structured framework for designing optimum machine learning applications in a data-driven manner. The framework supports ingestion of local institutional electronic health records (EHRs) and process standardisation. The study design and parameter optimisisation is done in a fully data-driven evidence-based approach. It seamlessly integrating with existing quality control tools. To handle the unique characteristics of the EHR data, it offers customisable ensemble models. It enables the acquisition of EHR data from diverse systems and harmonise them into common formats following international standards.
Results
The effectiveness of the EHR-ML is demonstrated through a series of case studies. These studies highlight its capability to develop high-performance models in a fully automated manner, consistently surpassing the performance of traditional methodologies. Furthermore, they exhibited strong generalisability across diverse healthcare settings.
Discussion and Conclusion
EHR-ML enhances the clinical relevance and accuracy of predictive models by incorporating local context into machine learning applications. Additionally, by providing an user-friendly fully-automated framework, it facilitates rapid hypothesis testing aimed to generate localised biomedical knowledge.
Tan, Chang Wei; Herrmann, Matthieu; Salehi, Mahsa; Webb, Geoffrey I.
Proximity forest 2.0: a new effective and scalable similarity-based classifier for time series
Data Mining and Knowledge Discovery, vol. 39, no. 2, 2025, ISSN: 1573-756X.
@article{Tan2025,
title = {Proximity forest 2.0: a new effective and scalable similarity-based classifier for time series},
author = {Chang Wei Tan and Matthieu Herrmann and Mahsa Salehi and Geoffrey I. Webb},
doi = {10.1007/s10618-024-01085-0},
issn = {1573-756X},
year = {2025},
date = {2025-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {39},
number = {2},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Dempster, Angus; Tan, Chang Wei; Miller, Lynn; Foumani, Navid Mohammadi; Schmidt, Daniel F.; Webb, Geoffrey I.
Highly Scalable Time Series Classification for Very Large Datasets
Advanced Analytics and Learning on Temporal Data, pp. 80–95, Springer Nature Switzerland, 2025, ISSN: 1611-3349.
@inbook{Dempster2025,
title = {Highly Scalable Time Series Classification for Very Large Datasets},
author = {Angus Dempster and Chang Wei Tan and Lynn Miller and Navid Mohammadi Foumani and Daniel F. Schmidt and Geoffrey I. Webb},
doi = {10.1007/978-3-031-77066-1_5},
issn = {1611-3349},
year = {2025},
date = {2025-01-01},
booktitle = {Advanced Analytics and Learning on Temporal Data},
pages = {80\textendash95},
publisher = {Springer Nature Switzerland},
abstract = {Relatively little work in the field of time series classification focuses on learning effectively from very large quantities of data. Large datasets present significant practical challenges in terms of computational cost and memory complexity. We present strategies for extending two recent state-of-the-art methods for time series classification\textemdashnamely, Hydra and Quant\textemdashto very large datasets. This allows for training these methods on large quantities of data with a fixed memory cost, while making effective use of appropriate computational resources. For Hydra, we fit a ridge regression classifier iteratively, using a single pass through the data, integrating the Hydra transform with the process of fitting the ridge regression model, allowing for a fixed memory cost, and allowing almost all computation to be performed on GPU. For Quant, we ‘spread’ subsets of extremely randomised trees over a given dataset such that each tree is trained using as much data as possible for a given amount of memory while minimising reads from the data, allowing for a simple tradeoff between error and computational cost. This allows for the straightforward application of both methods to very large quantities of data. We demonstrate these approaches with results (including learning curves) on a selection of large datasets with between approximately 85, 000 and 47 million training examples.},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Relatively little work in the field of time series classification focuses on learning effectively from very large quantities of data. Large datasets present significant practical challenges in terms of computational cost and memory complexity. We present strategies for extending two recent state-of-the-art methods for time series classification—namely, Hydra and Quant—to very large datasets. This allows for training these methods on large quantities of data with a fixed memory cost, while making effective use of appropriate computational resources. For Hydra, we fit a ridge regression classifier iteratively, using a single pass through the data, integrating the Hydra transform with the process of fitting the ridge regression model, allowing for a fixed memory cost, and allowing almost all computation to be performed on GPU. For Quant, we ‘spread’ subsets of extremely randomised trees over a given dataset such that each tree is trained using as much data as possible for a given amount of memory while minimising reads from the data, allowing for a simple tradeoff between error and computational cost. This allows for the straightforward application of both methods to very large quantities of data. We demonstrate these approaches with results (including learning curves) on a selection of large datasets with between approximately 85, 000 and 47 million training examples.
Darban, Zahra Zamanzadeh; Yang, Yiyuan; Webb, Geoffrey I.; Aggarwal, Charu C.; Wen, Qingsong; Pan, Shirui; Salehi, Mahsa
DACAD: Domain Adaptation Contrastive Learning for Anomaly Detection in Multivariate Time Series
IEEE Transactions on Knowledge and Data Engineering, pp. 1-12, 2025, ISSN: 2326-3865.
@article{Darban2025a,
title = {DACAD: Domain Adaptation Contrastive Learning for Anomaly Detection in Multivariate Time Series},
author = {Zahra Zamanzadeh Darban and Yiyuan Yang and Geoffrey I. Webb and Charu C. Aggarwal and Qingsong Wen and Shirui Pan and Mahsa Salehi},
doi = {10.1109/tkde.2025.3569909},
issn = {2326-3865},
year = {2025},
date = {2025-01-01},
journal = {IEEE Transactions on Knowledge and Data Engineering},
pages = {1-12},
publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Chen, Qingfeng; Li, Shiyuan; Liu, Yixin; Pan, Shirui; Webb, Geoffrey I.; Zhang, Shichao
Uncertainty-Aware Graph Neural Networks: A Multihop Evidence Fusion Approach
IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 10, pp. 18463-18477, 2025, ISSN: 2162-2388.
@article{Chen2025,
title = {Uncertainty-Aware Graph Neural Networks: A Multihop Evidence Fusion Approach},
author = {Qingfeng Chen and Shiyuan Li and Yixin Liu and Shirui Pan and Geoffrey I. Webb and Shichao Zhang},
doi = {10.1109/tnnls.2025.3581083},
issn = {2162-2388},
year = {2025},
date = {2025-01-01},
journal = {IEEE Transactions on Neural Networks and Learning Systems},
volume = {36},
number = {10},
pages = {18463-18477},
publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Pan, Tong; Webb, Geoffrey I.; Imoto, Seiya; Song, Jiangning
Integrating Gene Ontology Relationships for Protein Function Prediction Using PFresGO
Protein Function Prediction, pp. 161-169, Springer US, 2025, ISSN: 1940-6029.
@inbook{Pan2025,
title = {Integrating Gene Ontology Relationships for Protein Function Prediction Using PFresGO},
author = {Tong Pan and Geoffrey I. Webb and Seiya Imoto and Jiangning Song},
doi = {10.1007/978-1-0716-4662-5_9},
issn = {1940-6029},
year = {2025},
date = {2025-01-01},
booktitle = {Protein Function Prediction},
pages = {161-169},
publisher = {Springer US},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Koh, Huan Yee; Zheng, Yizhen; Yang, Madeleine; Arora, Rohit; Webb, Geoffrey I.; Pan, Shirui; Li, Li; Church, George M.
Nature Reviews Bioengineering, 2025, ISSN: 2731-6092.
@article{Koh2025,
title = {AI-driven protein design},
author = {Huan Yee Koh and Yizhen Zheng and Madeleine Yang and Rohit Arora and Geoffrey I. Webb and Shirui Pan and Li Li and George M. Church},
doi = {10.1038/s44222-025-00349-8},
issn = {2731-6092},
year = {2025},
date = {2025-01-01},
journal = {Nature Reviews Bioengineering},
abstract = {Protein design is undergoing a revolution driven by artificial intelligence (AI), transforming how we engineer proteins for applications in drug discovery, biotechnology and synthetic biology. By navigating the immense complexity of protein sequence space and overcoming the limitations of structural and functional data, AI enables unprecedented precision and speed in designing novel proteins with tailored functions. Central to this Review is a comprehensive and actionable roadmap for designers, providing step-by-step guidance on how to integrate state-of-the-art AI tools into protein design workflows, including tools for structural and functional prediction as well as generative models for de novo design. To illustrate this roadmap in practice, we present case studies showcasing AI-driven protein design, from engineering therapeutic proteins to designing novel proteins that unlock enzyme functions and reprogramme biomolecular systems. Looking ahead, we outline future directions highlighting the vast potential of AI to revolutionize synthetic biology, expedite drug development and drive sustainable biotechnology, positioning it as a transformative force at the forefront of protein design.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Protein design is undergoing a revolution driven by artificial intelligence (AI), transforming how we engineer proteins for applications in drug discovery, biotechnology and synthetic biology. By navigating the immense complexity of protein sequence space and overcoming the limitations of structural and functional data, AI enables unprecedented precision and speed in designing novel proteins with tailored functions. Central to this Review is a comprehensive and actionable roadmap for designers, providing step-by-step guidance on how to integrate state-of-the-art AI tools into protein design workflows, including tools for structural and functional prediction as well as generative models for de novo design. To illustrate this roadmap in practice, we present case studies showcasing AI-driven protein design, from engineering therapeutic proteins to designing novel proteins that unlock enzyme functions and reprogramme biomolecular systems. Looking ahead, we outline future directions highlighting the vast potential of AI to revolutionize synthetic biology, expedite drug development and drive sustainable biotechnology, positioning it as a transformative force at the forefront of protein design.
Pan, Tong; Bi, Yue; Wang, Xiaoyu; Zhang, Ying; Webb, Geoffrey I; Gasser, Robin B; Kurgan, Lukasz; Song, Jiangning
Genomics, Proteomics & Bioinformatics, 2024, ISSN: 2210-3244.
@article{Pan2024,
title = {SCREEN: A Graph-based Contrastive Learning Tool to Infer Catalytic Residues and Assess Enzyme Mutations},
author = {Tong Pan and Yue Bi and Xiaoyu Wang and Ying Zhang and Geoffrey I Webb and Robin B Gasser and Lukasz Kurgan and Jiangning Song},
doi = {10.1093/gpbjnl/qzae094},
issn = {2210-3244},
year = {2024},
date = {2024-12-01},
journal = {Genomics, Proteomics \& Bioinformatics},
publisher = {Oxford University Press (OUP)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Darban, Zahra Zamanzadeh; Webb, Geoffrey I.; Pan, Shirui; Aggarwal, Charu; Salehi, Mahsa
Deep Learning for Time Series Anomaly Detection: A Survey
ACM Computing Surveys, vol. 57, no. 1, pp. 1–42, 2024, ISSN: 1557-7341.
@article{ZamanzadehDarban2024,
title = {Deep Learning for Time Series Anomaly Detection: A Survey},
author = {Zahra Zamanzadeh Darban and Geoffrey I. Webb and Shirui Pan and Charu Aggarwal and Mahsa Salehi},
doi = {10.1145/3691338},
issn = {1557-7341},
year = {2024},
date = {2024-10-01},
journal = {ACM Computing Surveys},
volume = {57},
number = {1},
pages = {1\textendash42},
publisher = {Association for Computing Machinery (ACM)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Bi, Yue; Li, Fuyi; Wang, Cong; Pan, Tong; Davidovich, Chen; Webb, Geoffrey I; Song, Jiangning
Advancing microRNA target site prediction with transformer and base-pairing patterns
Nucleic Acids Research, vol. 52, no. 19, pp. 11455–11465, 2024, ISSN: 1362-4962.
@article{Bi2024,
title = {Advancing microRNA target site prediction with transformer and base-pairing patterns},
author = {Yue Bi and Fuyi Li and Cong Wang and Tong Pan and Chen Davidovich and Geoffrey I Webb and Jiangning Song},
doi = {10.1093/nar/gkae782},
issn = {1362-4962},
year = {2024},
date = {2024-09-01},
journal = {Nucleic Acids Research},
volume = {52},
number = {19},
pages = {11455\textendash11465},
publisher = {Oxford University Press (OUP)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Nguyen, Hoai-An; Peleg, Anton Y.; Song, Jiangning; Antony, Bhavna; Webb, Geoffrey I.; Wisniewski, Jessica A.; Blakeway, Luke V.; Badoordeen, Gnei Z.; Theegala, Ravali; Zisis, Helen; Dowe, David L.; Macesic, Nenad
mSystems, 2024, ISSN: 2379-5077.
@article{Nguyen2024,
title = {Predicting Pseudomonas aeruginosa drug resistance using artificial intelligence and clinical MALDI-TOF mass spectra},
author = {Hoai-An Nguyen and Anton Y. Peleg and Jiangning Song and Bhavna Antony and Geoffrey I. Webb and Jessica A. Wisniewski and Luke V. Blakeway and Gnei Z. Badoordeen and Ravali Theegala and Helen Zisis and David L. Dowe and Nenad Macesic},
editor = {Yu-Liang Yang},
doi = {10.1128/msystems.00789-24},
issn = {2379-5077},
year = {2024},
date = {2024-08-01},
journal = {mSystems},
publisher = {American Society for Microbiology},
abstract = {Matrix-assisted laser desorption/ionization\textendashtime of flight mass spectrometry (MALDI-TOF MS) is widely used in clinical microbiology laboratories for bacterial identification but its use for detection of antimicrobial resistance (AMR) remains limited. Here, we used MALDI-TOF MS with artificial intelligence (AI) approaches to successfully predict AMR in Pseudomonas aeruginosa, a priority pathogen with complex AMR mechanisms. The highest performance was achieved for modern β-lactam/β-lactamase inhibitor drugs, namely, ceftazidime/avibactam and ceftolozane/tazobactam. For these drugs, the model demonstrated area under the receiver operating characteristic curve (AUROC) of 0.869 and 0.856, specificity of 0.925 and 0.897, and sensitivity of 0.731 and 0.714, respectively. As part of this work, we developed dynamic binning, a feature engineering technique that effectively reduces the high-dimensional feature set and has wide-ranging applicability to MALDI-TOF MS data. Compared to conventional feature engineering approaches, the dynamic binning method yielded highest performance in 7 of 10 antimicrobials. Moreover, we showcased the efficacy of transfer learning in enhancing the AUROC performance for 8 of 11 antimicrobials. By assessing the contribution of features to the model's prediction, we identified proteins that may contribute to AMR mechanisms. Our findings demonstrate the potential of combining AI with MALDI-TOF MS as a rapid AMR diagnostic tool for Pseudomonas aeruginosa.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Matrix-assisted laser desorption/ionization–time of flight mass spectrometry (MALDI-TOF MS) is widely used in clinical microbiology laboratories for bacterial identification but its use for detection of antimicrobial resistance (AMR) remains limited. Here, we used MALDI-TOF MS with artificial intelligence (AI) approaches to successfully predict AMR in Pseudomonas aeruginosa, a priority pathogen with complex AMR mechanisms. The highest performance was achieved for modern β-lactam/β-lactamase inhibitor drugs, namely, ceftazidime/avibactam and ceftolozane/tazobactam. For these drugs, the model demonstrated area under the receiver operating characteristic curve (AUROC) of 0.869 and 0.856, specificity of 0.925 and 0.897, and sensitivity of 0.731 and 0.714, respectively. As part of this work, we developed dynamic binning, a feature engineering technique that effectively reduces the high-dimensional feature set and has wide-ranging applicability to MALDI-TOF MS data. Compared to conventional feature engineering approaches, the dynamic binning method yielded highest performance in 7 of 10 antimicrobials. Moreover, we showcased the efficacy of transfer learning in enhancing the AUROC performance for 8 of 11 antimicrobials. By assessing the contribution of features to the model's prediction, we identified proteins that may contribute to AMR mechanisms. Our findings demonstrate the potential of combining AI with MALDI-TOF MS as a rapid AMR diagnostic tool for Pseudomonas aeruginosa.
Bagnall, Anthony; Middlehurst, Matthew; Forestier, Germain; Ismail-Fawaz, Ali; Guillaume, Antoine; Guijo-Rubio, David; Tan, Chang Wei; Dempster, Angus; Webb, Geoffrey I.
A Hands-on Introduction to Time Series Classification and Regression
Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 6410-6411, ACM, 2024.
@inproceedings{Bagnall2024,
title = {A Hands-on Introduction to Time Series Classification and Regression},
author = {Anthony Bagnall and Matthew Middlehurst and Germain Forestier and Ali Ismail-Fawaz and Antoine Guillaume and David Guijo-Rubio and Chang Wei Tan and Angus Dempster and Geoffrey I. Webb},
doi = {10.1145/3637528.3671443},
year = {2024},
date = {2024-08-01},
booktitle = {Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages = {6410-6411},
publisher = {ACM},
series = {KDD ’24},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Nguyen, Anh T. N.; Nguyen, Diep T. N.; Koh, Huan Yee; Toskov, Jason; MacLean, William; Xu, Andrew; Zhang, Daokun; Webb, Geoffrey I.; May, Lauren T.; Halls, Michelle L.
The application of artificial intelligence to accelerate G protein-coupled receptor drug discovery
British Journal of Pharmacology, vol. 181, no. 14, pp. 2371-2384, 2024.
@article{Nguyen,
title = {The application of artificial intelligence to accelerate G protein-coupled receptor drug discovery},
author = {Anh T. N. Nguyen and Diep T. N. Nguyen and Huan Yee Koh and Jason Toskov and William MacLean and Andrew Xu and Daokun Zhang and Geoffrey I. Webb and Lauren T. May and Michelle L. Halls},
doi = {10.1111/bph.16140},
year = {2024},
date = {2024-01-01},
journal = {British Journal of Pharmacology},
volume = {181},
number = {14},
pages = {2371-2384},
abstract = {The application of artificial intelligence (AI) approaches to drug discovery for G protein-coupled receptors (GPCRs) is a rapidly expanding area. Artificial intelligence can be used at multiple stages during the drug discovery process, from aiding our understanding of the fundamental actions of GPCRs to the discovery of new ligand-GPCR interactions or the prediction of clinical responses. Here, we provide an overview of the concepts behind artificial intelligence, including the subfields of machine learning and deep learning. We summarise the published applications of artificial intelligence to different stages of the GPCR drug discovery process. Finally, we reflect on the benefits and limitations of artificial intelligence and share our vision for the exciting potential for further development of applications to aid GPCR drug discovery. In addition to making the drug discovery process faster, smarter and cheaper, we anticipate that the application of artificial intelligence will create exciting new opportunities for GPCR drug discovery.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The application of artificial intelligence (AI) approaches to drug discovery for G protein-coupled receptors (GPCRs) is a rapidly expanding area. Artificial intelligence can be used at multiple stages during the drug discovery process, from aiding our understanding of the fundamental actions of GPCRs to the discovery of new ligand-GPCR interactions or the prediction of clinical responses. Here, we provide an overview of the concepts behind artificial intelligence, including the subfields of machine learning and deep learning. We summarise the published applications of artificial intelligence to different stages of the GPCR drug discovery process. Finally, we reflect on the benefits and limitations of artificial intelligence and share our vision for the exciting potential for further development of applications to aid GPCR drug discovery. In addition to making the drug discovery process faster, smarter and cheaper, we anticipate that the application of artificial intelligence will create exciting new opportunities for GPCR drug discovery.
Foumani, Navid Mohammadi; Tan, Chang Wei; Webb, Geoffrey I.; Salehi, Mahsa
Improving position encoding of transformers for multivariate time series classification
Data Mining and Knowledge Discovery, vol. 38, pp. 22-48, 2024.
@article{Foumani2023,
title = {Improving position encoding of transformers for multivariate time series classification},
author = {Navid Mohammadi Foumani and Chang Wei Tan and Geoffrey I. Webb and Mahsa Salehi},
doi = {10.1007/s10618-023-00948-2},
year = {2024},
date = {2024-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {38},
pages = {22-48},
publisher = {Springer Science and Business Media LLC},
abstract = {Transformers have demonstrated outstanding performance in many applications of deep learning. When applied to time series data, transformers require effective position encoding to capture the ordering of the time series data. The efficacy of position encoding in time series analysis is not well-studied and remains controversial, e.g., whether it is better to inject absolute position encoding or relative position encoding, or a combination of them. In order to clarify this, we first review existing absolute and relative position encoding methods when applied in time series classification. We then proposed a new absolute position encoding method dedicated to time series data called time Absolute Position Encoding (tAPE). Our new method incorporates the series length and input embedding dimension in absolute position encoding. Additionally, we propose computationally Efficient implementation of Relative Position Encoding (eRPE) to improve generalisability for time series. We then propose a novel multivariate time series classification model combining tAPE/eRPE and convolution-based input encoding named ConvTran to improve the position and data embedding of time series data. The proposed absolute and relative position encoding methods are simple and efficient. They can be easily integrated into transformer blocks and used for downstream tasks such as forecasting, extrinsic regression, and anomaly detection. Extensive experiments on 32 multivariate time-series datasets show that our model is significantly more accurate than state-of-the-art convolution and transformer-based models. Code and models are open-sourced at https://github.com/Navidfoumani/ConvTran.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Transformers have demonstrated outstanding performance in many applications of deep learning. When applied to time series data, transformers require effective position encoding to capture the ordering of the time series data. The efficacy of position encoding in time series analysis is not well-studied and remains controversial, e.g., whether it is better to inject absolute position encoding or relative position encoding, or a combination of them. In order to clarify this, we first review existing absolute and relative position encoding methods when applied in time series classification. We then proposed a new absolute position encoding method dedicated to time series data called time Absolute Position Encoding (tAPE). Our new method incorporates the series length and input embedding dimension in absolute position encoding. Additionally, we propose computationally Efficient implementation of Relative Position Encoding (eRPE) to improve generalisability for time series. We then propose a novel multivariate time series classification model combining tAPE/eRPE and convolution-based input encoding named ConvTran to improve the position and data embedding of time series data. The proposed absolute and relative position encoding methods are simple and efficient. They can be easily integrated into transformer blocks and used for downstream tasks such as forecasting, extrinsic regression, and anomaly detection. Extensive experiments on 32 multivariate time-series datasets show that our model is significantly more accurate than state-of-the-art convolution and transformer-based models. Code and models are open-sourced at https://github.com/Navidfoumani/ConvTran.
Liu, Tongliang; Webb, Geoff; Yue, Lin; Wang, Dadong (Ed.)
Springer Nature Singapore, 2024, ISSN: 1611-3349.
@proceedings{2024,
title = {AI 2023: Advances in Artificial Intelligence: 36th Australasian Joint Conference on Artificial Intelligence, AI 2023, Brisbane, QLD, Australia, November 28 December 1, 2023, Proceedings, Part I},
editor = {Tongliang Liu and Geoff Webb and Lin Yue and Dadong Wang},
doi = {10.1007/978-981-99-8388-9},
issn = {1611-3349},
year = {2024},
date = {2024-01-01},
publisher = {Springer Nature Singapore},
series = {Lecture Notes in Computer Science},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Foumani, Navid Mohammadi; Miller, Lynn; Tan, Chang Wei; Webb, Geoffrey I.; Forestier, Germain; Salehi, Mahsa
Deep Learning for Time Series Classification and Extrinsic Regression: A Current Survey
ACM Computing Surveys, vol. 56, no. 9, 2024, ISSN: 0360-0300.
@article{Foumani2024,
title = {Deep Learning for Time Series Classification and Extrinsic Regression: A Current Survey},
author = {Navid Mohammadi Foumani and Lynn Miller and Chang Wei Tan and Geoffrey I. Webb and Germain Forestier and Mahsa Salehi},
doi = {10.1145/3649448},
issn = {0360-0300},
year = {2024},
date = {2024-01-01},
journal = {ACM Computing Surveys},
volume = {56},
number = {9},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
abstract = {Time Series Classification and Extrinsic Regression are important and challenging machine learning tasks. Deep learning has revolutionized natural language processing and computer vision and holds great promise in other fields such as time series analysis where the relevant features must often be abstracted from the raw data but are not known a priori. This paper surveys the current state of the art in the fast-moving field of deep learning for time series classification and extrinsic regression. We review different network architectures and training methods used for these tasks and discuss the challenges and opportunities when applying deep learning to time series data. We also summarize two critical applications of time series classification and extrinsic regression, human activity recognition and satellite earth observation.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Time Series Classification and Extrinsic Regression are important and challenging machine learning tasks. Deep learning has revolutionized natural language processing and computer vision and holds great promise in other fields such as time series analysis where the relevant features must often be abstracted from the raw data but are not known a priori. This paper surveys the current state of the art in the fast-moving field of deep learning for time series classification and extrinsic regression. We review different network architectures and training methods used for these tasks and discuss the challenges and opportunities when applying deep learning to time series data. We also summarize two critical applications of time series classification and extrinsic regression, human activity recognition and satellite earth observation.
Gao, Mengmeng; Zhang, Daokun; Chen, Yi; Zhang, Yiwen; Wang, Zhikang; Wang, Xiaoyu; Li, Shanshan; Guo, Yuming; Webb, Geoffrey I.; Nguyen, Anh T. N.; May, Lauren; Song, Jiangning
GraphormerDTI: A graph transformer-based approach for drug-target interaction prediction
Computers in Biology and Medicine, pp. 108339, 2024, ISSN: 0010-4825.
@article{Gao2024,
title = {GraphormerDTI: A graph transformer-based approach for drug-target interaction prediction},
author = {Mengmeng Gao and Daokun Zhang and Yi Chen and Yiwen Zhang and Zhikang Wang and Xiaoyu Wang and Shanshan Li and Yuming Guo and Geoffrey I. Webb and Anh T. N. Nguyen and Lauren May and Jiangning Song},
url = {https://www.sciencedirect.com/science/article/pii/S0010482524004232},
doi = {https://doi.org/10.1016/j.compbiomed.2024.108339},
issn = {0010-4825},
year = {2024},
date = {2024-01-01},
journal = {Computers in Biology and Medicine},
pages = {108339},
abstract = {The application of Artificial Intelligence (AI) to screen drug molecules with potential therapeutic effects has revolutionized the drug discovery process, with significantly lower economic cost and time consumption than the traditional drug discovery pipeline. With the great power of AI, it is possible to rapidly search the vast chemical space for potential drug-target interactions (DTIs) between candidate drug molecules and disease protein targets. However, only a small proportion of molecules have labelled DTIs, consequently limiting the performance of AI-based drug screening. To solve this problem, a machine learning-based approach with great ability to generalize DTI prediction across molecules is desirable. Many existing machine learning approaches for DTI identification failed to exploit the full information with respect to the topological structures of candidate molecules. To develop a better approach for DTI prediction, we propose GraphormerDTI, which employs the powerful Graph Transformer neural network to model molecular structures. GraphormerDTI embeds molecular graphs into vector-format representations through iterative Transformer-based message passing, which encodes molecules' structural characteristics by node centrality encoding, node spatial encoding and edge encoding. With a strong structural inductive bias, the proposed GraphormerDTI approach can effectively infer informative representations for out-of-sample molecules and as such, it is capable of predicting DTIs across molecules with an exceptional performance. GraphormerDTI integrates the Graph Transformer neural network with a 1-dimensional Convolutional Neural Network (1D-CNN) to extract the drugs' and target proteins' representations and leverages an attention mechanism to model the interactions between them. To examine GraphormerDTI's performance for DTI prediction, we conduct experiments on three benchmark datasets, where GraphormerDTI achieves a superior performance than five state-of-the-art baselines for out-of-molecule DTI prediction, including GNN-CPI, GNN-PT, DeepEmbedding-DTI, MolTrans and HyperAttentionDTI, and is on a par with the best baseline for transductive DTI prediction. The source codes and datasets are publicly accessible at https://github.com/mengmeng34/GraphormerDTI.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The application of Artificial Intelligence (AI) to screen drug molecules with potential therapeutic effects has revolutionized the drug discovery process, with significantly lower economic cost and time consumption than the traditional drug discovery pipeline. With the great power of AI, it is possible to rapidly search the vast chemical space for potential drug-target interactions (DTIs) between candidate drug molecules and disease protein targets. However, only a small proportion of molecules have labelled DTIs, consequently limiting the performance of AI-based drug screening. To solve this problem, a machine learning-based approach with great ability to generalize DTI prediction across molecules is desirable. Many existing machine learning approaches for DTI identification failed to exploit the full information with respect to the topological structures of candidate molecules. To develop a better approach for DTI prediction, we propose GraphormerDTI, which employs the powerful Graph Transformer neural network to model molecular structures. GraphormerDTI embeds molecular graphs into vector-format representations through iterative Transformer-based message passing, which encodes molecules' structural characteristics by node centrality encoding, node spatial encoding and edge encoding. With a strong structural inductive bias, the proposed GraphormerDTI approach can effectively infer informative representations for out-of-sample molecules and as such, it is capable of predicting DTIs across molecules with an exceptional performance. GraphormerDTI integrates the Graph Transformer neural network with a 1-dimensional Convolutional Neural Network (1D-CNN) to extract the drugs' and target proteins' representations and leverages an attention mechanism to model the interactions between them. To examine GraphormerDTI's performance for DTI prediction, we conduct experiments on three benchmark datasets, where GraphormerDTI achieves a superior performance than five state-of-the-art baselines for out-of-molecule DTI prediction, including GNN-CPI, GNN-PT, DeepEmbedding-DTI, MolTrans and HyperAttentionDTI, and is on a par with the best baseline for transductive DTI prediction. The source codes and datasets are publicly accessible at https://github.com/mengmeng34/GraphormerDTI.
Miller, Lynn; Pelletier, Charlotte; Webb, Geoffrey I.
Deep Learning for Satellite Image Time-Series Analysis: A review
IEEE Geoscience and Remote Sensing Magazine, vol. 12, no. 3, pp. 81-124, 2024.
@article{Miller2024,
title = {Deep Learning for Satellite Image Time-Series Analysis: A review},
author = {Lynn Miller and Charlotte Pelletier and Geoffrey I. Webb},
doi = {10.1109/MGRS.2024.3393010},
year = {2024},
date = {2024-01-01},
journal = {IEEE Geoscience and Remote Sensing Magazine},
volume = {12},
number = {3},
pages = {81-124},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Dempster, Angus; Schmidt, Daniel F.; Webb, Geoffrey I.
Quant: a minimalist interval method for time series classification
Data Mining and Knowledge Discovery, vol. 38, pp. 2377-2402, 2024, ISSN: 1573-756X.
@article{Dempster2024,
title = {Quant: a minimalist interval method for time series classification},
author = {Angus Dempster and Daniel F. Schmidt and Geoffrey I. Webb},
doi = {10.1007/s10618-024-01036-9},
issn = {1573-756X},
year = {2024},
date = {2024-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {38},
pages = {2377-2402},
publisher = {Springer Science and Business Media LLC},
abstract = {We show that it is possible to achieve the same accuracy, on average, as the most accurate existing interval methods for time series classification on a standard set of benchmark datasets using a single type of feature (quantiles), fixed intervals, and an ‘off the shelf’ classifier. This distillation of interval-based approaches represents a fast and accurate method for time series classification, achieving state-of-the-art accuracy on the expanded set of 142 datasets in the UCR archive with a total compute time (training and inference) of less than 15 min using a single CPU core.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We show that it is possible to achieve the same accuracy, on average, as the most accurate existing interval methods for time series classification on a standard set of benchmark datasets using a single type of feature (quantiles), fixed intervals, and an ‘off the shelf’ classifier. This distillation of interval-based approaches represents a fast and accurate method for time series classification, achieving state-of-the-art accuracy on the expanded set of 142 datasets in the UCR archive with a total compute time (training and inference) of less than 15 min using a single CPU core.
Koh, Huan Yee; Nguyen, Anh T. N.; Pan, Shirui; May, Lauren T.; Webb, Geoffrey I.
Nature Machine Intelligence, vol. 6, pp. 673–687, 2024, ISSN: 2522-5839.
@article{Koh2024,
title = {Physicochemical graph neural network for learning protein-ligand interaction fingerprints from sequence data},
author = {Huan Yee Koh and Anh T. N. Nguyen and Shirui Pan and Lauren T. May and Geoffrey I. Webb},
doi = {10.1038/s42256-024-00847-1},
issn = {2522-5839},
year = {2024},
date = {2024-01-01},
journal = {Nature Machine Intelligence},
volume = {6},
pages = {673\textendash687},
abstract = {In drug discovery, determining the binding affinity and functional effects of small-molecule ligands on proteins is critical. Current computational methods can predict these protein-ligand interaction properties but often lose accuracy without high-resolution protein structures and falter in predicting functional effects. Here we introduce PSICHIC (PhySIcoCHemICal graph neural network), a framework incorporating physicochemical constraints to decode interaction fingerprints directly from sequence data alone. This enables PSICHIC to attain capabilities in decoding mechanisms underlying protein-ligand interactions, achieving state-of-the-art accuracy and interpretability. Trained on identical protein-ligand pairs without structural data, PSICHIC matched and even surpassed leading structure-based methods in binding-affinity prediction. In an experimental library screening for adenosine A1 receptor agonists, PSICHIC discerned functional effects effectively, ranking the sole novel agonist within the top three. PSICHIC’s interpretable fingerprints identified protein residues and ligand atoms involved in interactions, and helped in unveiling selectivity determinants of protein-ligand interaction. We foresee PSICHIC reshaping virtual screening and deepening our understanding of protein-ligand interactions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In drug discovery, determining the binding affinity and functional effects of small-molecule ligands on proteins is critical. Current computational methods can predict these protein-ligand interaction properties but often lose accuracy without high-resolution protein structures and falter in predicting functional effects. Here we introduce PSICHIC (PhySIcoCHemICal graph neural network), a framework incorporating physicochemical constraints to decode interaction fingerprints directly from sequence data alone. This enables PSICHIC to attain capabilities in decoding mechanisms underlying protein-ligand interactions, achieving state-of-the-art accuracy and interpretability. Trained on identical protein-ligand pairs without structural data, PSICHIC matched and even surpassed leading structure-based methods in binding-affinity prediction. In an experimental library screening for adenosine A1 receptor agonists, PSICHIC discerned functional effects effectively, ranking the sole novel agonist within the top three. PSICHIC’s interpretable fingerprints identified protein residues and ligand atoms involved in interactions, and helped in unveiling selectivity determinants of protein-ligand interaction. We foresee PSICHIC reshaping virtual screening and deepening our understanding of protein-ligand interactions.
Foumani, Navid Mohammadi; Tan, Chang Wei; Webb, Geoffrey I.; Rezatofighi, Hamid; Salehi, Mahsa
Series2vec: similarity-based self-supervised representation learning for time series classification
Data Mining and Knowledge Discovery, 2024, ISSN: 1573-756X.
@article{Foumani2024a,
title = {Series2vec: similarity-based self-supervised representation learning for time series classification},
author = {Navid Mohammadi Foumani and Chang Wei Tan and Geoffrey I. Webb and Hamid Rezatofighi and Mahsa Salehi},
doi = {10.1007/s10618-024-01043-w},
issn = {1573-756X},
year = {2024},
date = {2024-01-01},
journal = {Data Mining and Knowledge Discovery},
abstract = {We argue that time series analysis is fundamentally different in nature to either vision or natural language processing with respect to the forms of meaningful self-supervised learning tasks that can be defined. Motivated by this insight, we introduce a novel approach called Series2Vec for self-supervised representation learning. Unlike the state-of-the-art methods in time series which rely on hand-crafted data augmentation, Series2Vec is trained by predicting the similarity between two series in both temporal and spectral domains through a self-supervised task. By leveraging the similarity prediction task, which has inherent meaning for a wide range of time series analysis tasks, Series2Vec eliminates the need for hand-crafted data augmentation. To further enforce the network to learn similar representations for similar time series, we propose a novel approach that applies order-invariant attention to each representation within the batch during training. Our evaluation of Series2Vec on nine large real-world datasets, along with the UCR/UEA archive, shows enhanced performance compared to current state-of-the-art self-supervised techniques for time series. Additionally, our extensive experiments show that Series2Vec performs comparably with fully supervised training and offers high efficiency in datasets with limited-labeled data. Finally, we show that the fusion of Series2Vec with other representation learning models leads to enhanced performance for time series classification. Code and models are open-source at https://github.com/Navidfoumani/Series2Vec},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We argue that time series analysis is fundamentally different in nature to either vision or natural language processing with respect to the forms of meaningful self-supervised learning tasks that can be defined. Motivated by this insight, we introduce a novel approach called Series2Vec for self-supervised representation learning. Unlike the state-of-the-art methods in time series which rely on hand-crafted data augmentation, Series2Vec is trained by predicting the similarity between two series in both temporal and spectral domains through a self-supervised task. By leveraging the similarity prediction task, which has inherent meaning for a wide range of time series analysis tasks, Series2Vec eliminates the need for hand-crafted data augmentation. To further enforce the network to learn similar representations for similar time series, we propose a novel approach that applies order-invariant attention to each representation within the batch during training. Our evaluation of Series2Vec on nine large real-world datasets, along with the UCR/UEA archive, shows enhanced performance compared to current state-of-the-art self-supervised techniques for time series. Additionally, our extensive experiments show that Series2Vec performs comparably with fully supervised training and offers high efficiency in datasets with limited-labeled data. Finally, we show that the fusion of Series2Vec with other representation learning models leads to enhanced performance for time series classification. Code and models are open-source at https://github.com/Navidfoumani/Series2Vec
Jin, Ming; Koh, Huan Yee; Wen, Qingsong; Zambon, Daniele; Alippi, Cesare; Webb, Geoffrey I.; King, Irwin; Pan, Shirui
IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1-20, 2024, ISSN: 1939-3539.
@article{Jin2024,
title = {A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anomaly Detection},
author = {Ming Jin and Huan Yee Koh and Qingsong Wen and Daniele Zambon and Cesare Alippi and Geoffrey I. Webb and Irwin King and Shirui Pan},
doi = {10.1109/tpami.2024.3443141},
issn = {1939-3539},
year = {2024},
date = {2024-01-01},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
pages = {1-20},
publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Lee, Loong Kuan; Webb, Geoffrey I.; Schmidt, Daniel F.; Piatkowski, Nico
Knowledge and Information Systems, vol. 66, pp. 7527-7556, 2024, ISSN: 0219-3116.
@article{Lee2024,
title = {Computing marginal and conditional divergences between decomposable models with applications in quantum computing and earth observation},
author = {Loong Kuan Lee and Geoffrey I. Webb and Daniel F. Schmidt and Nico Piatkowski},
doi = {10.1007/s10115-024-02191-7},
issn = {0219-3116},
year = {2024},
date = {2024-01-01},
journal = {Knowledge and Information Systems},
volume = {66},
pages = {7527-7556},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Li, Shiyuan; Liu, Yixin; Chen, Qingfeng; Webb, Geoffrey I.; Pan, Shirui
Noise-Resilient Unsupervised Graph Representation Learning via Multi-Hop Feature Quality Estimation
Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, pp. 1255-1265, ACM, 2024.
@inproceedings{Li2024,
title = {Noise-Resilient Unsupervised Graph Representation Learning via Multi-Hop Feature Quality Estimation},
author = {Shiyuan Li and Yixin Liu and Qingfeng Chen and Geoffrey I. Webb and Shirui Pan},
doi = {10.1145/3627673.3679758},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 33rd ACM International Conference on Information and Knowledge Management},
pages = {1255-1265},
publisher = {ACM},
series = {CIKM ’24},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Lee, Loong Kuan; Piatkowski, Nico; Petitjean, Francois; Webb, Geoffrey I.
Computing Divergences between Discrete Decomposable Models
Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 10, pp. 12243-12251, 2023.
@article{Lee2023,
title = {Computing Divergences between Discrete Decomposable Models},
author = {Loong Kuan Lee and Nico Piatkowski and Francois Petitjean and Geoffrey I. Webb},
doi = {10.1609/aaai.v37i10.26443},
year = {2023},
date = {2023-06-01},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
volume = {37},
number = {10},
pages = {12243-12251},
abstract = {There are many applications that benefit from computing the exact divergence between 2 discrete probability measures, including machine learning. Unfortunately, in the absence of any assumptions on the structure or independencies within these distributions, computing the divergence between them is an intractable problem in high dimensions. We show that we are able to compute a wide family of functionals and divergences, such as the alpha-beta divergence, between two decomposable models, i.e. chordal Markov networks, in time exponential to the treewidth of these models. The alpha-beta divergence is a family of divergences that include popular divergences such as the Kullback-Leibler divergence, the Hellinger distance, and the chi-squared divergence. Thus, we can accurately compute the exact values of any of this broad class of divergences to the extent to which we can accurately model the two distributions using decomposable models.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
There are many applications that benefit from computing the exact divergence between 2 discrete probability measures, including machine learning. Unfortunately, in the absence of any assumptions on the structure or independencies within these distributions, computing the divergence between them is an intractable problem in high dimensions. We show that we are able to compute a wide family of functionals and divergences, such as the alpha-beta divergence, between two decomposable models, i.e. chordal Markov networks, in time exponential to the treewidth of these models. The alpha-beta divergence is a family of divergences that include popular divergences such as the Kullback-Leibler divergence, the Hellinger distance, and the chi-squared divergence. Thus, we can accurately compute the exact values of any of this broad class of divergences to the extent to which we can accurately model the two distributions using decomposable models.
Jung, Monica; Lukose, Dickson; Nielsen, Suzanne; Bell, J. Simon; Webb, Geoffrey I.; Ilomaki, Jenni
British Journal of Clinical Pharmacology, vol. 89, no. 2, pp. 914-920, 2023.
@article{Jung,
title = {COVID-19 restrictions and the incidence and prevalence of prescription opioid use in Australia - a nation-wide study},
author = {Monica Jung and Dickson Lukose and Suzanne Nielsen and J. Simon Bell and Geoffrey I. Webb and Jenni Ilomaki},
doi = {10.1111/bcp.15577},
year = {2023},
date = {2023-01-01},
journal = {British Journal of Clinical Pharmacology},
volume = {89},
number = {2},
pages = {914-920},
abstract = {The COVID-19 pandemic has disrupted seeking and delivery of healthcare. Different Australian jurisdictions implemented different COVID-19 restrictions. We used Australian national pharmacy dispensing data to conduct interrupted time series analyses to examine the incidence and prevalence of opioid dispensing in different jurisdictions. Following nationwide COVID-19 restrictions, the incidence dropped by -0.40 [-0.50, -0.31], -0.33 [-0.46, -0.21] and -0.21 [-0.37, -0.04] /1000 people/week and prevalence dropped by -0.85 [-1.39, -0.31], -0.54 [-1.01, -0.07] and -0.62 [-0.99, -0.25] /1000 people/week in Victoria, New South Wales and other jurisdictions, respectively. Incidence and prevalence increased by 0.29 [0.13, 0.44] and 0.72 [0.11, 1.33] /1000 people/week, respectively in Victoria post-lockdown; no significant changes were observed in other jurisdictions. No significant changes were observed in the initiation of long-term opioid use in any jurisdictions. More stringent restrictions coincided with more pronounced reductions in overall opioid initiation, but initiation of long-term opioid use did not change.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The COVID-19 pandemic has disrupted seeking and delivery of healthcare. Different Australian jurisdictions implemented different COVID-19 restrictions. We used Australian national pharmacy dispensing data to conduct interrupted time series analyses to examine the incidence and prevalence of opioid dispensing in different jurisdictions. Following nationwide COVID-19 restrictions, the incidence dropped by -0.40 [-0.50, -0.31], -0.33 [-0.46, -0.21] and -0.21 [-0.37, -0.04] /1000 people/week and prevalence dropped by -0.85 [-1.39, -0.31], -0.54 [-1.01, -0.07] and -0.62 [-0.99, -0.25] /1000 people/week in Victoria, New South Wales and other jurisdictions, respectively. Incidence and prevalence increased by 0.29 [0.13, 0.44] and 0.72 [0.11, 1.33] /1000 people/week, respectively in Victoria post-lockdown; no significant changes were observed in other jurisdictions. No significant changes were observed in the initiation of long-term opioid use in any jurisdictions. More stringent restrictions coincided with more pronounced reductions in overall opioid initiation, but initiation of long-term opioid use did not change.
Tan, Chang Wei; Herrmann, Matthieu; Webb, Geoffrey I.
Knowledge and Information Systems, vol. 65, pp. 2123-2157, 2023.
@article{Tan2023,
title = {Ultra-fast meta-parameter optimization for time series similarity measures with application to nearest neighbour classification},
author = {Chang Wei Tan and Matthieu Herrmann and Geoffrey I. Webb},
doi = {10.1007/s10115-022-01827-w},
year = {2023},
date = {2023-01-01},
journal = {Knowledge and Information Systems},
volume = {65},
pages = {2123-2157},
publisher = {Springer Science and Business Media LLC},
abstract = {Nearest neighbour similarity measures are widely used in many time series data analysis applications. They compute a measure of similarity between two time series. Most applications require tuning of these measures’ meta-parameters in order to achieve good performance. However, most measures have at least complexity O(L^2), making them computationally expensive and the process of learning their meta-parameters burdensome, requiring days even for datasets containing only a few thousand series. In this paper, we propose UltraFastMPSearch, a family of algorithms to learn the meta-parameters for different types of time series distance measures. These algorithms are significantly faster than the prior state of the art. Our algorithms build upon the state of the art, exploiting the properties of a new efficient exact algorithm which supports early abandoning and pruning for most time series distance measures. We show on 128 datasets from the UCR archive that our new family of algorithms are up to an order of magnitude faster than the previous state of the art.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Nearest neighbour similarity measures are widely used in many time series data analysis applications. They compute a measure of similarity between two time series. Most applications require tuning of these measures’ meta-parameters in order to achieve good performance. However, most measures have at least complexity O(L^2), making them computationally expensive and the process of learning their meta-parameters burdensome, requiring days even for datasets containing only a few thousand series. In this paper, we propose UltraFastMPSearch, a family of algorithms to learn the meta-parameters for different types of time series distance measures. These algorithms are significantly faster than the prior state of the art. Our algorithms build upon the state of the art, exploiting the properties of a new efficient exact algorithm which supports early abandoning and pruning for most time series distance measures. We show on 128 datasets from the UCR archive that our new family of algorithms are up to an order of magnitude faster than the previous state of the art.
Shifaz, Ahmed; Pelletier, Charlotte; Petitjean, François; Webb, Geoffrey I.
Elastic similarity and distance measures for multivariate time series
Knowledge and Information Systems, vol. 65, pp. 2665-2698, 2023.
@article{Shifaz2023,
title = {Elastic similarity and distance measures for multivariate time series},
author = {Ahmed Shifaz and Charlotte Pelletier and Fran\c{c}ois Petitjean and Geoffrey I. Webb},
doi = {10.1007/s10115-023-01835-4},
year = {2023},
date = {2023-01-01},
journal = {Knowledge and Information Systems},
volume = {65},
pages = {2665-2698},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Pan, Tong; Li, Chen; Bi, Yue; Wang, Zhikang; Gasser, Robin B; Purcell, Anthony W; Akutsu, Tatsuya; Webb, Geoffrey I; Imoto, Seiya; Song, Jiangning
Bioinformatics, 2023, ISSN: 1367-4811.
@article{10.1093/bioinformatics/btad094,
title = {PFresGO: an attention mechanism-based deep-learning approach for protein annotation by integrating gene ontology inter-relationships},
author = {Tong Pan and Chen Li and Yue Bi and Zhikang Wang and Robin B Gasser and Anthony W Purcell and Tatsuya Akutsu and Geoffrey I Webb and Seiya Imoto and Jiangning Song},
doi = {10.1093/bioinformatics/btad094},
issn = {1367-4811},
year = {2023},
date = {2023-01-01},
journal = {Bioinformatics},
abstract = {The rapid accumulation of high-throughput sequence data demands the development of effective and efficient data-driven computational methods to functionally annotate proteins. However, most current approaches used for functional annotation simply focus on the use of protein-level information but ignore inter-relationships among annotations.Here, we established PFresGO, an attention-based deep-learning approach that incorporates hierarchical structures in Gene Ontology (GO) graphs and advances in natural language processing algorithms for the functional annotation of proteins. PFresGO employs a self-attention operation to capture the inter-relationships of GO terms, updates its embedding accordingly, and uses a cross-attention operation to project protein representations and GO embedding into a common latent space to identify global protein sequence patterns and local functional residues. We demonstrate that PFresGO consistently achieves superior performance across GO categories when compared with state-of-the-art methods. Importantly, we show that PFresGO can identify functionally important residues in protein sequences by assessing the distribution of attention weightings. PFresGO should serve as an effective tool for the accurate functional annotation of proteins and functional domains within proteins.PFresGO is available for academic purposes at https://github.com/BioColLab/PFresGO.Supplementary data are available at Bioinformatics online.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The rapid accumulation of high-throughput sequence data demands the development of effective and efficient data-driven computational methods to functionally annotate proteins. However, most current approaches used for functional annotation simply focus on the use of protein-level information but ignore inter-relationships among annotations.Here, we established PFresGO, an attention-based deep-learning approach that incorporates hierarchical structures in Gene Ontology (GO) graphs and advances in natural language processing algorithms for the functional annotation of proteins. PFresGO employs a self-attention operation to capture the inter-relationships of GO terms, updates its embedding accordingly, and uses a cross-attention operation to project protein representations and GO embedding into a common latent space to identify global protein sequence patterns and local functional residues. We demonstrate that PFresGO consistently achieves superior performance across GO categories when compared with state-of-the-art methods. Importantly, we show that PFresGO can identify functionally important residues in protein sequences by assessing the distribution of attention weightings. PFresGO should serve as an effective tool for the accurate functional annotation of proteins and functional domains within proteins.PFresGO is available for academic purposes at https://github.com/BioColLab/PFresGO.Supplementary data are available at Bioinformatics online.
Godahewa, Rakshitha; Webb, Geoffrey I.; Schmidt, Daniel; Bergmeir, Christoph
SETAR-Tree: a novel and accurate tree algorithm for global time series forecasting
Machine Learning, vol. 112, pp. 2555-2591, 2023, ISSN: 1573-0565.
@article{Godahewa2023,
title = {SETAR-Tree: a novel and accurate tree algorithm for global time series forecasting},
author = {Rakshitha Godahewa and Geoffrey I. Webb and Daniel Schmidt and Christoph Bergmeir},
doi = {10.1007/s10994-023-06316-x},
issn = {1573-0565},
year = {2023},
date = {2023-01-01},
journal = {Machine Learning},
volume = {112},
pages = {2555-2591},
abstract = {Threshold Autoregressive (TAR) models have been widely used by statisticians for non-linear time series forecasting during the past few decades, due to their simplicity and mathematical properties. On the other hand, in the forecasting community, general-purpose tree-based regression algorithms (forests, gradient-boosting) have become popular recently due to their ease of use and accuracy. In this paper, we explore the close connections between TAR models and regression trees. These enable us to use the rich methodology from the literature on TAR models to define a hierarchical TAR model as a regression tree that trains globally across series, which we call SETAR-Tree. In contrast to the general-purpose tree-based models that do not primarily focus on forecasting, and calculate averages at the leaf nodes, we introduce a new forecasting-specific tree algorithm that trains global Pooled Regression (PR) models in the leaves allowing the models to learn cross-series information and also uses some time-series-specific splitting and stopping procedures. The depth of the tree is controlled by conducting a statistical linearity test commonly employed in TAR models, as well as measuring the error reduction percentage at each node split. Thus, the proposed tree model requires minimal external hyperparameter tuning and provides competitive results under its default configuration. We also use this tree algorithm to develop a forest where the forecasts provided by a collection of diverse SETAR-Trees are combined during the forecasting process. In our evaluation on eight publicly available datasets, the proposed tree and forest models are able to achieve significantly higher accuracy than a set of state-of-the-art tree-based algorithms and forecasting benchmarks across four evaluation metrics.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Threshold Autoregressive (TAR) models have been widely used by statisticians for non-linear time series forecasting during the past few decades, due to their simplicity and mathematical properties. On the other hand, in the forecasting community, general-purpose tree-based regression algorithms (forests, gradient-boosting) have become popular recently due to their ease of use and accuracy. In this paper, we explore the close connections between TAR models and regression trees. These enable us to use the rich methodology from the literature on TAR models to define a hierarchical TAR model as a regression tree that trains globally across series, which we call SETAR-Tree. In contrast to the general-purpose tree-based models that do not primarily focus on forecasting, and calculate averages at the leaf nodes, we introduce a new forecasting-specific tree algorithm that trains global Pooled Regression (PR) models in the leaves allowing the models to learn cross-series information and also uses some time-series-specific splitting and stopping procedures. The depth of the tree is controlled by conducting a statistical linearity test commonly employed in TAR models, as well as measuring the error reduction percentage at each node split. Thus, the proposed tree model requires minimal external hyperparameter tuning and provides competitive results under its default configuration. We also use this tree algorithm to develop a forest where the forecasts provided by a collection of diverse SETAR-Trees are combined during the forecasting process. In our evaluation on eight publicly available datasets, the proposed tree and forest models are able to achieve significantly higher accuracy than a set of state-of-the-art tree-based algorithms and forecasting benchmarks across four evaluation metrics.
Miller, Lynn; Zhu, Liujun; Yebra, Marta; Rudiger, Christoph; Webb, Geoffrey I
Projecting live fuel moisture content via deep learning
International Journal of Wildland Fire, 2023.
@article{Miller2023Projecting,
title = {Projecting live fuel moisture content via deep learning},
author = {Lynn Miller and Liujun Zhu and Marta Yebra and Christoph Rudiger and Geoffrey I Webb},
doi = {10.1071/WF22188},
year = {2023},
date = {2023-01-01},
journal = {International Journal of Wildland Fire},
abstract = {Background: Live fuel moisture content (LFMC) is a key environmental indicator used to monitor for high wildfire risk conditions. Many statistical models have been proposed to predict LFMC from remotely sensed data; however, almost all these estimate current LFMC (nowcasting models). Accurate modelling of LFMC in advance (projection models) would provide wildfire managers with more timely information for assessing and preparing for wildfire risk.
Aims: The aim of this study was to investigate the potential for deep learning models to predict LFMC across the continental United States 3 months in advance.
Method: Temporal convolutional networks were trained and evaluated using a large database of field measured samples, as well as year-long time series of MODerate resolution Imaging Spectroradiometer (MODIS) reflectance data and Parameter-elevation Relationships on Independent Slopes Model (PRISM) meteorological data.
Key results: The proposed 3-month projection model achieved an accuracy (root mean squared error (RMSE) 27.52%; R2 0.47) close to that of the nowcasting model (RMSE 26.52%; R2 0.51).
Conclusions: The study is the first to predict LFMC with a 3-month lead-time, demonstrating the potential for deep learning models to make reliable LFMC projections.
Implications: These findings are beneficial for wildfire management and risk assessment, showing proof-of-concept for providing advance information useful to help mitigate the effect of catastrophic wildfires.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Background: Live fuel moisture content (LFMC) is a key environmental indicator used to monitor for high wildfire risk conditions. Many statistical models have been proposed to predict LFMC from remotely sensed data; however, almost all these estimate current LFMC (nowcasting models). Accurate modelling of LFMC in advance (projection models) would provide wildfire managers with more timely information for assessing and preparing for wildfire risk.
Aims: The aim of this study was to investigate the potential for deep learning models to predict LFMC across the continental United States 3 months in advance.
Method: Temporal convolutional networks were trained and evaluated using a large database of field measured samples, as well as year-long time series of MODerate resolution Imaging Spectroradiometer (MODIS) reflectance data and Parameter-elevation Relationships on Independent Slopes Model (PRISM) meteorological data.
Key results: The proposed 3-month projection model achieved an accuracy (root mean squared error (RMSE) 27.52%; R2 0.47) close to that of the nowcasting model (RMSE 26.52%; R2 0.51).
Conclusions: The study is the first to predict LFMC with a 3-month lead-time, demonstrating the potential for deep learning models to make reliable LFMC projections.
Implications: These findings are beneficial for wildfire management and risk assessment, showing proof-of-concept for providing advance information useful to help mitigate the effect of catastrophic wildfires.
Aims: The aim of this study was to investigate the potential for deep learning models to predict LFMC across the continental United States 3 months in advance.
Method: Temporal convolutional networks were trained and evaluated using a large database of field measured samples, as well as year-long time series of MODerate resolution Imaging Spectroradiometer (MODIS) reflectance data and Parameter-elevation Relationships on Independent Slopes Model (PRISM) meteorological data.
Key results: The proposed 3-month projection model achieved an accuracy (root mean squared error (RMSE) 27.52%; R2 0.47) close to that of the nowcasting model (RMSE 26.52%; R2 0.51).
Conclusions: The study is the first to predict LFMC with a 3-month lead-time, demonstrating the potential for deep learning models to make reliable LFMC projections.
Implications: These findings are beneficial for wildfire management and risk assessment, showing proof-of-concept for providing advance information useful to help mitigate the effect of catastrophic wildfires.
Zhang, Huan; Jiang, Liangxiao; Webb, Geoffrey I.
Rigorous non-disjoint discretization for naive Bayes
Pattern Recognition, vol. 140, 2023, ISSN: 0031-3203.
@article{Zhang2023,
title = {Rigorous non-disjoint discretization for naive Bayes},
author = {Huan Zhang and Liangxiao Jiang and Geoffrey I. Webb},
doi = {10.1016/j.patcog.2023.109554},
issn = {0031-3203},
year = {2023},
date = {2023-01-01},
journal = {Pattern Recognition},
volume = {140},
abstract = {Naive Bayes is a classical machine learning algorithm for which discretization is commonly used to transform quantitative attributes into qualitative attributes. Of numerous discretization methods, Non-Disjoint Discretization (NDD) proposes a novel perspective by forming overlapping intervals and always locating a value toward the middle of an interval. However, existing approaches to NDD fail to adequately consider the effect of multiple occurrences of a single value - a commonly occurring circumstance in practice. By necessity, all occurrences of a single value fall within the same interval. As a result, it is often not possible to discretize an attribute into intervals containing equal numbers of training instances. Current methods address this issue in an ad hoc manner, reducing the specificity of the resulting atomic intervals. In this study, we propose a non-disjoint discretization method for NB, called Rigorous Non-Disjoint Discretization (RNDD), that handles multiple occurrences of a single value in a systematic manner. Our extensive experimental results suggest that RNDD significantly outperforms NDD along with all other existing state-of-the-art competitors.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Naive Bayes is a classical machine learning algorithm for which discretization is commonly used to transform quantitative attributes into qualitative attributes. Of numerous discretization methods, Non-Disjoint Discretization (NDD) proposes a novel perspective by forming overlapping intervals and always locating a value toward the middle of an interval. However, existing approaches to NDD fail to adequately consider the effect of multiple occurrences of a single value - a commonly occurring circumstance in practice. By necessity, all occurrences of a single value fall within the same interval. As a result, it is often not possible to discretize an attribute into intervals containing equal numbers of training instances. Current methods address this issue in an ad hoc manner, reducing the specificity of the resulting atomic intervals. In this study, we propose a non-disjoint discretization method for NB, called Rigorous Non-Disjoint Discretization (RNDD), that handles multiple occurrences of a single value in a systematic manner. Our extensive experimental results suggest that RNDD significantly outperforms NDD along with all other existing state-of-the-art competitors.
Herrmann, Matthieu; Webb, Geoffrey I.
Amercing: An Intuitive and Effective Constraint for Dynamic Time Warping
Pattern Recognition, vol. 137, 2023, ISSN: 0031-3203.
@article{Herrmann2023a,
title = {Amercing: An Intuitive and Effective Constraint for Dynamic Time Warping},
author = {Matthieu Herrmann and Geoffrey I. Webb},
doi = {10.1016/j.patcog.2023.109333},
issn = {0031-3203},
year = {2023},
date = {2023-01-01},
journal = {Pattern Recognition},
volume = {137},
abstract = {Dynamic Time Warping (DTW) is a time series distance measure that allows non-linear alignments between series. Constraints on the alignments in the form of windows and weights have been introduced because unconstrained DTW is too permissive in its alignments. However, windowing introduces a crude step function, allowing unconstrained flexibility within the window, and none beyond it. While not entailing a step function, a multiplicative weight is relative to the distances between aligned points along a warped path, rather than being a direct function of the amount of warping that is introduced. In this paper, we introduce Amerced Dynamic Time Warping (ADTW), a new, intuitive, DTW variant that penalizes the act of warping by a fixed additive cost. Like windowing and weighting, ADTW constrains the amount of warping. However, it avoids both abrupt discontinuities in the amount of warping allowed and the limitations of a multiplicative penalty. We formally introduce ADTW, prove some of its properties, and discuss its parameterization. We show on a simple example how it can be parameterized to achieve an intuitive outcome, and demonstrate its usefulness on a standard time series classification benchmark. We provide a demonstration application in C++ [1].},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Dynamic Time Warping (DTW) is a time series distance measure that allows non-linear alignments between series. Constraints on the alignments in the form of windows and weights have been introduced because unconstrained DTW is too permissive in its alignments. However, windowing introduces a crude step function, allowing unconstrained flexibility within the window, and none beyond it. While not entailing a step function, a multiplicative weight is relative to the distances between aligned points along a warped path, rather than being a direct function of the amount of warping that is introduced. In this paper, we introduce Amerced Dynamic Time Warping (ADTW), a new, intuitive, DTW variant that penalizes the act of warping by a fixed additive cost. Like windowing and weighting, ADTW constrains the amount of warping. However, it avoids both abrupt discontinuities in the amount of warping allowed and the limitations of a multiplicative penalty. We formally introduce ADTW, prove some of its properties, and discuss its parameterization. We show on a simple example how it can be parameterized to achieve an intuitive outcome, and demonstrate its usefulness on a standard time series classification benchmark. We provide a demonstration application in C++ [1].
Herrmann, Matthieu; Tan, Chang Wei; Webb, Geoffrey I.
Data Mining and Knowledge Discovery, vol. 37, pp. 2024-2045, 2023.
@article{Herrmann2023Parameterizing,
title = {Parameterizing the cost function of dynamic time warping with application to time series classification},
author = {Matthieu Herrmann and Chang Wei Tan and Geoffrey I. Webb},
doi = {10.1007/s10618-023-00926-8},
year = {2023},
date = {2023-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {37},
pages = {2024-2045},
abstract = {Dynamic time warping (DTW) is a popular time series distance measure that aligns the points in two series with one another. These alignments support warping of the time dimension to allow for processes that unfold at differing rates. The distance is the minimum sum of costs of the resulting alignments over any allowable warping of the time dimension. The cost of an alignment of two points is a function of the difference in the values of those points. The original cost function was the absolute value of this difference. Other cost functions have been proposed. A popular alternative is the square of the difference. However, to our knowledge, this is the first investigation of both the relative impacts of using different cost functions and the potential to tune cost functions to different time series classification tasks. We do so in this paper by using a tunable cost function ?? with parameter ?. We show that higher values of ? place greater weight on larger pairwise differences, while lower values place greater weight on smaller pairwise differences. We demonstrate that training ? significantly improves the accuracy of both the DTW nearest neighbor and Proximity Forest classifiers.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Dynamic time warping (DTW) is a popular time series distance measure that aligns the points in two series with one another. These alignments support warping of the time dimension to allow for processes that unfold at differing rates. The distance is the minimum sum of costs of the resulting alignments over any allowable warping of the time dimension. The cost of an alignment of two points is a function of the difference in the values of those points. The original cost function was the absolute value of this difference. Other cost functions have been proposed. A popular alternative is the square of the difference. However, to our knowledge, this is the first investigation of both the relative impacts of using different cost functions and the potential to tune cost functions to different time series classification tasks. We do so in this paper by using a tunable cost function ?? with parameter ?. We show that higher values of ? place greater weight on larger pairwise differences, while lower values place greater weight on smaller pairwise differences. We demonstrate that training ? significantly improves the accuracy of both the DTW nearest neighbor and Proximity Forest classifiers.
Zhu, Yan; Li, Fuyi; Guo, Xudong; Wang, Xiaoyu; Coin, Lachlan J M; Webb, Geoffrey I; Song, Jiangning; Jia, Cangzhi
Briefings in Bioinformatics, vol. 24, no. 4, 2023, ISSN: 1477-4054.
@article{Zhu2023,
title = {TIMER is a Siamese neural network-based framework for identifying both general and species-specific bacterial promoters},
author = {Yan Zhu and Fuyi Li and Xudong Guo and Xiaoyu Wang and Lachlan J M Coin and Geoffrey I Webb and Jiangning Song and Cangzhi Jia},
doi = {10.1093/bib/bbad209},
issn = {1477-4054},
year = {2023},
date = {2023-01-01},
journal = {Briefings in Bioinformatics},
volume = {24},
number = {4},
abstract = {Promoters are DNA regions that initiate the transcription of specific genes near the transcription start sites. In bacteria, promoters are recognized by RNA polymerases and associated sigma factors. Effective promoter recognition is essential for synthesizing the gene-encoded products by bacteria to grow and adapt to different environmental conditions. A variety of machine learning-based predictors for bacterial promoters have been developed; however, most of them were designed specifically for a particular species. To date, only a few predictors are available for identifying general bacterial promoters with limited predictive performance.In this study, we developed TIMER, a Siamese neural network-based approach for identifying both general and species-specific bacterial promoters. Specifically, TIMER uses DNA sequences as the input and employs three Siamese neural networks with the attention layers to train and optimize the models for a total of 13 species-specific and general bacterial promoters. Extensive 10-fold cross-validation and independent tests demonstrated that TIMER achieves a competitive performance and outperforms several existing methods on both general and species-specific promoter prediction. As an implementation of the proposed method, the web server of TIMER is publicly accessible at http://web.unimelb-bioinfortools.cloud.edu.au/TIMER/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Promoters are DNA regions that initiate the transcription of specific genes near the transcription start sites. In bacteria, promoters are recognized by RNA polymerases and associated sigma factors. Effective promoter recognition is essential for synthesizing the gene-encoded products by bacteria to grow and adapt to different environmental conditions. A variety of machine learning-based predictors for bacterial promoters have been developed; however, most of them were designed specifically for a particular species. To date, only a few predictors are available for identifying general bacterial promoters with limited predictive performance.In this study, we developed TIMER, a Siamese neural network-based approach for identifying both general and species-specific bacterial promoters. Specifically, TIMER uses DNA sequences as the input and employs three Siamese neural networks with the attention layers to train and optimize the models for a total of 13 species-specific and general bacterial promoters. Extensive 10-fold cross-validation and independent tests demonstrated that TIMER achieves a competitive performance and outperforms several existing methods on both general and species-specific promoter prediction. As an implementation of the proposed method, the web server of TIMER is publicly accessible at http://web.unimelb-bioinfortools.cloud.edu.au/TIMER/.
Dempster, Angus; Schmidt, Daniel F.; Webb, Geoffrey I.
HYDRA: competing convolutional kernels for fast and accurate time series classification
Data Mining and Knowledge Discovery, vol. 37, no. 5, pp. 1779-1805, 2023.
@article{Dempster2023,
title = {HYDRA: competing convolutional kernels for fast and accurate time series classification},
author = {Angus Dempster and Daniel F. Schmidt and Geoffrey I. Webb},
doi = {10.1007/s10618-023-00939-3},
year = {2023},
date = {2023-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {37},
number = {5},
pages = {1779-1805},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ramakrishnaiah, Yashpal; Macesic, Nenad; Webb, Geoff; Peleg, Anton Y.; Tyagi, Sonika
Journal of Biomedical Informatics, pp. 104509, 2023, ISSN: 1532-0464.
@article{Ramakrishnaiah2023,
title = {EHR-QC: A streamlined pipeline for automated electronic health records standardisation and preprocessing to predict clinical outcomes},
author = {Yashpal Ramakrishnaiah and Nenad Macesic and Geoff Webb and Anton Y. Peleg and Sonika Tyagi},
doi = {10.1016/j.jbi.2023.104509},
issn = {1532-0464},
year = {2023},
date = {2023-01-01},
journal = {Journal of Biomedical Informatics},
pages = {104509},
abstract = {The adoption of electronic health records (EHRs) has created opportunities to analyze historical data for predicting clinical outcomes and improving patient care. However, non-standardized data representations and anomalies pose major challenges to the use of EHRs in digital health research. To address these challenges, we have developed EHR-QC, a tool comprising two modules: the data standardization module and the preprocessing module. The data standardization module migrates source EHR data to a standard format using advanced concept mapping techniques, surpassing expert curation in benchmarking analysis. The preprocessing module includes several functions designed specifically to handle healthcare data subtleties. We provide automated detection of data anomalies and solutions to handle those anomalies. We believe that the development and adoption of tools like EHR-QC is critical for advancing digital health. Our ultimate goal is to accelerate clinical research by enabling rapid experimentation with data-driven observational research to generate robust, generalisable biomedical knowledge.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The adoption of electronic health records (EHRs) has created opportunities to analyze historical data for predicting clinical outcomes and improving patient care. However, non-standardized data representations and anomalies pose major challenges to the use of EHRs in digital health research. To address these challenges, we have developed EHR-QC, a tool comprising two modules: the data standardization module and the preprocessing module. The data standardization module migrates source EHR data to a standard format using advanced concept mapping techniques, surpassing expert curation in benchmarking analysis. The preprocessing module includes several functions designed specifically to handle healthcare data subtleties. We provide automated detection of data anomalies and solutions to handle those anomalies. We believe that the development and adoption of tools like EHR-QC is critical for advancing digital health. Our ultimate goal is to accelerate clinical research by enabling rapid experimentation with data-driven observational research to generate robust, generalisable biomedical knowledge.
Pialla, Gautier; Fawaz, Hassan Ismail; Devanne, Maxime; Weber, Jonathan; Idoumghar, Lhassane; Muller, Pierre-Alain; Bergmeir, Christoph; Schmidt, Daniel F.; Webb, Geoffrey I.; Forestier, Germain
Time series adversarial attacks: an investigation of smooth perturbations and defense approaches
International Journal of Data Science and Analytics, 2023.
@article{Pialla2023,
title = {Time series adversarial attacks: an investigation of smooth perturbations and defense approaches},
author = {Gautier Pialla and Hassan Ismail Fawaz and Maxime Devanne and Jonathan Weber and Lhassane Idoumghar and Pierre-Alain Muller and Christoph Bergmeir and Daniel F. Schmidt and Geoffrey I. Webb and Germain Forestier},
doi = {10.1007/s41060-023-00438-0},
year = {2023},
date = {2023-01-01},
journal = {International Journal of Data Science and Analytics},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Li, Fuyi; Wang, Cong; Guo, Xudong; Akutsu, Tatsuya; Webb, Geoffrey I; Coin, Lachlan J M; Kurgan, Lukasz; Song, Jiangning
Briefings in Bioinformatics, vol. 24, no. 6, 2023.
@article{Li2023,
title = {ProsperousPlus: a one-stop and comprehensive platform for accurate protease-specific substrate cleavage prediction and machine-learning model construction},
author = {Fuyi Li and Cong Wang and Xudong Guo and Tatsuya Akutsu and Geoffrey I Webb and Lachlan J M Coin and Lukasz Kurgan and Jiangning Song},
doi = {10.1093/bib/bbad372},
year = {2023},
date = {2023-01-01},
journal = {Briefings in Bioinformatics},
volume = {24},
number = {6},
publisher = {Oxford University Press (OUP)},
abstract = {Proteases contribute to a broad spectrum of cellular functions. Given a relatively limited amount of experimental data, developing accurate sequence-based predictors of substrate cleavage sites facilitates a better understanding of protease functions and substrate specificity. While many protease-specific predictors of substrate cleavage sites were developed, these efforts are outpaced by the growth of the protease substrate cleavage data. In particular, since data for 100+ protease types are available and this number continues to grow, it becomes impractical to publish predictors for new protease types, and instead it might be better to provide a computational platform that helps users to quickly and efficiently build predictors that address their specific needs. To this end, we conceptualized, developed, tested and released a versatile bioinformatics platform, ProsperousPlus, that empowers users, even those with no programming or little bioinformatics background, to build fast and accurate predictors of substrate cleavage sites. ProsperousPlus facilitates the use of the rapidly accumulating substrate cleavage data to train, empirically assess and deploy predictive models for user-selected substrate types. Benchmarking tests on test datasets show that our platform produces predictors that on average exceed the predictive performance of current state-of-the-art approaches. ProsperousPlus is available as a webserver and a stand-alone software package at http://prosperousplus.unimelb-biotools.cloud.edu.au/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Proteases contribute to a broad spectrum of cellular functions. Given a relatively limited amount of experimental data, developing accurate sequence-based predictors of substrate cleavage sites facilitates a better understanding of protease functions and substrate specificity. While many protease-specific predictors of substrate cleavage sites were developed, these efforts are outpaced by the growth of the protease substrate cleavage data. In particular, since data for 100+ protease types are available and this number continues to grow, it becomes impractical to publish predictors for new protease types, and instead it might be better to provide a computational platform that helps users to quickly and efficiently build predictors that address their specific needs. To this end, we conceptualized, developed, tested and released a versatile bioinformatics platform, ProsperousPlus, that empowers users, even those with no programming or little bioinformatics background, to build fast and accurate predictors of substrate cleavage sites. ProsperousPlus facilitates the use of the rapidly accumulating substrate cleavage data to train, empirically assess and deploy predictive models for user-selected substrate types. Benchmarking tests on test datasets show that our platform produces predictors that on average exceed the predictive performance of current state-of-the-art approaches. ProsperousPlus is available as a webserver and a stand-alone software package at http://prosperousplus.unimelb-biotools.cloud.edu.au/.
Ismail-Fawaz, Ali; Fawaz, Hassan Ismail; Petitjean, François; Devanne, Maxime; Weber, Jonathan; Berretti, Stefano; Webb, Geoffrey I.; Forestier, Germain
ShapeDBA: Generating Effective Time Series Prototypes Using ShapeDTW Barycenter Averaging
Ifrim, Georgiana; Tavenard, Romain; Bagnall, Anthony; Schaefer, Patrick; Malinowski, Simon; Guyet, Thomas; Lemaire, Vincent (Ed.): Advanced Analytics and Learning on Temporal Data, pp. 127–142, Springer Nature Switzerland, Cham, 2023, ISBN: 978-3-031-49896-1.
@inproceedings{IsmailFawaz2023,
title = {ShapeDBA: Generating Effective Time Series Prototypes Using ShapeDTW Barycenter Averaging},
author = {Ali Ismail-Fawaz and Hassan Ismail Fawaz and Fran\c{c}ois Petitjean and Maxime Devanne and Jonathan Weber and Stefano Berretti and Geoffrey I. Webb and Germain Forestier},
editor = {Georgiana Ifrim and Romain Tavenard and Anthony Bagnall and Patrick Schaefer and Simon Malinowski and Thomas Guyet and Vincent Lemaire},
isbn = {978-3-031-49896-1},
year = {2023},
date = {2023-01-01},
booktitle = {Advanced Analytics and Learning on Temporal Data},
pages = {127\textendash142},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {Time series data can be found in almost every domain, ranging from the medical field to manufacturing and wireless communication. Generating realistic and useful exemplars and prototypes is a fundamental data analysis task. In this paper, we investigate a novel approach to generating realistic and useful exemplars and prototypes for time series data. Our approach uses a new form of time series average, the ShapeDTW Barycentric Average. We therefore turn our attention to accurately generating time series prototypes with a novel approach. The existing time series prototyping approaches rely on the Dynamic Time Warping (DTW) similarity measure such as DTW Barycentering Average (DBA) and SoftDBA. These last approaches suffer from a common problem of generating out-of-distribution artifacts in their prototypes. This is mostly caused by the DTW variant used and its incapability of detecting neighborhood similarities, instead it detects absolute similarities. Our proposed method, ShapeDBA, uses the ShapeDTW variant of DTW, that overcomes this issue. We chose time series clustering, a popular form of time series analysis to evaluate the outcome of ShapeDBA compared to the other prototyping approaches. Coupled with the k-means clustering algorithm, and evaluated on a total of 123 datasets from the UCR archive, our proposed averaging approach is able to achieve new state-of-the-art results in terms of Adjusted Rand Index.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Time series data can be found in almost every domain, ranging from the medical field to manufacturing and wireless communication. Generating realistic and useful exemplars and prototypes is a fundamental data analysis task. In this paper, we investigate a novel approach to generating realistic and useful exemplars and prototypes for time series data. Our approach uses a new form of time series average, the ShapeDTW Barycentric Average. We therefore turn our attention to accurately generating time series prototypes with a novel approach. The existing time series prototyping approaches rely on the Dynamic Time Warping (DTW) similarity measure such as DTW Barycentering Average (DBA) and SoftDBA. These last approaches suffer from a common problem of generating out-of-distribution artifacts in their prototypes. This is mostly caused by the DTW variant used and its incapability of detecting neighborhood similarities, instead it detects absolute similarities. Our proposed method, ShapeDBA, uses the ShapeDTW variant of DTW, that overcomes this issue. We chose time series clustering, a popular form of time series analysis to evaluate the outcome of ShapeDBA compared to the other prototyping approaches. Coupled with the k-means clustering algorithm, and evaluated on a total of 123 datasets from the UCR archive, our proposed averaging approach is able to achieve new state-of-the-art results in terms of Adjusted Rand Index.
Lee, Loong Kuan; Webb, Geoffrey I; Schmidt, Daniel F; Piatkowski, Nico
Computing Marginal and Conditional Divergences between Decomposable Models with Applications
Proceedings of the IEEE International Conference on Data Mining (ICDM), pp. 239-248, IEEE 2023.
@inproceedings{Lee2023a,
title = {Computing Marginal and Conditional Divergences between Decomposable Models with Applications},
author = {Loong Kuan Lee and Geoffrey I Webb and Daniel F Schmidt and Nico Piatkowski},
doi = {10.1109/ICDM58522.2023.00033},
year = {2023},
date = {2023-01-01},
booktitle = {Proceedings of the IEEE International Conference on Data Mining (ICDM)},
pages = {239-248},
organization = {IEEE},
abstract = {The ability to compute the exact divergence between two high-dimensional distributions is useful in many applications but doing so naively is intractable. Computing the alpha-beta divergence\textemdasha family of divergences that includes the Kullback-Leibler divergence and Hellinger distance\textemdashbetween the joint distribution of two decomposable models, i.e chordal Markov networks, can be done in time exponential in the treewidth of these models. However, reducing the dissimilarity between two high-dimensional objects to a single scalar value can be uninformative. Furthermore, in applications such as supervised learning, the divergence over a conditional distribution might be of more interest. Therefore, we propose an approach to compute the exact alpha-beta divergence between any marginal or conditional distribution of two decomposable models. Doing so tractably is non-trivial as we need to decompose the divergence between these distributions and therefore, require a decomposition over the marginal and conditional distributions of these models. Consequently, we provide such a decomposition and also extend existing work to compute the marginal and conditional alpha-beta divergence between these decompositions. We then show how our method can be used to analyze distributional changes by first applying it to a benchmark image dataset. Finally, based on our framework, we propose a novel way to quantify the error in contemporary superconducting quantum computers. Code for all experiments is available at: https://lklee.dev/pub/2023-icdm/code.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The ability to compute the exact divergence between two high-dimensional distributions is useful in many applications but doing so naively is intractable. Computing the alpha-beta divergence—a family of divergences that includes the Kullback-Leibler divergence and Hellinger distance—between the joint distribution of two decomposable models, i.e chordal Markov networks, can be done in time exponential in the treewidth of these models. However, reducing the dissimilarity between two high-dimensional objects to a single scalar value can be uninformative. Furthermore, in applications such as supervised learning, the divergence over a conditional distribution might be of more interest. Therefore, we propose an approach to compute the exact alpha-beta divergence between any marginal or conditional distribution of two decomposable models. Doing so tractably is non-trivial as we need to decompose the divergence between these distributions and therefore, require a decomposition over the marginal and conditional distributions of these models. Consequently, we provide such a decomposition and also extend existing work to compute the marginal and conditional alpha-beta divergence between these decompositions. We then show how our method can be used to analyze distributional changes by first applying it to a benchmark image dataset. Finally, based on our framework, we propose a novel way to quantify the error in contemporary superconducting quantum computers. Code for all experiments is available at: https://lklee.dev/pub/2023-icdm/code.
Huynh, Viet; Say, Buser; Vogel, Peter; Cao, Lucy; Webb, Geoffrey I; Aleti, Aldeida
Rapid Identification of Protein Formulations with Bayesian Optimisation
2023 International Conference on Machine Learning and Applications (ICMLA), pp. 776-781, 2023.
@inproceedings{Huynh2023,
title = {Rapid Identification of Protein Formulations with Bayesian Optimisation},
author = {Viet Huynh and Buser Say and Peter Vogel and Lucy Cao and Geoffrey I Webb and Aldeida Aleti},
doi = {10.1109/ICMLA58977.2023.00113},
year = {2023},
date = {2023-01-01},
booktitle = {2023 International Conference on Machine Learning and Applications (ICMLA)},
pages = {776-781},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Lucas, Benjamin; Pelletier, Charlotte; Schmidt, Daniel; Webb, Geoffrey I; Petitjean, François
Machine Learning, vol. 112, pp. 1941-1973, 2023.
@article{lucas2021bayesian,
title = {A Bayesian-inspired, deep learning-based, semi-supervised domain adaptation technique for land cover mapping},
author = {Benjamin Lucas and Charlotte Pelletier and Daniel Schmidt and Geoffrey I Webb and Fran\c{c}ois Petitjean},
doi = {10.1007/s10994-020-05942-z},
year = {2023},
date = {2023-01-01},
journal = {Machine Learning},
volume = {112},
pages = {1941-1973},
publisher = {Springer US},
abstract = {Land cover maps are a vital input variable to many types of environmental research and management. While they can be produced automatically by machine learning techniques, these techniques require substantial training data to achieve high levels of accuracy, which are not always available. One technique researchers use when labelled training data are scarce is domain adaptation (DA) - where data from an alternate region, known as the source domain, are used to train a classifier and this model is adapted to map the study region, or target domain. The scenario we address in this paper is known as semi-supervised DA, where some labelled samples are available in the target domain. In this paper we present Sourcerer, a Bayesian-inspired, deep learning-based, semi-supervised DA technique for producing land cover maps from satellite image time series (SITS) data. The technique takes a convolutional neural network trained on a source domain and then trains further on the available target domain with a novel regularizer applied to the model weights. The regularizer adjusts the degree to which the model is modified to fit the target data, limiting the degree of change when the target data are few in number and increasing it as target data quantity increases. Our experiments on Sentinel-2 time series images compare Sourcerer with two state-of-the-art semi-supervised domain adaptation techniques and four baseline models. We show that on two different source-target domain pairings Sourcerer outperforms all other methods for any quantity of labelled target data available. In fact, the results on the more difficult target domain show that the starting accuracy of Sourcerer (when no labelled target data are available), 74.2%, is greater than the next-best state-of-the-art method trained on 20,000 labelled target instances.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Land cover maps are a vital input variable to many types of environmental research and management. While they can be produced automatically by machine learning techniques, these techniques require substantial training data to achieve high levels of accuracy, which are not always available. One technique researchers use when labelled training data are scarce is domain adaptation (DA) - where data from an alternate region, known as the source domain, are used to train a classifier and this model is adapted to map the study region, or target domain. The scenario we address in this paper is known as semi-supervised DA, where some labelled samples are available in the target domain. In this paper we present Sourcerer, a Bayesian-inspired, deep learning-based, semi-supervised DA technique for producing land cover maps from satellite image time series (SITS) data. The technique takes a convolutional neural network trained on a source domain and then trains further on the available target domain with a novel regularizer applied to the model weights. The regularizer adjusts the degree to which the model is modified to fit the target data, limiting the degree of change when the target data are few in number and increasing it as target data quantity increases. Our experiments on Sentinel-2 time series images compare Sourcerer with two state-of-the-art semi-supervised domain adaptation techniques and four baseline models. We show that on two different source-target domain pairings Sourcerer outperforms all other methods for any quantity of labelled target data available. In fact, the results on the more difficult target domain show that the starting accuracy of Sourcerer (when no labelled target data are available), 74.2%, is greater than the next-best state-of-the-art method trained on 20,000 labelled target instances.
Godahewa, Rakshitha; Bergmeir, Christoph; Webb, Geoffrey I.; Montero-Manso, Pablo
An accurate and fully-automated ensemble model for weekly time series forecasting
International Journal of Forecasting, vol. 39, iss. 2, pp. 641-658, 2023, ISSN: 0169-2070.
@article{GODAHEWA2022,
title = {An accurate and fully-automated ensemble model for weekly time series forecasting},
author = {Rakshitha Godahewa and Christoph Bergmeir and Geoffrey I. Webb and Pablo Montero-Manso},
doi = {10.1016/j.ijforecast.2022.01.008},
issn = {0169-2070},
year = {2023},
date = {2023-01-01},
journal = {International Journal of Forecasting},
volume = {39},
issue = {2},
pages = {641-658},
abstract = {Many businesses and industries require accurate forecasts for weekly time series nowadays. However, the forecasting literature does not currently provide easy-to-use, automatic, reproducible and accurate approaches dedicated to this task. We propose a forecasting method in this domain to fill this gap, leveraging state-of-the-art forecasting techniques, such as forecast combination, meta-learning, and global modelling. We consider different meta-learning architectures, algorithms, and base model pools. Based on all considered model variants, we propose to use a stacking approach with lasso regression which optimally combines the forecasts of four base models: a global Recurrent Neural Network (RNN) model, Theta, Trigonometric Box-Cox ARMA Trend Seasonal (TBATS), and Dynamic Harmonic Regression ARIMA (DHR-ARIMA), as it shows the overall best performance across seven experimental weekly datasets on four evaluation metrics. Our proposed method also consistently outperforms a set of benchmarks and state-of-the-art weekly forecasting models by a considerable margin with statistical significance. Our method can produce the most accurate forecasts, in terms of mean sMAPE, for the M4 weekly dataset among all benchmarks and all original competition participants.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Many businesses and industries require accurate forecasts for weekly time series nowadays. However, the forecasting literature does not currently provide easy-to-use, automatic, reproducible and accurate approaches dedicated to this task. We propose a forecasting method in this domain to fill this gap, leveraging state-of-the-art forecasting techniques, such as forecast combination, meta-learning, and global modelling. We consider different meta-learning architectures, algorithms, and base model pools. Based on all considered model variants, we propose to use a stacking approach with lasso regression which optimally combines the forecasts of four base models: a global Recurrent Neural Network (RNN) model, Theta, Trigonometric Box-Cox ARMA Trend Seasonal (TBATS), and Dynamic Harmonic Regression ARIMA (DHR-ARIMA), as it shows the overall best performance across seven experimental weekly datasets on four evaluation metrics. Our proposed method also consistently outperforms a set of benchmarks and state-of-the-art weekly forecasting models by a considerable margin with statistical significance. Our method can produce the most accurate forecasts, in terms of mean sMAPE, for the M4 weekly dataset among all benchmarks and all original competition participants.
Manapragada, Chaitanya; Salehi, Mahsa; Webb, Geoffrey I.
Extremely Fast Hoeffding Adaptive Tree
Proceedings of the 2022 IEEE International Conference on Data Mining (ICDM), pp. 319-328, IEEE, 2022.
@inproceedings{ManapragadaEtAl22,
title = {Extremely Fast Hoeffding Adaptive Tree},
author = {Chaitanya Manapragada and Mahsa Salehi and Geoffrey I. Webb},
doi = {10.1109/ICDM54844.2022.00042},
year = {2022},
date = {2022-01-01},
booktitle = {Proceedings of the 2022 IEEE International Conference on Data Mining (ICDM)},
pages = {319-328},
publisher = {IEEE},
abstract = {Many real-world data streams are non-stationary.
Subject to concept drift, the distributions change over time.
To retain accuracy in the face of such drift, online decision
tree learners must discard parts of the tree that are no longer
accurate and replace them by new subtrees that reflect the
new distribution. The longstanding state-of-the-art online
decision tree learner for non-stationary streams is Hoeffding
Adaptive Tree (HAT), which adds a drift detection and response
mechanism to the classic Very Fast Decision Tree (VFDT) online
decision tree learner. However, for stationary distributions,
VFDT has been superseded by Extremely Fast Decision Tree
(EFDT), which uses a statistically more efficient learning
mechanism than VFDT. This learning mechanism needs to be
coupled with a compensatory revision mechanism that can
compensate for circumstances where the learning mechanism is
too eager.
The current work develops a strategy to combine the best
of both these state-of-the-art approaches, exploiting both the
statistically efficient learning mechanism from EFDT and the
highly effective drift detection and response mechanism of HAT.
To do so requires decoupling of the EFDT splitting and revision
mechanisms, as the latter incorrectly triggers the HAT drift
detection mechanism. The resulting learner, Extremely Fast
Hoeffding Adaptive Tree, responds to drift more rapidly and
effectively than either HAT or EFDT, and attains a statistically
significant advantage in accuracy even on stationary streams.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Many real-world data streams are non-stationary.
Subject to concept drift, the distributions change over time.
To retain accuracy in the face of such drift, online decision
tree learners must discard parts of the tree that are no longer
accurate and replace them by new subtrees that reflect the
new distribution. The longstanding state-of-the-art online
decision tree learner for non-stationary streams is Hoeffding
Adaptive Tree (HAT), which adds a drift detection and response
mechanism to the classic Very Fast Decision Tree (VFDT) online
decision tree learner. However, for stationary distributions,
VFDT has been superseded by Extremely Fast Decision Tree
(EFDT), which uses a statistically more efficient learning
mechanism than VFDT. This learning mechanism needs to be
coupled with a compensatory revision mechanism that can
compensate for circumstances where the learning mechanism is
too eager.
The current work develops a strategy to combine the best
of both these state-of-the-art approaches, exploiting both the
statistically efficient learning mechanism from EFDT and the
highly effective drift detection and response mechanism of HAT.
To do so requires decoupling of the EFDT splitting and revision
mechanisms, as the latter incorrectly triggers the HAT drift
detection mechanism. The resulting learner, Extremely Fast
Hoeffding Adaptive Tree, responds to drift more rapidly and
effectively than either HAT or EFDT, and attains a statistically
significant advantage in accuracy even on stationary streams.
Subject to concept drift, the distributions change over time.
To retain accuracy in the face of such drift, online decision
tree learners must discard parts of the tree that are no longer
accurate and replace them by new subtrees that reflect the
new distribution. The longstanding state-of-the-art online
decision tree learner for non-stationary streams is Hoeffding
Adaptive Tree (HAT), which adds a drift detection and response
mechanism to the classic Very Fast Decision Tree (VFDT) online
decision tree learner. However, for stationary distributions,
VFDT has been superseded by Extremely Fast Decision Tree
(EFDT), which uses a statistically more efficient learning
mechanism than VFDT. This learning mechanism needs to be
coupled with a compensatory revision mechanism that can
compensate for circumstances where the learning mechanism is
too eager.
The current work develops a strategy to combine the best
of both these state-of-the-art approaches, exploiting both the
statistically efficient learning mechanism from EFDT and the
highly effective drift detection and response mechanism of HAT.
To do so requires decoupling of the EFDT splitting and revision
mechanisms, as the latter incorrectly triggers the HAT drift
detection mechanism. The resulting learner, Extremely Fast
Hoeffding Adaptive Tree, responds to drift more rapidly and
effectively than either HAT or EFDT, and attains a statistically
significant advantage in accuracy even on stationary streams.
Livori, Adam C; Lukose, Dickson; Bell, J Simon; Webb, Geoffrey I; Ilomaki, Jenni
Did Australia's COVID-19 restrictions impact statin incidence, prevalence or adherence?
Current Problems in Cardiology, pp. 101576, 2022, ISSN: 0146-2806.
@article{Livori2022,
title = {Did Australia's COVID-19 restrictions impact statin incidence, prevalence or adherence?},
author = {Adam C Livori and Dickson Lukose and J Simon Bell and Geoffrey I Webb and Jenni Ilomaki},
doi = {10.1016/j.cpcardiol.2022.101576},
issn = {0146-2806},
year = {2022},
date = {2022-01-01},
journal = {Current Problems in Cardiology},
pages = {101576},
abstract = {Objective
COVID-19 restrictions may have an unintended consequence of limiting access to cardiovascular care. Australia implemented adaptive interventions (e.g. telehealth consultations, digital image prescriptions, continued dispensing, medication delivery) to maintain medication access. This study investigated whether COVID-19 restrictions in different jurisdictions coincided with changes in statin incidence, prevalence and adherence.
Methods Analysis of a 10% random sample of national medication claims data from January 2018 to December 2020 was conducted across three Australian jurisdictions. Weekly incidence and prevalence were estimated by dividing the number statin initiations and any statin dispensing by the Australian population aged 18-99 years. Statin adherence was analysed across the jurisdictions and years, with adherence categorised as \<40%, 40-79% and \>=80% based on dispensings per calendar year.
Results Overall, 309,123, 315,703 and 324,906 people were dispensed and 39029, 39816, and 44979 initiated statins in 2018, 2019 and 2020 respectively. Two waves of COVID-19 restrictions in 2020 coincided with no meaningful change in statin incidence or prevalence per week when compared to 2018 and 2019. Incidence increased 0.3% from 23.7 to 26.2 per 1000 people across jurisdictions in 2020 compared to 2019. Prevalence increased 0.14% from 158.5 to 159.9 per 1000 people across jurisdictions in 2020 compared to 2019. The proportion of adults with \>=80% adherence increased by 3.3% in Victoria, 1.4% in NSW and 1.8% in other states and territories between 2019 and 2020.
Conclusions
COVID-19 restrictions did not coincide with meaningful changes in the incidence, prevalence or adherence to statins suggesting adaptive interventions succeeded in maintaining access to cardiovascular medications.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Objective
COVID-19 restrictions may have an unintended consequence of limiting access to cardiovascular care. Australia implemented adaptive interventions (e.g. telehealth consultations, digital image prescriptions, continued dispensing, medication delivery) to maintain medication access. This study investigated whether COVID-19 restrictions in different jurisdictions coincided with changes in statin incidence, prevalence and adherence.
Methods Analysis of a 10% random sample of national medication claims data from January 2018 to December 2020 was conducted across three Australian jurisdictions. Weekly incidence and prevalence were estimated by dividing the number statin initiations and any statin dispensing by the Australian population aged 18-99 years. Statin adherence was analysed across the jurisdictions and years, with adherence categorised as <40%, 40-79% and >=80% based on dispensings per calendar year.
Results Overall, 309,123, 315,703 and 324,906 people were dispensed and 39029, 39816, and 44979 initiated statins in 2018, 2019 and 2020 respectively. Two waves of COVID-19 restrictions in 2020 coincided with no meaningful change in statin incidence or prevalence per week when compared to 2018 and 2019. Incidence increased 0.3% from 23.7 to 26.2 per 1000 people across jurisdictions in 2020 compared to 2019. Prevalence increased 0.14% from 158.5 to 159.9 per 1000 people across jurisdictions in 2020 compared to 2019. The proportion of adults with >=80% adherence increased by 3.3% in Victoria, 1.4% in NSW and 1.8% in other states and territories between 2019 and 2020.
Conclusions
COVID-19 restrictions did not coincide with meaningful changes in the incidence, prevalence or adherence to statins suggesting adaptive interventions succeeded in maintaining access to cardiovascular medications.
COVID-19 restrictions may have an unintended consequence of limiting access to cardiovascular care. Australia implemented adaptive interventions (e.g. telehealth consultations, digital image prescriptions, continued dispensing, medication delivery) to maintain medication access. This study investigated whether COVID-19 restrictions in different jurisdictions coincided with changes in statin incidence, prevalence and adherence.
Methods Analysis of a 10% random sample of national medication claims data from January 2018 to December 2020 was conducted across three Australian jurisdictions. Weekly incidence and prevalence were estimated by dividing the number statin initiations and any statin dispensing by the Australian population aged 18-99 years. Statin adherence was analysed across the jurisdictions and years, with adherence categorised as <40%, 40-79% and >=80% based on dispensings per calendar year.
Results Overall, 309,123, 315,703 and 324,906 people were dispensed and 39029, 39816, and 44979 initiated statins in 2018, 2019 and 2020 respectively. Two waves of COVID-19 restrictions in 2020 coincided with no meaningful change in statin incidence or prevalence per week when compared to 2018 and 2019. Incidence increased 0.3% from 23.7 to 26.2 per 1000 people across jurisdictions in 2020 compared to 2019. Prevalence increased 0.14% from 158.5 to 159.9 per 1000 people across jurisdictions in 2020 compared to 2019. The proportion of adults with >=80% adherence increased by 3.3% in Victoria, 1.4% in NSW and 1.8% in other states and territories between 2019 and 2020.
Conclusions
COVID-19 restrictions did not coincide with meaningful changes in the incidence, prevalence or adherence to statins suggesting adaptive interventions succeeded in maintaining access to cardiovascular medications.
Abourayya, Amr; Kamp, Michael; Ayday, Erman; Kleesiek, Jens; Rao, Kanishka; Webb, Geoffrey I.; Rao, Bharat
AIMHI: Protecting Sensitive Data through Federated Co-Training
Workshop on Federated Learning: Recent Advances and New Challenges (in Conjunction with NeurIPS 2022), 2022.
@inproceedings{Abourayya2022,
title = {AIMHI: Protecting Sensitive Data through Federated Co-Training},
author = {Amr Abourayya and Michael Kamp and Erman Ayday and Jens Kleesiek and Kanishka Rao and Geoffrey I. Webb and Bharat Rao},
url = {https://openreview.net/forum?id=6uJo_mWAmHZ},
year = {2022},
date = {2022-01-01},
booktitle = {Workshop on Federated Learning: Recent Advances and New Challenges (in Conjunction with NeurIPS 2022)},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Li, Fuyi; Dong, Shuangyu; Leier, Andre; Han, Meiya; Guo, Xudong; Xu, Jing; Wang, Xiaoyu; Pan, Shirui; Jia, Cangzhi; Zhang, Yang; Webb, Geoffrey I; Coin, Lachlan J M; Li, Chen; Song, Jiangning
Positive-unlabeled learning in bioinformatics and computational biology: a brief review
Briefings in Bioinformatics, vol. 23, no. 1, 2022, ISSN: 1477-4054.
@article{10.1093/bib/bbab461,
title = {Positive-unlabeled learning in bioinformatics and computational biology: a brief review},
author = {Fuyi Li and Shuangyu Dong and Andre Leier and Meiya Han and Xudong Guo and Jing Xu and Xiaoyu Wang and Shirui Pan and Cangzhi Jia and Yang Zhang and Geoffrey I Webb and Lachlan J M Coin and Chen Li and Jiangning Song},
doi = {10.1093/bib/bbab461},
issn = {1477-4054},
year = {2022},
date = {2022-01-01},
journal = {Briefings in Bioinformatics},
volume = {23},
number = {1},
abstract = {Conventional supervised binary classification algorithms have been widely applied to address significant research questions using biological and biomedical data. This classification scheme requires two fully labeled classes of data (e.g. positive and negative samples) to train a classification model. However, in many bioinformatics applications, labeling data is laborious, and the negative samples might be potentially mislabeled due to the limited sensitivity of the experimental equipment. The positive unlabeled (PU) learning scheme was therefore proposed to enable the classifier to learn directly from limited positive samples and a large number of unlabeled samples (i.e. a mixture of positive or negative samples). To date, several PU learning algorithms have been developed to address various biological questions, such as sequence identification, functional site characterization and interaction prediction. In this paper, we revisit a collection of 29 state-of-the-art PU learning bioinformatic applications to address various biological questions. Various important aspects are extensively discussed, including PU learning methodology, biological application, classifier design and evaluation strategy. We also comment on the existing issues of PU learning and offer our perspectives for the future development of PU learning applications. We anticipate that our work serves as an instrumental guideline for a better understanding of the PU learning framework in bioinformatics and further developing next-generation PU learning frameworks for critical biological applications.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Conventional supervised binary classification algorithms have been widely applied to address significant research questions using biological and biomedical data. This classification scheme requires two fully labeled classes of data (e.g. positive and negative samples) to train a classification model. However, in many bioinformatics applications, labeling data is laborious, and the negative samples might be potentially mislabeled due to the limited sensitivity of the experimental equipment. The positive unlabeled (PU) learning scheme was therefore proposed to enable the classifier to learn directly from limited positive samples and a large number of unlabeled samples (i.e. a mixture of positive or negative samples). To date, several PU learning algorithms have been developed to address various biological questions, such as sequence identification, functional site characterization and interaction prediction. In this paper, we revisit a collection of 29 state-of-the-art PU learning bioinformatic applications to address various biological questions. Various important aspects are extensively discussed, including PU learning methodology, biological application, classifier design and evaluation strategy. We also comment on the existing issues of PU learning and offer our perspectives for the future development of PU learning applications. We anticipate that our work serves as an instrumental guideline for a better understanding of the PU learning framework in bioinformatics and further developing next-generation PU learning frameworks for critical biological applications.
Manapragada, Chaitanya; Gomes, Heitor M.; Salehi, Mahsa; Bifet, Albert; Webb, Geoffrey I.
An eager splitting strategy for online decision trees in ensembles
Data Mining and Knowledge Discovery, vol. 36, pp. 566-619, 2022.
@article{Manapragada_2022,
title = {An eager splitting strategy for online decision trees in ensembles},
author = {Chaitanya Manapragada and Heitor M. Gomes and Mahsa Salehi and Albert Bifet and Geoffrey I. Webb},
url = {https://rdcu.be/c1y4Z},
doi = {10.1007/s10618-021-00816-x},
year = {2022},
date = {2022-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {36},
pages = {566-619},
publisher = {Springer Science and Business Media LLC},
abstract = {Decision tree ensembles are widely used in practice. In this work, we study in ensemble settings the effectiveness of replacing the split strategy for the state-of-the-art online tree learner, Hoeffding Tree, with a rigorous but more eager splitting strategy that we had previously published as Hoeffding AnyTime Tree. Hoeffding AnyTime Tree (HATT), uses the Hoeffding Test to determine whether the current best candidate split is superior to the current split, with the possibility of revision, while Hoeffding Tree aims to determine whether the top candidate is better than the second best and if a test is selected, fixes it for all posterity. HATT converges to the ideal batch tree while Hoeffding Tree does not. We find that HATT is an efficacious base learner for online bagging and online boosting ensembles. On UCI and synthetic streams, HATT as a base learner outperforms HT at a 0.05 significance level for the majority of tested ensembles on what we believe is the largest and most comprehensive set of testbenches in the online learning literature. Our results indicate that HATT is a superior alternative to Hoeffding Tree in a large number of ensemble settings.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Decision tree ensembles are widely used in practice. In this work, we study in ensemble settings the effectiveness of replacing the split strategy for the state-of-the-art online tree learner, Hoeffding Tree, with a rigorous but more eager splitting strategy that we had previously published as Hoeffding AnyTime Tree. Hoeffding AnyTime Tree (HATT), uses the Hoeffding Test to determine whether the current best candidate split is superior to the current split, with the possibility of revision, while Hoeffding Tree aims to determine whether the top candidate is better than the second best and if a test is selected, fixes it for all posterity. HATT converges to the ideal batch tree while Hoeffding Tree does not. We find that HATT is an efficacious base learner for online bagging and online boosting ensembles. On UCI and synthetic streams, HATT as a base learner outperforms HT at a 0.05 significance level for the majority of tested ensembles on what we believe is the largest and most comprehensive set of testbenches in the online learning literature. Our results indicate that HATT is a superior alternative to Hoeffding Tree in a large number of ensemble settings.
Zhang, Meng; Jia, Cangzhi; Li, Fuyi; Li, Chen; Zhu, Yan; Akutsu, Tatsuya; Webb, Geoffrey I; Zou, Quan; Coin, Lachlan J M; Song, Jiangning
Critical assessment of computational tools for prokaryotic and eukaryotic promoter prediction
Briefings in Bioinformatics, vol. 23, iss. 2, 2022, ISSN: 1477-4054.
@article{10.1093/bib/bbab551,
title = {Critical assessment of computational tools for prokaryotic and eukaryotic promoter prediction},
author = {Meng Zhang and Cangzhi Jia and Fuyi Li and Chen Li and Yan Zhu and Tatsuya Akutsu and Geoffrey I Webb and Quan Zou and Lachlan J M Coin and Jiangning Song},
doi = {10.1093/bib/bbab551},
issn = {1477-4054},
year = {2022},
date = {2022-01-01},
journal = {Briefings in Bioinformatics},
volume = {23},
issue = {2},
abstract = {Promoters are crucial regulatory DNA regions for gene transcriptional activation. Rapid advances in next-generation sequencing technologies have accelerated the accumulation of genome sequences, providing increased training data to inform computational approaches for both prokaryotic and eukaryotic promoter prediction. However, it remains a significant challenge to accurately identify species-specific promoter sequences using computational approaches. To advance computational support for promoter prediction, in this study, we curated 58 comprehensive, up-to-date, benchmark datasets for 7 different species (i.e. Escherichia coli, Bacillus subtilis, Homo sapiens, Mus musculus, Arabidopsis thaliana, Zea mays and Drosophila melanogaster) to assist the research community to assess the relative functionality of alternative approaches and support future research on both prokaryotic and eukaryotic promoters. We revisited 106 predictors published since 2000 for promoter identification (40 for prokaryotic promoter, 61 for eukaryotic promoter, and 5 for both). We systematically evaluated their training datasets, computational methodologies, calculated features, performance and software usability. On the basis of these benchmark datasets, we benchmarked 19 predictors with functioning webservers/local tools and assessed their prediction performance. We found that deep learning and traditional machine learning-based approaches generally outperformed scoring function-based approaches. Taken together, the curated benchmark dataset repository and the benchmarking analysis in this study serve to inform the design and implementation of computational approaches for promoter prediction and facilitate more rigorous comparison of new techniques in the future.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Promoters are crucial regulatory DNA regions for gene transcriptional activation. Rapid advances in next-generation sequencing technologies have accelerated the accumulation of genome sequences, providing increased training data to inform computational approaches for both prokaryotic and eukaryotic promoter prediction. However, it remains a significant challenge to accurately identify species-specific promoter sequences using computational approaches. To advance computational support for promoter prediction, in this study, we curated 58 comprehensive, up-to-date, benchmark datasets for 7 different species (i.e. Escherichia coli, Bacillus subtilis, Homo sapiens, Mus musculus, Arabidopsis thaliana, Zea mays and Drosophila melanogaster) to assist the research community to assess the relative functionality of alternative approaches and support future research on both prokaryotic and eukaryotic promoters. We revisited 106 predictors published since 2000 for promoter identification (40 for prokaryotic promoter, 61 for eukaryotic promoter, and 5 for both). We systematically evaluated their training datasets, computational methodologies, calculated features, performance and software usability. On the basis of these benchmark datasets, we benchmarked 19 predictors with functioning webservers/local tools and assessed their prediction performance. We found that deep learning and traditional machine learning-based approaches generally outperformed scoring function-based approaches. Taken together, the curated benchmark dataset repository and the benchmarking analysis in this study serve to inform the design and implementation of computational approaches for promoter prediction and facilitate more rigorous comparison of new techniques in the future.
Wang, Xiaoyu; Li, Fuyi; Xu, Jing; Rong, Jia; Webb, Geoffrey I; Ge, Zongyuan; Li, Jian; Song, Jiangning
Briefings in Bioinformatics, vol. 23, no. 2, 2022, ISSN: 1477-4054.
@article{10.1093/bib/bbac031,
title = {ASPIRER: a new computational approach for identifying non-classical secreted proteins based on deep learning},
author = {Xiaoyu Wang and Fuyi Li and Jing Xu and Jia Rong and Geoffrey I Webb and Zongyuan Ge and Jian Li and Jiangning Song},
doi = {10.1093/bib/bbac031},
issn = {1477-4054},
year = {2022},
date = {2022-01-01},
journal = {Briefings in Bioinformatics},
volume = {23},
number = {2},
abstract = {Protein secretion has a pivotal role in many biological processes and is particularly important for intercellular communication, from the cytoplasm to the host or external environment. Gram-positive bacteria can secrete proteins through multiple secretion pathways. The non-classical secretion pathway has recently received increasing attention among these secretion pathways, but its exact mechanism remains unclear. Non-classical secreted proteins (NCSPs) are a class of secreted proteins lacking signal peptides and motifs. Several NCSP predictors have been proposed to identify NCSPs and most of them employed the whole amino acid sequence of NCSPs to construct the model. However, the sequence length of different proteins varies greatly. In addition, not all regions of the protein are equally important and some local regions are not relevant to the secretion. The functional regions of the protein, particularly in the N- and C-terminal regions, contain important determinants for secretion. In this study, we propose a new hybrid deep learning-based framework, referred to as ASPIRER, which improves the prediction of NCSPs from amino acid sequences. More specifically, it combines a whole sequence-based XGBoost model and an N-terminal sequence-based convolutional neural network model; 5-fold cross-validation and independent tests demonstrate that ASPIRER achieves superior performance than existing state-of-the-art approaches. The source code and curated datasets of ASPIRER are publicly available at https://github.com/yanwu20/ASPIRER/. ASPIRER is anticipated to be a useful tool for improved prediction of novel putative NCSPs from sequences information and prioritization of candidate proteins for follow-up experimental validation.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Protein secretion has a pivotal role in many biological processes and is particularly important for intercellular communication, from the cytoplasm to the host or external environment. Gram-positive bacteria can secrete proteins through multiple secretion pathways. The non-classical secretion pathway has recently received increasing attention among these secretion pathways, but its exact mechanism remains unclear. Non-classical secreted proteins (NCSPs) are a class of secreted proteins lacking signal peptides and motifs. Several NCSP predictors have been proposed to identify NCSPs and most of them employed the whole amino acid sequence of NCSPs to construct the model. However, the sequence length of different proteins varies greatly. In addition, not all regions of the protein are equally important and some local regions are not relevant to the secretion. The functional regions of the protein, particularly in the N- and C-terminal regions, contain important determinants for secretion. In this study, we propose a new hybrid deep learning-based framework, referred to as ASPIRER, which improves the prediction of NCSPs from amino acid sequences. More specifically, it combines a whole sequence-based XGBoost model and an N-terminal sequence-based convolutional neural network model; 5-fold cross-validation and independent tests demonstrate that ASPIRER achieves superior performance than existing state-of-the-art approaches. The source code and curated datasets of ASPIRER are publicly available at https://github.com/yanwu20/ASPIRER/. ASPIRER is anticipated to be a useful tool for improved prediction of novel putative NCSPs from sequences information and prioritization of candidate proteins for follow-up experimental validation.
Pialla, Gautier; Fawaz, Hassan Ismail; Devanne, Maxime; Weber, Jonathan; Idoumghar, Lhassane; Muller, Pierre-Alain; Bergmeir, Christoph; Schmidt, Daniel; Webb, Geoffrey I.; Forestier, Germain
Smooth Perturbations for Time Series Adversarial Attacks
Gama, João; Li, Tianrui; Yu, Yang; Chen, Enhong; Zheng, Yu; Teng, Fei (Ed.): Proceedings of the 2022 Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. 485-496, Springer International Publishing, Cham, 2022, ISBN: 978-3-031-05933-9.
@inproceedings{10.1007/978-3-031-05933-9_38,
title = {Smooth Perturbations for Time Series Adversarial Attacks},
author = {Gautier Pialla and Hassan Ismail Fawaz and Maxime Devanne and Jonathan Weber and Lhassane Idoumghar and Pierre-Alain Muller and Christoph Bergmeir and Daniel Schmidt and Geoffrey I. Webb and Germain Forestier},
editor = {Jo\~{a}o Gama and Tianrui Li and Yang Yu and Enhong Chen and Yu Zheng and Fei Teng},
doi = {10.1007/978-3-031-05933-9_38},
isbn = {978-3-031-05933-9},
year = {2022},
date = {2022-01-01},
booktitle = {Proceedings of the 2022 Pacific-Asia Conference on Knowledge Discovery and Data Mining},
pages = {485-496},
publisher = {Springer International Publishing},
address = {Cham},
abstract = {Adversarial attacks represent a threat to every deep neural network. They are particularly effective if they can perturb a given model while remaining undetectable. They have been initially introduced for image classifiers, and are well studied for this task. For time series, few attacks have yet been proposed. Most that have are adaptations of attacks previously proposed for image classifiers. Although these attacks are effective, they generate perturbations containing clearly discernible patterns such as sawtooth and spikes. Adversarial patterns are not perceptible on images, but the attacks proposed to date are readily perceptible in the case of time series. In order to generate stealthier adversarial attacks for time series, we propose a new attack that produces smoother perturbations. We find that smooth perturbations are harder to detect by the naked eye. We also show how adversarial training can improve model robustness against this attack, thus making models less vulnerable.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Adversarial attacks represent a threat to every deep neural network. They are particularly effective if they can perturb a given model while remaining undetectable. They have been initially introduced for image classifiers, and are well studied for this task. For time series, few attacks have yet been proposed. Most that have are adaptations of attacks previously proposed for image classifiers. Although these attacks are effective, they generate perturbations containing clearly discernible patterns such as sawtooth and spikes. Adversarial patterns are not perceptible on images, but the attacks proposed to date are readily perceptible in the case of time series. In order to generate stealthier adversarial attacks for time series, we propose a new attack that produces smoother perturbations. We find that smooth perturbations are harder to detect by the naked eye. We also show how adversarial training can improve model robustness against this attack, thus making models less vulnerable.
Wang, Yanan; Wang, Yu Guang; Hu, Changyuan; Li, Ming; Fan, Yanan; Otter, Nina; Sam, Ikuan; Gou, Hongquan; Hu, Yiqun; Kwok, Terry; Zalcberg, John; Boussioutas, Alex; Daly, Roger J.; Montfar, Guido; Li, Pietro; Xu, Dakang; Webb, Geoffrey I.; Song, Jiangning
Cell graph neural networks enable the precise prediction of patient survival in gastric cancer
npj Precision Oncology, vol. 6, no. 1, 2022, ISSN: 2397-768X.
@article{Wang2022,
title = {Cell graph neural networks enable the precise prediction of patient survival in gastric cancer},
author = {Yanan Wang and Yu Guang Wang and Changyuan Hu and Ming Li and Yanan Fan and Nina Otter and Ikuan Sam and Hongquan Gou and Yiqun Hu and Terry Kwok and John Zalcberg and Alex Boussioutas and Roger J. Daly and Guido Montfar and Pietro Li and Dakang Xu and Geoffrey I. Webb and Jiangning Song},
url = {https://rdcu.be/cQeFD},
doi = {10.1038/s41698-022-00285-5},
issn = {2397-768X},
year = {2022},
date = {2022-01-01},
journal = {npj Precision Oncology},
volume = {6},
number = {1},
abstract = {Gastric cancer is one of the deadliest cancers worldwide. An accurate prognosis is essential for effective clinical assessment and treatment. Spatial patterns in the tumor microenvironment (TME) are conceptually indicative of the staging and progression of gastric cancer patients. Using spatial patterns of the TME by integrating and transforming the multiplexed immunohistochemistry (mIHC) images as Cell-Graphs, we propose a graph neural network-based approach, termed Cell-Graph Signature or CGSignature, powered by artificial intelligence, for the digital staging of TME and precise prediction of patient survival in gastric cancer. In this study, patient survival prediction is formulated as either a binary (short-term and long-term) or ternary (short-term, medium-term, and long-term) classification task. Extensive benchmarking experiments demonstrate that the CGSignature achieves outstanding model performance, with Area Under the Receiver Operating Characteristic curve of 0.960+/-0.01, and 0.771+/-0.024 to 0.904+/-0.012 for the binary- and ternary-classification, respectively. Moreover, Kaplan-Meier survival analysis indicates that the 'digital grade' cancer staging produced by CGSignature provides a remarkable capability in discriminating both binary and ternary classes with statistical significance (P value \<=0.0001), significantly outperforming the AJCC 8th edition Tumor Node Metastasis staging system. Using Cell-Graphs extracted from mIHC images, CGSignature improves the assessment of the link between the TME spatial patterns and patient prognosis. Our study suggests the feasibility and benefits of such an artificial intelligence-powered digital staging system in diagnostic pathology and precision oncology.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Gastric cancer is one of the deadliest cancers worldwide. An accurate prognosis is essential for effective clinical assessment and treatment. Spatial patterns in the tumor microenvironment (TME) are conceptually indicative of the staging and progression of gastric cancer patients. Using spatial patterns of the TME by integrating and transforming the multiplexed immunohistochemistry (mIHC) images as Cell-Graphs, we propose a graph neural network-based approach, termed Cell-Graph Signature or CGSignature, powered by artificial intelligence, for the digital staging of TME and precise prediction of patient survival in gastric cancer. In this study, patient survival prediction is formulated as either a binary (short-term and long-term) or ternary (short-term, medium-term, and long-term) classification task. Extensive benchmarking experiments demonstrate that the CGSignature achieves outstanding model performance, with Area Under the Receiver Operating Characteristic curve of 0.960+/-0.01, and 0.771+/-0.024 to 0.904+/-0.012 for the binary- and ternary-classification, respectively. Moreover, Kaplan-Meier survival analysis indicates that the 'digital grade' cancer staging produced by CGSignature provides a remarkable capability in discriminating both binary and ternary classes with statistical significance (P value <=0.0001), significantly outperforming the AJCC 8th edition Tumor Node Metastasis staging system. Using Cell-Graphs extracted from mIHC images, CGSignature improves the assessment of the link between the TME spatial patterns and patient prognosis. Our study suggests the feasibility and benefits of such an artificial intelligence-powered digital staging system in diagnostic pathology and precision oncology.
Tan, Chang Wei; Dempster, Angus; Bergmeir, Christoph; Webb, Geoffrey I.
Data Mining and Knowledge Discovery, vol. 36, pp. 1623-1646, 2022, ISSN: 1573-756X.
@article{Tan2022,
title = {MultiRocket: multiple pooling operators and transformations for fast and effective time series classification},
author = {Chang Wei Tan and Angus Dempster and Christoph Bergmeir and Geoffrey I. Webb},
doi = {10.1007/s10618-022-00844-1},
issn = {1573-756X},
year = {2022},
date = {2022-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {36},
pages = {1623-1646},
abstract = {We propose MultiRocket, a fast time series classification (TSC) algorithm that achieves state-of-the-art accuracy with a tiny fraction of the time and without the complex ensembling structure of many state-of-the-art methods. MultiRocket improves on MiniRocket, one of the fastest TSC algorithms to date, by adding multiple pooling operators and transformations to improve the diversity of the features generated. In addition to processing the raw input series, MultiRocket also applies first order differences to transform the original series. Convolutions are applied to both representations, and four pooling operators are applied to the convolution outputs. When benchmarked using the University of California Riverside TSC benchmark datasets, MultiRocket is significantly more accurate than MiniRocket, and competitive with the best ranked current method in terms of accuracy, HIVE-COTE 2.0, while being orders of magnitude faster.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We propose MultiRocket, a fast time series classification (TSC) algorithm that achieves state-of-the-art accuracy with a tiny fraction of the time and without the complex ensembling structure of many state-of-the-art methods. MultiRocket improves on MiniRocket, one of the fastest TSC algorithms to date, by adding multiple pooling operators and transformations to improve the diversity of the features generated. In addition to processing the raw input series, MultiRocket also applies first order differences to transform the original series. Convolutions are applied to both representations, and four pooling operators are applied to the convolution outputs. When benchmarked using the University of California Riverside TSC benchmark datasets, MultiRocket is significantly more accurate than MiniRocket, and competitive with the best ranked current method in terms of accuracy, HIVE-COTE 2.0, while being orders of magnitude faster.
Wang, Yanan; Hu, Changyuan; Kwok, Terry; Bain, Christopher A; Xue, Xiangyang; Gasser, Robin B; Webb, Geoffrey I; Boussioutas, Alex; Shen, Xian; Daly, Roger J; Song, Jiangning
Bioinformatics, vol. 38, no. 17, pp. 4206-4213, 2022.
@article{Wang_2022,
title = {DEMoS: A Deep Learning-based Ensemble Approach for Predicting the Molecular Subtypes of Gastric Adenocarcinomas from Histopathological Images},
author = {Yanan Wang and Changyuan Hu and Terry Kwok and Christopher A Bain and Xiangyang Xue and Robin B Gasser and Geoffrey I Webb and Alex Boussioutas and Xian Shen and Roger J Daly and Jiangning Song},
editor = {Hanchuan Peng},
doi = {10.1093/bioinformatics/btac456},
year = {2022},
date = {2022-01-01},
journal = {Bioinformatics},
volume = {38},
number = {17},
pages = {4206-4213},
publisher = {Oxford University Press (OUP)},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Chen, Zhen; Liu, Xuhan; Li, Fuyi; Li, Chen; Marquez-Lago, Tatiana; Leier, André; Webb, Geoffrey I.; Xu, Dakang; Akutsu, Tatsuya; Song, Jiangning
Systematic Characterization of Lysine Post-translational Modification Sites Using MUscADEL
KC, Dukka B. (Ed.): Computational Methods for Predicting Post-Translational Modification Sites, pp. 205-219, Springer US, New York, NY, 2022, ISBN: 978-1-0716-2317-6.
@inbook{Chen2022,
title = {Systematic Characterization of Lysine Post-translational Modification Sites Using MUscADEL},
author = {Zhen Chen and Xuhan Liu and Fuyi Li and Chen Li and Tatiana Marquez-Lago and Andr\'{e} Leier and Geoffrey I. Webb and Dakang Xu and Tatsuya Akutsu and Jiangning Song},
editor = {Dukka B. KC},
doi = {10.1007/978-1-0716-2317-6_11},
isbn = {978-1-0716-2317-6},
year = {2022},
date = {2022-01-01},
booktitle = {Computational Methods for Predicting Post-Translational Modification Sites},
pages = {205-219},
publisher = {Springer US},
address = {New York, NY},
abstract = {Among various types of protein post-translational modifications (PTMs), lysineLysinesPTMs play an important role in regulating a wide range of functions and biological processes. Due to the generation and accumulation of enormous amount of protein sequence data by ongoing whole-genome sequencing projects, systematic identification of different types of lysineLysinesPTMPost-translational modification (PTM)substrates and their specific PTMPost-translational modification (PTM)sites in the entire proteome is increasingly important and has therefore received much attention. Accordingly, a variety of computational methods for lysineLysinesPTMPost-translational modification (PTM)identification have been developed based on the combination of various handcrafted sequence features and machine-learning techniques. In this chapter, we first briefly review existing computational methods for lysineLysinesPTMPost-translational modification (PTM)identification and then introduce a recently developed deep learning-based method, termed MUscADELMUscADEL (Multiple Scalable Accurate Deep Learner for lysineLysinesPTMs). Specifically, MUscADELMUscADEL employs bidirectional long short-term memoryLong short-term memory (LSTM) (BiLSTM) recurrent neural networks and is capable of predicting eight major types of lysineLysinesPTMs in both the human and mouse proteomes. The web server of MUscADELMUscADEL is publicly available at http://muscadel.erc.monash.edu/for the research community to use.},
keywords = {},
pubstate = {published},
tppubtype = {inbook}
}
Among various types of protein post-translational modifications (PTMs), lysineLysinesPTMs play an important role in regulating a wide range of functions and biological processes. Due to the generation and accumulation of enormous amount of protein sequence data by ongoing whole-genome sequencing projects, systematic identification of different types of lysineLysinesPTMPost-translational modification (PTM)substrates and their specific PTMPost-translational modification (PTM)sites in the entire proteome is increasingly important and has therefore received much attention. Accordingly, a variety of computational methods for lysineLysinesPTMPost-translational modification (PTM)identification have been developed based on the combination of various handcrafted sequence features and machine-learning techniques. In this chapter, we first briefly review existing computational methods for lysineLysinesPTMPost-translational modification (PTM)identification and then introduce a recently developed deep learning-based method, termed MUscADELMUscADEL (Multiple Scalable Accurate Deep Learner for lysineLysinesPTMs). Specifically, MUscADELMUscADEL employs bidirectional long short-term memoryLong short-term memory (LSTM) (BiLSTM) recurrent neural networks and is capable of predicting eight major types of lysineLysinesPTMs in both the human and mouse proteomes. The web server of MUscADELMUscADEL is publicly available at http://muscadel.erc.monash.edu/for the research community to use.
Miller, Lynn; Zhu, Liujun; Yebra, Marta; Rudiger, Christoph; Webb, Geoffrey I.
Multi-modal temporal CNNs for live fuel moisture content estimation
Environmental Modelling & Software, pp. 105467, 2022, ISSN: 1364-8152.
@article{Miller2022,
title = {Multi-modal temporal CNNs for live fuel moisture content estimation},
author = {Lynn Miller and Liujun Zhu and Marta Yebra and Christoph Rudiger and Geoffrey I. Webb},
doi = {10.1016/j.envsoft.2022.105467},
issn = {1364-8152},
year = {2022},
date = {2022-01-01},
journal = {Environmental Modelling \& Software},
pages = {105467},
abstract = {Live fuel moisture content (LFMC) is an important environmental indicator used to measure vegetation conditions and monitor for high fire risk conditions. However, LFMC is challenging to measure on a wide scale, thus reliable models for estimating LFMC are needed. Therefore, this paper proposes a new deep learning architecture for LFMC estimation. The architecture comprises an ensemble of temporal convolutional neural networks that learn from year-long time series of meteorological and reflectance data, and a few auxiliary inputs including the climate zone. LFMC estimation models are designed for two training and evaluation scenarios, one for sites where historical LFMC measurements are available (within-site), the other for sites without historical LFMC measurements (out-of-site). The models were trained and evaluated using a large database of LFMC samples measured in the field from 2001 to 2017 and achieved an RMSE of 20.87% for the within-site scenario and 25.36% for the out-of-site scenario.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Live fuel moisture content (LFMC) is an important environmental indicator used to measure vegetation conditions and monitor for high fire risk conditions. However, LFMC is challenging to measure on a wide scale, thus reliable models for estimating LFMC are needed. Therefore, this paper proposes a new deep learning architecture for LFMC estimation. The architecture comprises an ensemble of temporal convolutional neural networks that learn from year-long time series of meteorological and reflectance data, and a few auxiliary inputs including the climate zone. LFMC estimation models are designed for two training and evaluation scenarios, one for sites where historical LFMC measurements are available (within-site), the other for sites without historical LFMC measurements (out-of-site). The models were trained and evaluated using a large database of LFMC samples measured in the field from 2001 to 2017 and achieved an RMSE of 20.87% for the within-site scenario and 25.36% for the out-of-site scenario.
Iqbal, Shahid; Ge, Fang; Li, Fuyi; Akutsu, Tatsuya; Zheng, Yuanting; Gasser, Robin B.; Yu, Dong-Jun; Webb, Geoffrey I.; Song, Jiangning
Journal of Chemical Information and Modeling, vol. 62, no. 17, pp. 4270-4282, 2022.
@article{Iqbal2022,
title = {PROST: AlphaFold2-aware Sequence-Based Predictor to Estimate Protein Stability Changes upon Missense Mutations},
author = {Shahid Iqbal and Fang Ge and Fuyi Li and Tatsuya Akutsu and Yuanting Zheng and Robin B. Gasser and Dong-Jun Yu and Geoffrey I. Webb and Jiangning Song},
doi = {10.1021/acs.jcim.2c00799},
year = {2022},
date = {2022-01-01},
journal = {Journal of Chemical Information and Modeling},
volume = {62},
number = {17},
pages = {4270-4282},
publisher = {American Chemical Society (ACS)},
abstract = {An essential step in engineering proteins and understanding disease-causing missense mutations is to accurately model protein stability changes when such mutations occur. Here, we developed a new sequence-based predictor for protein stability (PROST) change upon single-point missense mutation. PROST extracts multiple descriptors from the most promising sequence-based predictors, such as BoostDDG, SAAFEC-SEQ, and DDGun. RPOST also extracts descriptors from iFeature and AlphaFold2. The extracted descriptors include sequence-based features, physicochemical properties, evolutionary information, evolutionary-based physicochemical properties, and predicted structural features. The PROST predictor is a weighted average ensemble model based on extreme gradient boosting (XGBoost) decision trees and extra-trees regressor, PROST is trained on both direct and hypothetical reverse mutations using the S5294 (S2647 direct mutations + S2647 inverse mutations). The parameters for the PROST model are optimized using grid searching with 5-fold cross-validation, and feature importance analysis unveils the most relevant features. The performance of PROST is evaluated in a blinded manner, employing nine distinct datasets and existing state-of-the-art sequence-based and structure-based predictors. This method consistently performs well on Frataxin, S217, S349, Ssym, Myoglobin, and CAGI5 datasets in blind tests, and similarly to the state-of-the-art predictors for p53 and S276 datasets. When the performance of PROST is compared with the latest predictors such as BoostDDG, SAAFEC-SEQ, ACDC-NN-seq, and DDGun, PROST dominates these predictors. A case study of mutation scanning of the Frataxin protein for nine wild-type residues demonstrates the utility of PROST. Taken together, these findings indicate that PROST is a well-suited predictor when no protein structural information is available. The source code of PROST, datasets, examples, pre-trained models, along with how to use PROST are available at https://github.com/ShahidIqb/PROST and https://prost.erc.monash.edu/seq.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
An essential step in engineering proteins and understanding disease-causing missense mutations is to accurately model protein stability changes when such mutations occur. Here, we developed a new sequence-based predictor for protein stability (PROST) change upon single-point missense mutation. PROST extracts multiple descriptors from the most promising sequence-based predictors, such as BoostDDG, SAAFEC-SEQ, and DDGun. RPOST also extracts descriptors from iFeature and AlphaFold2. The extracted descriptors include sequence-based features, physicochemical properties, evolutionary information, evolutionary-based physicochemical properties, and predicted structural features. The PROST predictor is a weighted average ensemble model based on extreme gradient boosting (XGBoost) decision trees and extra-trees regressor, PROST is trained on both direct and hypothetical reverse mutations using the S5294 (S2647 direct mutations + S2647 inverse mutations). The parameters for the PROST model are optimized using grid searching with 5-fold cross-validation, and feature importance analysis unveils the most relevant features. The performance of PROST is evaluated in a blinded manner, employing nine distinct datasets and existing state-of-the-art sequence-based and structure-based predictors. This method consistently performs well on Frataxin, S217, S349, Ssym, Myoglobin, and CAGI5 datasets in blind tests, and similarly to the state-of-the-art predictors for p53 and S276 datasets. When the performance of PROST is compared with the latest predictors such as BoostDDG, SAAFEC-SEQ, ACDC-NN-seq, and DDGun, PROST dominates these predictors. A case study of mutation scanning of the Frataxin protein for nine wild-type residues demonstrates the utility of PROST. Taken together, these findings indicate that PROST is a well-suited predictor when no protein structural information is available. The source code of PROST, datasets, examples, pre-trained models, along with how to use PROST are available at https://github.com/ShahidIqb/PROST and https://prost.erc.monash.edu/seq.
Bi, Yue; Li, Fuyi; Guo, Xudong; Wang, Zhikang; Pan, Tong; Guo, Yuming; Webb, Geoffrey I; Yao, Jianhua; Jia, Cangzhi; Song, Jiangning
Briefings in Bioinformatics, vol. 23, no. 6, 2022, ISSN: 1477-4054.
@article{Bi2022,
title = {Clarion is a multi-label problem transformation method for identifying mRNA subcellular localizations},
author = {Yue Bi and Fuyi Li and Xudong Guo and Zhikang Wang and Tong Pan and Yuming Guo and Geoffrey I Webb and Jianhua Yao and Cangzhi Jia and Jiangning Song},
doi = {10.1093/bib/bbac467},
issn = {1477-4054},
year = {2022},
date = {2022-01-01},
journal = {Briefings in Bioinformatics},
volume = {23},
number = {6},
abstract = {Subcellular localization of messenger RNAs (mRNAs) plays a key role in the spatial regulation of gene activity. The functions of mRNAs have been shown to be closely linked with their localizations. As such, understanding of the subcellular localizations of mRNAs can help elucidate gene regulatory networks. Despite several computational methods that have been developed to predict mRNA localizations within cells, there is still much room for improvement in predictive performance, especially for the multiple-location prediction. In this study, we proposed a novel multi-label multi-class predictor, termed Clarion, for mRNA subcellular localization prediction. Clarion was developed based on a manually curated benchmark dataset and leveraged the weighted series method for multi-label transformation. Extensive benchmarking tests demonstrated Clarion achieved competitive predictive performance and the weighted series method plays a crucial role in securing superior performance of Clarion. In addition, the independent test results indicate that Clarion outperformed the state-of-the-art methods and can secure accuracy of 81.47, 91.29, 79.77, 92.10, 89.15, 83.74, 80.74, 79.23 and 84.74% for chromatin, cytoplasm, cytosol, exosome, membrane, nucleolus, nucleoplasm, nucleus and ribosome, respectively. The webserver and local stand-alone tool of Clarion is freely available at http://monash.bioweb.cloud.edu.au/Clarion/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Subcellular localization of messenger RNAs (mRNAs) plays a key role in the spatial regulation of gene activity. The functions of mRNAs have been shown to be closely linked with their localizations. As such, understanding of the subcellular localizations of mRNAs can help elucidate gene regulatory networks. Despite several computational methods that have been developed to predict mRNA localizations within cells, there is still much room for improvement in predictive performance, especially for the multiple-location prediction. In this study, we proposed a novel multi-label multi-class predictor, termed Clarion, for mRNA subcellular localization prediction. Clarion was developed based on a manually curated benchmark dataset and leveraged the weighted series method for multi-label transformation. Extensive benchmarking tests demonstrated Clarion achieved competitive predictive performance and the weighted series method plays a crucial role in securing superior performance of Clarion. In addition, the independent test results indicate that Clarion outperformed the state-of-the-art methods and can secure accuracy of 81.47, 91.29, 79.77, 92.10, 89.15, 83.74, 80.74, 79.23 and 84.74% for chromatin, cytoplasm, cytosol, exosome, membrane, nucleolus, nucleoplasm, nucleus and ribosome, respectively. The webserver and local stand-alone tool of Clarion is freely available at http://monash.bioweb.cloud.edu.au/Clarion/.
Godahewa, Rakshitha W; Bergmeir, Christoph; Webb, Geoffrey; Hyndman, Rob; Montero-Manso, Pablo
Monash Time Series Forecasting Archive
Vanschoren, J.; Yeung, S. (Ed.): Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks, 2021.
@inproceedings{Godahewa2021,
title = {Monash Time Series Forecasting Archive},
author = {Rakshitha W Godahewa and Christoph Bergmeir and Geoffrey Webb and Rob Hyndman and Pablo Montero-Manso},
editor = {J. Vanschoren and S. Yeung},
url = {https://datasets-benchmarks-proceedings.neurips.cc/paper_files/paper/2021/file/eddea82ad2755b24c4e168c5fc2ebd40-Paper-round2.pdf},
year = {2021},
date = {2021-01-01},
booktitle = {Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks},
volume = {1},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Tan, Chang Wei; Bergmeir, Christoph; Petitjean, Francois; Webb, Geoffrey I.
Time series extrinsic regression: Predicting numeric values from time series data
Data Mining and Knowledge Discovery, vol. 35, no. 3, pp. 1032-1060, 2021, ISSN: 1573-756X.
@article{tan2021regression,
title = {Time series extrinsic regression: Predicting numeric values from time series data},
author = {Chang Wei Tan and Christoph Bergmeir and Francois Petitjean and Geoffrey I. Webb},
url = {https://rdcu.be/cgCAn},
doi = {10.1007/s10618-021-00745-9},
issn = {1573-756X},
year = {2021},
date = {2021-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {35},
number = {3},
pages = {1032-1060},
publisher = {Springer US},
abstract = {This paper studies time series extrinsic regression (TSER): a regression task of which the aim is to learn the relationship between a time series and a continuous scalar variable; a task closely related to time series classification (TSC), which aims to learn the relationship between a time series and a categorical class label. This task generalizes time series forecasting, relaxing the requirement that the value predicted be a future value of the input series or primarily depend on more recent values. In this paper, we motivate and study this task, and benchmark existing solutions and adaptations of TSC algorithms on a novel archive of 19 TSER datasets which we have assembled. Our results show that the state-of-the-art TSC algorithm Rocket, when adapted for regression, achieves the highest overall accuracy compared to adaptations of other TSC algorithms and state-of-the-art machine learning (ML) algorithms such as XGBoost, Random Forest and Support Vector Regression. More importantly, we show that much research is needed in this field to improve the accuracy of ML models. We also find evidence that further research has excellent prospects of improving upon these straightforward baselines.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper studies time series extrinsic regression (TSER): a regression task of which the aim is to learn the relationship between a time series and a continuous scalar variable; a task closely related to time series classification (TSC), which aims to learn the relationship between a time series and a categorical class label. This task generalizes time series forecasting, relaxing the requirement that the value predicted be a future value of the input series or primarily depend on more recent values. In this paper, we motivate and study this task, and benchmark existing solutions and adaptations of TSC algorithms on a novel archive of 19 TSER datasets which we have assembled. Our results show that the state-of-the-art TSC algorithm Rocket, when adapted for regression, achieves the highest overall accuracy compared to adaptations of other TSC algorithms and state-of-the-art machine learning (ML) algorithms such as XGBoost, Random Forest and Support Vector Regression. More importantly, we show that much research is needed in this field to improve the accuracy of ML models. We also find evidence that further research has excellent prospects of improving upon these straightforward baselines.
Li, M.; Wang, Y.; Li, F.; Zhao, Y.; Liu, M.; Zhang, S.; Bin, Y.; Smith, A. I.; Webb, G.; Li, J.; Song, J.; Xia, J.
A Deep Learning-Based Method for Identification of Bacteriophage-Host Interaction
IEEE/ACM Trans Comput Biol Bioinform, vol. 18, iss. 5, pp. 1801-1810, 2021, ISSN: 1545-5963.
@article{RN3447,
title = {A Deep Learning-Based Method for Identification of Bacteriophage-Host Interaction},
author = {M. Li and Y. Wang and F. Li and Y. Zhao and M. Liu and S. Zhang and Y. Bin and A. I. Smith and G. Webb and J. Li and J. Song and J. Xia},
doi = {10.1109/tcbb.2020.3017386},
issn = {1545-5963},
year = {2021},
date = {2021-01-01},
journal = {IEEE/ACM Trans Comput Biol Bioinform},
volume = {18},
issue = {5},
pages = {1801-1810},
abstract = {Multi-drug resistance (MDR) has become one of the greatest threats to human health worldwide, and novel treatment methods of infections caused by MDR bacteria are urgently needed. Phage therapy is a promising alternative to solve this problem, to which the key is correctly matching target pathogenic bacteria with the corresponding therapeutic phage. Deep learning is powerful for mining complex patterns to generate accurate predictions. In this study, we develop PredPHI (Predicting Phage-Host Interactions), a deep learning-based tool capable of predicting the host of phages from sequence data. We collect \>3000 phage-host pairs along with their protein sequences from PhagesDB and GenBank databases and extract a set of features. Then we select high-quality negative samples based on the K-Means clustering method and construct a balanced training set. Finally, we employ a deep convolutional neural network to build the predictive model. The results indicate that PredPHI can achieve a predictive performance of 81% in terms of the area under the receiver operating characteristic curve on the test set, and the clustering-based method is significantly more robust than that based on randomly selecting negative samples. These results highlight that PredPHI is a useful and accurate tool for identifying phage-host interactions from sequence data.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Multi-drug resistance (MDR) has become one of the greatest threats to human health worldwide, and novel treatment methods of infections caused by MDR bacteria are urgently needed. Phage therapy is a promising alternative to solve this problem, to which the key is correctly matching target pathogenic bacteria with the corresponding therapeutic phage. Deep learning is powerful for mining complex patterns to generate accurate predictions. In this study, we develop PredPHI (Predicting Phage-Host Interactions), a deep learning-based tool capable of predicting the host of phages from sequence data. We collect >3000 phage-host pairs along with their protein sequences from PhagesDB and GenBank databases and extract a set of features. Then we select high-quality negative samples based on the K-Means clustering method and construct a balanced training set. Finally, we employ a deep convolutional neural network to build the predictive model. The results indicate that PredPHI can achieve a predictive performance of 81% in terms of the area under the receiver operating characteristic curve on the test set, and the clustering-based method is significantly more robust than that based on randomly selecting negative samples. These results highlight that PredPHI is a useful and accurate tool for identifying phage-host interactions from sequence data.
Wang, Yanan; Li, Fuyi; Bharathwaj, Manasa; Rosas, Natalia C; Leier, André; Akutsu, Tatsuya; Webb, Geoffrey I; Marquez-Lago, Tatiana T; Li, Jian; Lithgow, Trevor; Song, Jiangning
DeepBL: a deep learning-based approach for in silico discovery of beta-lactamases
Briefings in Bioinformatics, vol. 22, no. 4, 2021.
@article{Wang2020,
title = {DeepBL: a deep learning-based approach for in silico discovery of beta-lactamases},
author = {Yanan Wang and Fuyi Li and Manasa Bharathwaj and Natalia C Rosas and Andr\'{e} Leier and Tatsuya Akutsu and Geoffrey I Webb and Tatiana T Marquez-Lago and Jian Li and Trevor Lithgow and Jiangning Song},
doi = {10.1093/bib/bbaa301},
year = {2021},
date = {2021-01-01},
journal = {Briefings in Bioinformatics},
volume = {22},
number = {4},
publisher = {Oxford University Press (OUP)},
abstract = {Beta-lactamases (BLs) are enzymes localized in the periplasmic space of bacterial pathogens, where they confer resistance to beta-lactam antibiotics. Experimental identification of BLs is costly yet crucial to understand beta-lactam resistance mechanisms. To address this issue, we present DeepBL, a deep learning-based approach by incorporating sequence-derived features to enable high-throughput prediction of BLs. Specifically, DeepBL is implemented based on the Small VGGNet architecture and the TensorFlow deep learning library. Furthermore, the performance of DeepBL models is investigated in relation to the sequence redundancy level and negative sample selection in the benchmark dataset. The models are trained on datasets of varying sequence redundancy thresholds, and the model performance is evaluated by extensive benchmarking tests. Using the optimized DeepBL model, we perform proteome-wide screening for all reviewed bacterium protein sequences available from the UniProt database. These results are freely accessible at the DeepBL webserver at http://deepbl.erc.monash.edu.au/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Beta-lactamases (BLs) are enzymes localized in the periplasmic space of bacterial pathogens, where they confer resistance to beta-lactam antibiotics. Experimental identification of BLs is costly yet crucial to understand beta-lactam resistance mechanisms. To address this issue, we present DeepBL, a deep learning-based approach by incorporating sequence-derived features to enable high-throughput prediction of BLs. Specifically, DeepBL is implemented based on the Small VGGNet architecture and the TensorFlow deep learning library. Furthermore, the performance of DeepBL models is investigated in relation to the sequence redundancy level and negative sample selection in the benchmark dataset. The models are trained on datasets of varying sequence redundancy thresholds, and the model performance is evaluated by extensive benchmarking tests. Using the optimized DeepBL model, we perform proteome-wide screening for all reviewed bacterium protein sequences available from the UniProt database. These results are freely accessible at the DeepBL webserver at http://deepbl.erc.monash.edu.au/.
Mei, Shutao; Li, Fuyi; Xiang, Dongxu; Ayala, Rochelle; Faridi, Pouya; Webb, Geoffrey I; Illing, Patricia T; Rossjohn, Jamie; Akutsu, Tatsuya; Croft, Nathan P; Purcell, Anthony W; Song, Jiangning
Briefings in Bioinformatics, vol. 22, iss. 5, 2021, ISSN: 1477-4054.
@article{Mei2021,
title = {Anthem: a user customised tool for fast and accurate prediction of binding between peptides and HLA class I molecules},
author = {Shutao Mei and Fuyi Li and Dongxu Xiang and Rochelle Ayala and Pouya Faridi and Geoffrey I Webb and Patricia T Illing and Jamie Rossjohn and Tatsuya Akutsu and Nathan P Croft and Anthony W Purcell and Jiangning Song},
doi = {10.1093/bib/bbaa415},
issn = {1477-4054},
year = {2021},
date = {2021-01-01},
journal = {Briefings in Bioinformatics},
volume = {22},
issue = {5},
abstract = {Neopeptide-based immunotherapy has been recognised as a promising approach for the treatment of cancers. For neopeptides to be recognised by CD8+ T cells and induce an immune response, their binding to human leukocyte antigen class I (HLA-I) molecules is a necessary first step. Most epitope prediction tools thus rely on the prediction of such binding. With the use of mass spectrometry, the scale of naturally presented HLA ligands that could be used to develop such predictors has been expanded. However, there are rarely efforts that focus on the integration of these experimental data with computational algorithms to efficiently develop up-to-date predictors. Here, we present Anthem for accurate HLA-I binding prediction. In particular, we have developed a user-friendly framework to support the development of customisable HLA-I binding prediction models to meet challenges associated with the rapidly increasing availability of large amounts of immunopeptidomic data. Our extensive evaluation, using both independent and experimental datasets shows that Anthem achieves an overall similar or higher area under curve value compared with other contemporary tools. It is anticipated that Anthem will provide a unique opportunity for the non-expert user to analyse and interpret their own in-house or publicly deposited datasets.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neopeptide-based immunotherapy has been recognised as a promising approach for the treatment of cancers. For neopeptides to be recognised by CD8+ T cells and induce an immune response, their binding to human leukocyte antigen class I (HLA-I) molecules is a necessary first step. Most epitope prediction tools thus rely on the prediction of such binding. With the use of mass spectrometry, the scale of naturally presented HLA ligands that could be used to develop such predictors has been expanded. However, there are rarely efforts that focus on the integration of these experimental data with computational algorithms to efficiently develop up-to-date predictors. Here, we present Anthem for accurate HLA-I binding prediction. In particular, we have developed a user-friendly framework to support the development of customisable HLA-I binding prediction models to meet challenges associated with the rapidly increasing availability of large amounts of immunopeptidomic data. Our extensive evaluation, using both independent and experimental datasets shows that Anthem achieves an overall similar or higher area under curve value compared with other contemporary tools. It is anticipated that Anthem will provide a unique opportunity for the non-expert user to analyse and interpret their own in-house or publicly deposited datasets.
Webb, Geoffrey I.; Petitjean, François
Tight lower bounds for Dynamic Time Warping
Pattern Recognition, vol. 115, 2021, ISSN: 0031-3203.
@article{WEBB2021107895,
title = {Tight lower bounds for Dynamic Time Warping},
author = {Geoffrey I. Webb and Fran\c{c}ois Petitjean},
doi = {10.1016/j.patcog.2021.107895},
issn = {0031-3203},
year = {2021},
date = {2021-01-01},
journal = {Pattern Recognition},
volume = {115},
abstract = {Dynamic Time Warping (DTW) is a popular similarity measure for aligning and comparing time series. Due to DTW's high computation time, lower bounds are often employed to screen poor matches. Many alternative lower bounds have been proposed, providing a range of different trade-offs between tightness and computational efficiency. LB_KEOGH provides a useful trade-off in many applications. Two recent lower bounds, LB_IMPROVED and LB_ENHANCED, are substantially tighter than LB_KEOGH. All three have the same worst case computational complexity - linear with respect to series length and constant with respect to window size. We present four new DTW lower bounds in the same complexity class. LB_PETITJEAN is substantially tighter than LB_IMPROVED, with only modest additional computational overhead. LB_WEBB is more efficient than LB_IMPROVED, while often providing a tighter bound. LB_WEBB is always tighter than LB_KEOGH. The parameter free LB_WEBB is usually tighter than LB_ENHANCED. A parameterized variant, LB_Webb_Enhanced, is always tighter than LB_ENHANCED. A further variant, LB_WEBB*, is useful for some constrained distance functions. In extensive experiments, LB_WEBB proves to be very effective for nearest neighbor search.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Dynamic Time Warping (DTW) is a popular similarity measure for aligning and comparing time series. Due to DTW's high computation time, lower bounds are often employed to screen poor matches. Many alternative lower bounds have been proposed, providing a range of different trade-offs between tightness and computational efficiency. LB_KEOGH provides a useful trade-off in many applications. Two recent lower bounds, LB_IMPROVED and LB_ENHANCED, are substantially tighter than LB_KEOGH. All three have the same worst case computational complexity - linear with respect to series length and constant with respect to window size. We present four new DTW lower bounds in the same complexity class. LB_PETITJEAN is substantially tighter than LB_IMPROVED, with only modest additional computational overhead. LB_WEBB is more efficient than LB_IMPROVED, while often providing a tighter bound. LB_WEBB is always tighter than LB_KEOGH. The parameter free LB_WEBB is usually tighter than LB_ENHANCED. A parameterized variant, LB_Webb_Enhanced, is always tighter than LB_ENHANCED. A further variant, LB_WEBB*, is useful for some constrained distance functions. In extensive experiments, LB_WEBB proves to be very effective for nearest neighbor search.
Krempl, Georg; Hofer, Vera; Webb, Geoffrey; Hullermeier, Eyke
Beyond Adaptation: Understanding Distributional Changes (Dagstuhl Seminar 20372)
Schloss Dagstuhl - Leibniz-Zentrum fur Informatik 2021.
@techreport{Krempl2021,
title = {Beyond Adaptation: Understanding Distributional Changes (Dagstuhl Seminar 20372)},
author = {Georg Krempl and Vera Hofer and Geoffrey Webb and Eyke Hullermeier},
doi = {10.4230/DAGREP.10.4.1},
year = {2021},
date = {2021-01-01},
institution = {Schloss Dagstuhl - Leibniz-Zentrum fur Informatik},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
Chen, Zhen; Zhao, Pei; Li, Chen; Li, Fuyi; Xiang, Dongxu; Chen, Yong-Zi; Akutsu, Tatsuya; Daly, Roger J; Webb, Geoffrey I; Zhao, Quanzhi; Kurgan, Lukasz; Song, Jiangning
Nucleic Acids Research, vol. 49, no. 10, 2021, ISSN: 0305-1048.
@article{ChenEtAl21,
title = {iLearnPlus: a comprehensive and automated machine-learning platform for nucleic acid and protein sequence analysis, prediction and visualization},
author = {Zhen Chen and Pei Zhao and Chen Li and Fuyi Li and Dongxu Xiang and Yong-Zi Chen and Tatsuya Akutsu and Roger J Daly and Geoffrey I Webb and Quanzhi Zhao and Lukasz Kurgan and Jiangning Song},
doi = {10.1093/nar/gkab122},
issn = {0305-1048},
year = {2021},
date = {2021-01-01},
journal = {Nucleic Acids Research},
volume = {49},
number = {10},
abstract = {Sequence-based analysis and prediction are fundamental bioinformatic tasks that facilitate understanding of the sequence(-structure)-function paradigm for DNAs, RNAs and proteins. Rapid accumulation of sequences requires equally pervasive development of new predictive models, which depends on the availability of effective tools that support these efforts. We introduce iLearnPlus, the first machine-learning platform with graphical- and web-based interfaces for the construction of machine-learning pipelines for analysis and predictions using nucleic acid and protein sequences. iLearnPlus provides a comprehensive set of algorithms and automates sequence-based feature extraction and analysis, construction and deployment of models, assessment of predictive performance, statistical analysis, and data visualization; all without programming. iLearnPlus includes a wide range of feature sets which encode information from the input sequences and over twenty machine-learning algorithms that cover several deep-learning approaches, outnumbering the current solutions by a wide margin. Our solution caters to experienced bioinformaticians, given the broad range of options, and biologists with no programming background, given the point-and-click interface and easy-to-follow design process. We showcase iLearnPlus with two case studies concerning prediction of long noncoding RNAs (lncRNAs) from RNA transcripts and prediction of crotonylation sites in protein chains. iLearnPlus is an open-source platform available at https://github.com/Superzchen/iLearnPlus/ with the webserver at http://ilearnplus.erc.monash.edu/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Sequence-based analysis and prediction are fundamental bioinformatic tasks that facilitate understanding of the sequence(-structure)-function paradigm for DNAs, RNAs and proteins. Rapid accumulation of sequences requires equally pervasive development of new predictive models, which depends on the availability of effective tools that support these efforts. We introduce iLearnPlus, the first machine-learning platform with graphical- and web-based interfaces for the construction of machine-learning pipelines for analysis and predictions using nucleic acid and protein sequences. iLearnPlus provides a comprehensive set of algorithms and automates sequence-based feature extraction and analysis, construction and deployment of models, assessment of predictive performance, statistical analysis, and data visualization; all without programming. iLearnPlus includes a wide range of feature sets which encode information from the input sequences and over twenty machine-learning algorithms that cover several deep-learning approaches, outnumbering the current solutions by a wide margin. Our solution caters to experienced bioinformaticians, given the broad range of options, and biologists with no programming background, given the point-and-click interface and easy-to-follow design process. We showcase iLearnPlus with two case studies concerning prediction of long noncoding RNAs (lncRNAs) from RNA transcripts and prediction of crotonylation sites in protein chains. iLearnPlus is an open-source platform available at https://github.com/Superzchen/iLearnPlus/ with the webserver at http://ilearnplus.erc.monash.edu/.
Boley, Mario; Teshuva, Simon; Bodic, Pierre Le; Webb, Geoffrey I
Better Short than Greedy: Interpretable Models through Optimal Rule Boosting
Proceedings of the 2021 SIAM International Conference on Data Mining (SDM), pp. 351-359, SIAM 2021.
@inproceedings{boley2021better,
title = {Better Short than Greedy: Interpretable Models through Optimal Rule Boosting},
author = {Mario Boley and Simon Teshuva and Pierre Le Bodic and Geoffrey I Webb},
url = {https://arxiv.org/abs/2101.08380},
year = {2021},
date = {2021-01-01},
booktitle = {Proceedings of the 2021 SIAM International Conference on Data Mining (SDM)},
pages = {351-359},
organization = {SIAM},
abstract = {Rule ensembles are designed to provide a useful trade-off between predictive accuracy and model interpretability. However, the myopic and random search components of current rule ensemble methods can compromise this goal: they often need more rules than necessary to reach a certain accuracy level or can even outright fail to accurately model a distribution that can actually be described well with a few rules. Here, we present a novel approach aiming to fit rule ensembles of maximal predictive power for a given ensemble size (and thus model comprehensibility). In particular, we present an efficient branch-and-bound algorithm that optimally solves the per-rule objective function of the popular second-order gradient boosting framework. Our main insight is that the boosting objective can be tightly bounded in linear time of the number of covered data points. Along with an additional novel pruning technique related to rule redundancy, this leads to a computationally feasible approach for boosting optimal rules that, as we demonstrate on a wide range of common benchmark problems, consistently outperforms the predictive performance of boosting greedy rules.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Rule ensembles are designed to provide a useful trade-off between predictive accuracy and model interpretability. However, the myopic and random search components of current rule ensemble methods can compromise this goal: they often need more rules than necessary to reach a certain accuracy level or can even outright fail to accurately model a distribution that can actually be described well with a few rules. Here, we present a novel approach aiming to fit rule ensembles of maximal predictive power for a given ensemble size (and thus model comprehensibility). In particular, we present an efficient branch-and-bound algorithm that optimally solves the per-rule objective function of the popular second-order gradient boosting framework. Our main insight is that the boosting objective can be tightly bounded in linear time of the number of covered data points. Along with an additional novel pruning technique related to rule redundancy, this leads to a computationally feasible approach for boosting optimal rules that, as we demonstrate on a wide range of common benchmark problems, consistently outperforms the predictive performance of boosting greedy rules.
Wang, Yanan; Coudray, Nicolas; Zhao, Yun; Li, Fuyi; Hu, Changyuan; Zhang, Yao-Zhong; Imoto, Seiya; Tsirigos, Aristotelis; Webb, Geoffrey I; Daly, Roger J; Song, Jiangning
HEAL: an automated deep learning framework for cancer histopathology image analysis
Bioinformatics, vol. 37, no. 22, pp. 4291-4295, 2021.
@article{Wang2021,
title = {HEAL: an automated deep learning framework for cancer histopathology image analysis},
author = {Yanan Wang and Nicolas Coudray and Yun Zhao and Fuyi Li and Changyuan Hu and Yao-Zhong Zhang and Seiya Imoto and Aristotelis Tsirigos and Geoffrey I Webb and Roger J Daly and Jiangning Song},
doi = {10.1093/bioinformatics/btab380},
year = {2021},
date = {2021-01-01},
journal = {Bioinformatics},
volume = {37},
number = {22},
pages = {4291-4295},
publisher = {Oxford University Press (OUP)},
abstract = {Digital pathology supports analysis of histopathological images using deep learning methods at a large-scale. However, applications of deep learning in this area have been limited by the complexities of configuration of the computational environment and of hyperparameter optimization, which hinder deployment and reduce reproducibility.Here, we propose HEAL, a deep learning-based automated framework for easy, flexible, and multi-faceted histopathological image analysis. We demonstrate its utility and functionality by performing two case studies on lung cancer and one on colon cancer. Leveraging the capability of Docker, HEAL represents an ideal end-to-end tool to conduct complex histopathological analysis and enables deep learning in a broad range of applications for cancer image analysis.Supplementary data are available at Bioinformatics online.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Digital pathology supports analysis of histopathological images using deep learning methods at a large-scale. However, applications of deep learning in this area have been limited by the complexities of configuration of the computational environment and of hyperparameter optimization, which hinder deployment and reduce reproducibility.Here, we propose HEAL, a deep learning-based automated framework for easy, flexible, and multi-faceted histopathological image analysis. We demonstrate its utility and functionality by performing two case studies on lung cancer and one on colon cancer. Leveraging the capability of Docker, HEAL represents an ideal end-to-end tool to conduct complex histopathological analysis and enables deep learning in a broad range of applications for cancer image analysis.Supplementary data are available at Bioinformatics online.
Dempster, Angus; Schmidt, Daniel F.; Webb, Geoffrey I.
MINIROCKET: A Very Fast (Almost) Deterministic Transform for Time Series Classification
Proceedings of the 27thACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 248-257, 2021.
@inproceedings{dempsteretal21kdd,
title = {MINIROCKET: A Very Fast (Almost) Deterministic Transform for Time Series Classification},
author = {Angus Dempster and Daniel F. Schmidt and Geoffrey I. Webb},
url = {https://arxiv.org/abs/2012.08791},
doi = {10.1145/3447548.3467231},
year = {2021},
date = {2021-01-01},
booktitle = {Proceedings of the 27thACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages = {248-257},
abstract = {Until recently, the most accurate methods for time series classification were limited by high computational complexity. ROCKET achieves state-of-the-art accuracy with a fraction of the computational expense of most existing methods by transforming input time series using random convolutional kernels, and using the transformed features to train a linear classifier. We reformulate ROCKET into a new method, MINIROCKET, making it up to 75 times faster on larger datasets, and making it almost deterministic (and optionally, with additional computational expense, fully deterministic), while maintaining essentially the same accuracy. Using this method, it is possible to train and test a classifier on all of 109 datasets from the UCR archive to state-of-the-art accuracy in less than 10 minutes. MINIROCKET is significantly faster than any other method of comparable accuracy (including ROCKET), and significantly more accurate than any other method of even roughly-similar computational expense. As such, we suggest that MINIROCKET should now be considered and used as the default variant of ROCKET.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Until recently, the most accurate methods for time series classification were limited by high computational complexity. ROCKET achieves state-of-the-art accuracy with a fraction of the computational expense of most existing methods by transforming input time series using random convolutional kernels, and using the transformed features to train a linear classifier. We reformulate ROCKET into a new method, MINIROCKET, making it up to 75 times faster on larger datasets, and making it almost deterministic (and optionally, with additional computational expense, fully deterministic), while maintaining essentially the same accuracy. Using this method, it is possible to train and test a classifier on all of 109 datasets from the UCR archive to state-of-the-art accuracy in less than 10 minutes. MINIROCKET is significantly faster than any other method of comparable accuracy (including ROCKET), and significantly more accurate than any other method of even roughly-similar computational expense. As such, we suggest that MINIROCKET should now be considered and used as the default variant of ROCKET.
Iqbal, Shahid; Li, Fuyi; Akutsu, Tatsuya; Ascher, David B; Webb, Geoffrey I; Song, Jiangning
Briefings in Bioinformatics, vol. 22, no. 6, 2021, ISSN: 1477-4054.
@article{10.1093/bib/bbab184,
title = {Assessing the performance of computational predictors for estimating protein stability changes upon missense mutations},
author = {Shahid Iqbal and Fuyi Li and Tatsuya Akutsu and David B Ascher and Geoffrey I Webb and Jiangning Song},
doi = {10.1093/bib/bbab184},
issn = {1477-4054},
year = {2021},
date = {2021-01-01},
journal = {Briefings in Bioinformatics},
volume = {22},
number = {6},
abstract = {Understanding how a mutation might affect protein stability is of significant importance to protein engineering and for understanding protein evolution genetic diseases. While a number of computational tools have been developed to predict the effect of missense mutations on protein stability protein stability upon mutations, they are known to exhibit large biases imparted in part by the data used to train and evaluate them. Here, we provide a comprehensive overview of predictive tools, which has provided an evolving insight into the importance and relevance of features that can discern the effects of mutations on protein stability. A diverse selection of these freely available tools was benchmarked using a large mutation-level blind dataset of 1342 experimentally characterised mutations across 130 proteins from ThermoMutDB, a second test dataset encompassing 630 experimentally characterised mutations across 39 proteins from iStable2.0 and a third blind test dataset consisting of 268 mutations in 27 proteins from the newly published ProThermDB. The performance of the methods was further evaluated with respect to the site of mutation, type of mutant residue and by ranging the pH and temperature. Additionally, the classification performance was also evaluated by classifying the mutations as stabilizing (delta delta G\>=0) or destabilizing (delta delta G\<0). The results reveal that the performance of the predictors is affected by the site of mutation and the type of mutant residue. Further, the results show very low performance for pH values 6-8 and temperature higher than 65 for all predictors except iStable2.0 on the S630 dataset. To illustrate how stability and structure change upon single point mutation, we considered four stabilizing, two destabilizing and two stabilizing mutations from two proteins, namely the toxin protein and bovine liver cytochrome. Overall, the results on S268, S630 and S1342 datasets show that the performance of the integrated predictors is better than the mechanistic or individual machine learning predictors. We expect that this paper will provide useful guidance for the design and development of next-generation bioinformatic tools for predicting protein stability changes upon mutations.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Understanding how a mutation might affect protein stability is of significant importance to protein engineering and for understanding protein evolution genetic diseases. While a number of computational tools have been developed to predict the effect of missense mutations on protein stability protein stability upon mutations, they are known to exhibit large biases imparted in part by the data used to train and evaluate them. Here, we provide a comprehensive overview of predictive tools, which has provided an evolving insight into the importance and relevance of features that can discern the effects of mutations on protein stability. A diverse selection of these freely available tools was benchmarked using a large mutation-level blind dataset of 1342 experimentally characterised mutations across 130 proteins from ThermoMutDB, a second test dataset encompassing 630 experimentally characterised mutations across 39 proteins from iStable2.0 and a third blind test dataset consisting of 268 mutations in 27 proteins from the newly published ProThermDB. The performance of the methods was further evaluated with respect to the site of mutation, type of mutant residue and by ranging the pH and temperature. Additionally, the classification performance was also evaluated by classifying the mutations as stabilizing (delta delta G>=0) or destabilizing (delta delta G<0). The results reveal that the performance of the predictors is affected by the site of mutation and the type of mutant residue. Further, the results show very low performance for pH values 6-8 and temperature higher than 65 for all predictors except iStable2.0 on the S630 dataset. To illustrate how stability and structure change upon single point mutation, we considered four stabilizing, two destabilizing and two stabilizing mutations from two proteins, namely the toxin protein and bovine liver cytochrome. Overall, the results on S268, S630 and S1342 datasets show that the performance of the integrated predictors is better than the mechanistic or individual machine learning predictors. We expect that this paper will provide useful guidance for the design and development of next-generation bioinformatic tools for predicting protein stability changes upon mutations.
Wang, Yanan; Yang, Litao; Webb, Geoffrey I; Ge, Zongyuan; Song, Jiangning
OCTID: a one-class learning-based Python package for tumor image detection
Bioinformatics, vol. 37, no. 21, pp. 3986-3988, 2021, ISSN: 1367-4803.
@article{10.1093/bioinformatics/btab416,
title = {OCTID: a one-class learning-based Python package for tumor image detection},
author = {Yanan Wang and Litao Yang and Geoffrey I Webb and Zongyuan Ge and Jiangning Song},
doi = {10.1093/bioinformatics/btab416},
issn = {1367-4803},
year = {2021},
date = {2021-01-01},
journal = {Bioinformatics},
volume = {37},
number = {21},
pages = {3986-3988},
abstract = {Tumor tile selection is a necessary prerequisite in patch-based cancer whole slide image analysis, which is labor-intensive and requires expertise. Whole slides are annotated as tumor or tumor free, but tiles within a tumor slide are not. As all tiles within a tumor free slide are tumor free, these can be used to capture tumor-free patterns using the one-class learning strategy. We present a Python package, termed OCTID, which combines a pretrained convolutional neural network (CNN) model, Uniform Manifold Approximation and Projection (UMAP) and one-class support vector machine to achieve accurate tumor tile classification using a training set of tumor free tiles. Benchmarking experiments on four H\&E image datasets achieved remarkable performance in terms of F1-score (0.90?+/-0.06), Matthews correlation coefficient (0.93?+/-0.05) and accuracy (0.94?+/-0.03).Detailed information can be found in the Supplementary File.Supplementary data are available at Bioinformatics online.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Tumor tile selection is a necessary prerequisite in patch-based cancer whole slide image analysis, which is labor-intensive and requires expertise. Whole slides are annotated as tumor or tumor free, but tiles within a tumor slide are not. As all tiles within a tumor free slide are tumor free, these can be used to capture tumor-free patterns using the one-class learning strategy. We present a Python package, termed OCTID, which combines a pretrained convolutional neural network (CNN) model, Uniform Manifold Approximation and Projection (UMAP) and one-class support vector machine to achieve accurate tumor tile classification using a training set of tumor free tiles. Benchmarking experiments on four H&E image datasets achieved remarkable performance in terms of F1-score (0.90?+/-0.06), Matthews correlation coefficient (0.93?+/-0.05) and accuracy (0.94?+/-0.03).Detailed information can be found in the Supplementary File.Supplementary data are available at Bioinformatics online.
Zhu, Liujun; Webb, Geoffrey I.; Yebra, Marta; Scortechini, Gianluca; Miller, Lynn; Petitjean, Francois
Live fuel moisture content estimation from MODIS: A deep learning approach
ISPRS Journal of Photogrammetry and Remote Sensing, vol. 179, pp. 81-91, 2021, ISSN: 0924-2716.
@article{ZHU202181,
title = {Live fuel moisture content estimation from MODIS: A deep learning approach},
author = {Liujun Zhu and Geoffrey I. Webb and Marta Yebra and Gianluca Scortechini and Lynn Miller and Francois Petitjean},
doi = {10.1016/j.isprsjprs.2021.07.010},
issn = {0924-2716},
year = {2021},
date = {2021-01-01},
journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
volume = {179},
pages = {81-91},
abstract = {Live fuel moisture content (LFMC) is an essential variable to model fire danger and behaviour. This paper presents the first application of deep learning to LFMC estimation based on the historical LFMC ground samples of the Globe-LFMC database, as a step towards operational daily LFMC mapping in the Contiguous United States (CONUS). One-year MODerate resolution Imaging Spectroradiometer (MODIS) time series preceding each LFMC sample were extracted as the primary data source for training. The proposed temporal convolutional neural network for LFMC (TempCNN-LFMC) comprises three 1-D convolutional layers that learn the multi-scale temporal dynamics (features) of one-year MODIS time series specific to LFMC estimation. The learned features, together with a few auxiliary variables (e.g., digital elevation model), are then passed to three fully connected layers to extract the non-linear relationships with LFMC. In the primary training and validation scenario, the neural network was trained using samples from 2002 to 2013 and then adopted to estimating the LFMC from 2014 to 2018, achieving an overall root mean square error (RMSE) of 25.57% and a correlation coefficient (R) of 0.74. Good consistency on spatial patterns and temporal trends of accuracy was observed. The trained model achieved a similar RMSE of 25.98%, 25.20% and 25.93% for forest, shrubland, and grassland, respectively, without requiring prior information on the vegetation type.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Live fuel moisture content (LFMC) is an essential variable to model fire danger and behaviour. This paper presents the first application of deep learning to LFMC estimation based on the historical LFMC ground samples of the Globe-LFMC database, as a step towards operational daily LFMC mapping in the Contiguous United States (CONUS). One-year MODerate resolution Imaging Spectroradiometer (MODIS) time series preceding each LFMC sample were extracted as the primary data source for training. The proposed temporal convolutional neural network for LFMC (TempCNN-LFMC) comprises three 1-D convolutional layers that learn the multi-scale temporal dynamics (features) of one-year MODIS time series specific to LFMC estimation. The learned features, together with a few auxiliary variables (e.g., digital elevation model), are then passed to three fully connected layers to extract the non-linear relationships with LFMC. In the primary training and validation scenario, the neural network was trained using samples from 2002 to 2013 and then adopted to estimating the LFMC from 2014 to 2018, achieving an overall root mean square error (RMSE) of 25.57% and a correlation coefficient (R) of 0.74. Good consistency on spatial patterns and temporal trends of accuracy was observed. The trained model achieved a similar RMSE of 25.98%, 25.20% and 25.93% for forest, shrubland, and grassland, respectively, without requiring prior information on the vegetation type.
Herrmann, Matthieu; Webb, Geoffrey I.
Early abandoning and pruning for elastic distances including dynamic time warping
Data Mining and Knowledge Discovery, vol. 35, no. 6, pp. 2577-2601, 2021.
@article{Herrmann_2021,
title = {Early abandoning and pruning for elastic distances including dynamic time warping},
author = {Matthieu Herrmann and Geoffrey I. Webb},
url = {https://rdcu.be/cuoN0},
doi = {10.1007/s10618-021-00782-4},
year = {2021},
date = {2021-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {35},
number = {6},
pages = {2577-2601},
publisher = {Springer Science and Business Media LLC},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Godahewa, Rakshitha; Bandara, Kasun; Webb, Geoffrey I.; Smyl, Slawek; Bergmeir, Christoph
Ensembles of localised models for time series forecasting
Knowledge-Based Systems, vol. 233, 2021, ISSN: 0950-7051.
@article{GODAHEWA2021107518,
title = {Ensembles of localised models for time series forecasting},
author = {Rakshitha Godahewa and Kasun Bandara and Geoffrey I. Webb and Slawek Smyl and Christoph Bergmeir},
doi = {10.1016/j.knosys.2021.107518},
issn = {0950-7051},
year = {2021},
date = {2021-01-01},
journal = {Knowledge-Based Systems},
volume = {233},
abstract = {With large quantities of data typically available nowadays, forecasting models that are trained across sets of time series, known as Global Forecasting Models (GFM), are regularly outperforming traditional univariate forecasting models that work on isolated series. As GFMs usually share the same set of parameters across all time series, they often have the problem of not being localised enough to a particular series, especially in situations where datasets are heterogeneous. We study how ensembling techniques can be used with generic GFMs and univariate models to solve this issue. Our work systematises and compares relevant current approaches, namely clustering series and training separate submodels per cluster, the so-called ensemble of specialists approach, and building heterogeneous ensembles of global and local models. We fill some gaps in the existing GFM localisation approaches, in particular by incorporating varied clustering techniques such as feature-based clustering, distance-based clustering and random clustering, and generalise them to use different underlying GFM model types. We then propose a new methodology of clustered ensembles where we train multiple GFMs on different clusters of series, obtained by changing the number of clusters and cluster seeds. Using Feed-forward Neural Networks, Recurrent Neural Networks, and Pooled Regression models as the underlying GFMs, in our evaluation on eight publicly available datasets, the proposed models are able to achieve significantly higher accuracy than baseline GFM models and univariate forecasting methods.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
With large quantities of data typically available nowadays, forecasting models that are trained across sets of time series, known as Global Forecasting Models (GFM), are regularly outperforming traditional univariate forecasting models that work on isolated series. As GFMs usually share the same set of parameters across all time series, they often have the problem of not being localised enough to a particular series, especially in situations where datasets are heterogeneous. We study how ensembling techniques can be used with generic GFMs and univariate models to solve this issue. Our work systematises and compares relevant current approaches, namely clustering series and training separate submodels per cluster, the so-called ensemble of specialists approach, and building heterogeneous ensembles of global and local models. We fill some gaps in the existing GFM localisation approaches, in particular by incorporating varied clustering techniques such as feature-based clustering, distance-based clustering and random clustering, and generalise them to use different underlying GFM model types. We then propose a new methodology of clustered ensembles where we train multiple GFMs on different clusters of series, obtained by changing the number of clusters and cluster seeds. Using Feed-forward Neural Networks, Recurrent Neural Networks, and Pooled Regression models as the underlying GFMs, in our evaluation on eight publicly available datasets, the proposed models are able to achieve significantly higher accuracy than baseline GFM models and univariate forecasting methods.
Tan, Chang Wei; Herrmann, Matthieu; Webb, Geoffrey I.
Ultra fast warping window optimization for Dynamic Time Warping
IEEE International Conference on Data Mining (ICDM-21), pp. 589-598, 2021.
@inproceedings{TanEtAlUltraFast2021,
title = {Ultra fast warping window optimization for Dynamic Time Warping},
author = {Chang Wei Tan and Matthieu Herrmann and Geoffrey I. Webb},
url = {https://changweitan.com/research/UltraFastWWSearch.pdf},
doi = {10.1109/ICDM51629.2021.00070},
year = {2021},
date = {2021-01-01},
booktitle = {IEEE International Conference on Data Mining (ICDM-21)},
pages = {589-598},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Chen, Zhen; Zhao, Pei; Li, Fuyi; Leier, André; Marquez-Lago, Tatiana T.; Webb, Geoffrey I.; Baggag, Abdelkader; Bensmail, Halima; Song, Jiangning
PROSPECT: A web server for predicting protein histidine phosphorylation sites
Journal of Bioinformatics and Computational Biology, vol. 18, no. 4, 2020.
@article{Chen2020,
title = {PROSPECT: A web server for predicting protein histidine phosphorylation sites},
author = {Zhen Chen and Pei Zhao and Fuyi Li and Andr\'{e} Leier and Tatiana T. Marquez-Lago and Geoffrey I. Webb and Abdelkader Baggag and Halima Bensmail and Jiangning Song},
doi = {10.1142/s0219720020500183},
year = {2020},
date = {2020-06-01},
journal = {Journal of Bioinformatics and Computational Biology},
volume = {18},
number = {4},
publisher = {World Scientific},
abstract = {Background: Phosphorylation of histidine residues plays crucial roles in signaling pathwaysand cell metabolism in prokaryotes such as bacteria. While evidence has emerged that proteinhistidine phosphorylation also occurs in more complex organisms, its role in mammalian cellshas remained largely uncharted. Thus, it is highly desirable to develop computational tools thatare able to identify histidine phosphorylation sites.Result:Here, we introduce PROSPECT thatenables fast and accurate prediction of proteome-wide histidine phosphorylation substrates andsites. Our tool is based on a hybrid method that integrates the outputs of two convolutional neuralnetwork (CNN)-based classifiers and a random forest-based classifier. Three features, includingthe one-of-K coding, enhanced grouped amino acids content (EGAAC) and composition of k-spaced amino acid group pairs (CKSAAGP) encoding, were taken as the input to three classifiers,respectively. Our results show that it is able to accurately predict histidine phosphorylation sitesfrom sequence information. Our PROSPECT web server is user-friendly and publicly available athttp://PROSPECT.erc.monash.edu/. Conclusions: PROSPECT is superior than other pHispredictors in both the running speed and prediction accuracy and we anticipate that thePROSPECT webserver will become a popular tool for identifying the pHis sites in bacteria.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Background: Phosphorylation of histidine residues plays crucial roles in signaling pathwaysand cell metabolism in prokaryotes such as bacteria. While evidence has emerged that proteinhistidine phosphorylation also occurs in more complex organisms, its role in mammalian cellshas remained largely uncharted. Thus, it is highly desirable to develop computational tools thatare able to identify histidine phosphorylation sites.Result:Here, we introduce PROSPECT thatenables fast and accurate prediction of proteome-wide histidine phosphorylation substrates andsites. Our tool is based on a hybrid method that integrates the outputs of two convolutional neuralnetwork (CNN)-based classifiers and a random forest-based classifier. Three features, includingthe one-of-K coding, enhanced grouped amino acids content (EGAAC) and composition of k-spaced amino acid group pairs (CKSAAGP) encoding, were taken as the input to three classifiers,respectively. Our results show that it is able to accurately predict histidine phosphorylation sitesfrom sequence information. Our PROSPECT web server is user-friendly and publicly available athttp://PROSPECT.erc.monash.edu/. Conclusions: PROSPECT is superior than other pHispredictors in both the running speed and prediction accuracy and we anticipate that thePROSPECT webserver will become a popular tool for identifying the pHis sites in bacteria.
Pratama, M.; Pedrycz, W.; Webb, G. I.
IEEE Transactions on Fuzzy Systems, vol. 28, no. 7, pp. 1315-1328, 2020, ISSN: 1063-6706.
@article{Pratama19,
title = {An Incremental Construction of Deep Neuro Fuzzy System for Continual Learning of Non-stationary Data Streams},
author = {M. Pratama and W. Pedrycz and G. I. Webb},
doi = {10.1109/TFUZZ.2019.2939993},
issn = {1063-6706},
year = {2020},
date = {2020-01-01},
journal = {IEEE Transactions on Fuzzy Systems},
volume = {28},
number = {7},
pages = {1315-1328},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Shifaz, Ahmed; Pelletier, Charlotte; Petitjean, Francois; Webb, Geoffrey I
TS-CHIEF: A Scalable and Accurate Forest Algorithm for Time Series Classification
Data Mining and Knowledge Discovery, vol. 34, no. 3, pp. 742-775, 2020.
@article{shifazetal2019,
title = {TS-CHIEF: A Scalable and Accurate Forest Algorithm for Time Series Classification},
author = {Ahmed Shifaz and Charlotte Pelletier and Francois Petitjean and Geoffrey I Webb},
url = {https://rdcu.be/c1zg6},
doi = {10.1007/s10618-020-00679-8},
year = {2020},
date = {2020-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {34},
number = {3},
pages = {742-775},
abstract = {Time Series Classification (TSC) has seen enormous progress over the last two decades. HIVE-COTE (Hierarchical Vote Collective of Transformation-based Ensembles) is the current state of the art in terms of classification accuracy. HIVE-COTE recognizes that time series data are a specific data type for which the traditional attribute-value representation, used predominantly in machine learning, fails to provide a relevant representation. HIVE-COTE combines multiple types of classifiers: each extracting information about a specific aspect of a time series, be it in the time domain, frequency domain or summarization of intervals within the series. However, HIVE-COTE (and its predecessor, FLAT-COTE) is often infeasible to run on even modest amounts of data. For instance, training HIVE-COTE on a dataset with only 1500 time series can require 8 days of CPU time. It has polynomial runtime with respect to the training set size, so this problem compounds as data quantity increases. We propose a novel TSC algorithm, TS-CHIEF (Time Series Combination of Heterogeneous and Integrated Embedding Forest), which rivals HIVE-COTE in accuracy but requires only a fraction of the runtime. TS-CHIEF constructs an ensemble classifier that integrates the most effective embeddings of time series that research has developed in the last decade. It uses tree-structured classifiers to do so efficiently. We assess TS-CHIEF on 85 datasets of the University of California Riverside (UCR) archive, where it achieves state-of-the-art accuracy with scalability and efficiency. We demonstrate that TS-CHIEF can be trained on 130 k time series in 2 days, a data quantity that is beyond the reach of any TSC algorithm with comparable accuracy.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Time Series Classification (TSC) has seen enormous progress over the last two decades. HIVE-COTE (Hierarchical Vote Collective of Transformation-based Ensembles) is the current state of the art in terms of classification accuracy. HIVE-COTE recognizes that time series data are a specific data type for which the traditional attribute-value representation, used predominantly in machine learning, fails to provide a relevant representation. HIVE-COTE combines multiple types of classifiers: each extracting information about a specific aspect of a time series, be it in the time domain, frequency domain or summarization of intervals within the series. However, HIVE-COTE (and its predecessor, FLAT-COTE) is often infeasible to run on even modest amounts of data. For instance, training HIVE-COTE on a dataset with only 1500 time series can require 8 days of CPU time. It has polynomial runtime with respect to the training set size, so this problem compounds as data quantity increases. We propose a novel TSC algorithm, TS-CHIEF (Time Series Combination of Heterogeneous and Integrated Embedding Forest), which rivals HIVE-COTE in accuracy but requires only a fraction of the runtime. TS-CHIEF constructs an ensemble classifier that integrates the most effective embeddings of time series that research has developed in the last decade. It uses tree-structured classifiers to do so efficiently. We assess TS-CHIEF on 85 datasets of the University of California Riverside (UCR) archive, where it achieves state-of-the-art accuracy with scalability and efficiency. We demonstrate that TS-CHIEF can be trained on 130 k time series in 2 days, a data quantity that is beyond the reach of any TSC algorithm with comparable accuracy.
Tan, Chang Wei; Petitjean, François; Webb, Geoffrey I.
FastEE: Fast Ensembles of Elastic Distances for time series classification
Data Mining and Knowledge Discovery, vol. 34, no. 1, pp. 231-272, 2020, ISSN: 1573-756X.
@article{Tan2019,
title = {FastEE: Fast Ensembles of Elastic Distances for time series classification},
author = {Chang Wei Tan and Fran\c{c}ois Petitjean and Geoffrey I. Webb},
url = {https://rdcu.be/c1y5a},
doi = {10.1007/s10618-019-00663-x},
issn = {1573-756X},
year = {2020},
date = {2020-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {34},
number = {1},
pages = {231-272},
abstract = {In recent years, many new ensemble-based time series classification (TSC) algorithms have been proposed. Each of them is significantly more accurate than their predecessors. The Hierarchical Vote Collective of Transformation-based Ensembles (HIVE-COTE) is currently the most accurate TSC algorithm when assessed on the UCR repository. It is a meta-ensemble of 5 state-of-the-art ensemble-based classifiers. The time complexity of HIVE-COTE\textemdashparticularly for training\textemdashis prohibitive for most datasets. There is thus a critical need to speed up the classifiers that compose HIVE-COTE. This paper focuses on speeding up one of its components: Ensembles of Elastic Distances (EE), which is the classifier that leverages on the decades of research into the development of time-dedicated measures. Training EE can be prohibitive for many datasets. For example, it takes a month on the ElectricDevices dataset with 9000 instances. This is because EE needs to cross-validate the hyper-parameters used for the 11 similarity measures it encompasses. In this work, Fast Ensembles of Elastic Distances is proposed to train EE faster. There are two versions to it. The exact version makes it possible to train EE 10 times faster. The approximate version is 40 times faster than EE without significantly impacting the classification accuracy. This translates to being able to train EE on ElectricDevices in 13h.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In recent years, many new ensemble-based time series classification (TSC) algorithms have been proposed. Each of them is significantly more accurate than their predecessors. The Hierarchical Vote Collective of Transformation-based Ensembles (HIVE-COTE) is currently the most accurate TSC algorithm when assessed on the UCR repository. It is a meta-ensemble of 5 state-of-the-art ensemble-based classifiers. The time complexity of HIVE-COTE—particularly for training—is prohibitive for most datasets. There is thus a critical need to speed up the classifiers that compose HIVE-COTE. This paper focuses on speeding up one of its components: Ensembles of Elastic Distances (EE), which is the classifier that leverages on the decades of research into the development of time-dedicated measures. Training EE can be prohibitive for many datasets. For example, it takes a month on the ElectricDevices dataset with 9000 instances. This is because EE needs to cross-validate the hyper-parameters used for the 11 similarity measures it encompasses. In this work, Fast Ensembles of Elastic Distances is proposed to train EE faster. There are two versions to it. The exact version makes it possible to train EE 10 times faster. The approximate version is 40 times faster than EE without significantly impacting the classification accuracy. This translates to being able to train EE on ElectricDevices in 13h.
Dempster, Angus; Petitjean, Francois; Webb, Geoffrey I.
Data Mining and Knowledge Discovery, vol. 34, iss. 5, pp. 1454-1495, 2020.
@article{dempster2020rocket,
title = {ROCKET: Exceptionally fast and accurate time series classification using random convolutional kernels},
author = {Angus Dempster and Francois Petitjean and Geoffrey I. Webb},
url = {https://rdcu.be/c1zg4},
doi = {10.1007/s10618-020-00701-z},
year = {2020},
date = {2020-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {34},
issue = {5},
pages = {1454-1495},
abstract = {Most methods for time series classification that attain state-of-the-art accuracy have high computational complexity, requiring significant training time even for smaller datasets, and are intractable for larger datasets. Additionally, many existing methods focus on a single type of feature such as shape or frequency. Building on the recent success of convolutional neural networks for time series classification, we show that simple linear classifiers using random convolutional kernels achieve state-of-the-art accuracy with a fraction of the computational expense of existing methods. Using this method, it is possible to train and test a classifier on all 85 'bake off' datasets in the UCR archive in \<2h, and it is possible to train a classifier on a large dataset of more than one million time series in approximately 1 h.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Most methods for time series classification that attain state-of-the-art accuracy have high computational complexity, requiring significant training time even for smaller datasets, and are intractable for larger datasets. Additionally, many existing methods focus on a single type of feature such as shape or frequency. Building on the recent success of convolutional neural networks for time series classification, we show that simple linear classifiers using random convolutional kernels achieve state-of-the-art accuracy with a fraction of the computational expense of existing methods. Using this method, it is possible to train and test a classifier on all 85 'bake off' datasets in the UCR archive in <2h, and it is possible to train a classifier on a large dataset of more than one million time series in approximately 1 h.
Li, Fuyi; Leier, Andre; Liu, Quanzhong; Wang, Yanan; Xiang, Dongxu; Akutsu, Tatsuya; Webb, Geoffrey I.; Smith, A. Ian; Marquez-Lago, Tatiana; Li, Jian; Song, Jiangning
Genomics, Proteomics & Bioinformatics, vol. 18, no. 1, pp. 52-64, 2020, ISSN: 1672-0229.
@article{LI2020,
title = {Procleave: Predicting Protease-specific Substrate Cleavage Sites by Combining Sequence and Structural Information},
author = {Fuyi Li and Andre Leier and Quanzhong Liu and Yanan Wang and Dongxu Xiang and Tatsuya Akutsu and Geoffrey I. Webb and A. Ian Smith and Tatiana Marquez-Lago and Jian Li and Jiangning Song},
doi = {10.1016/j.gpb.2019.08.002},
issn = {1672-0229},
year = {2020},
date = {2020-01-01},
journal = {Genomics, Proteomics \& Bioinformatics},
volume = {18},
number = {1},
pages = {52-64},
abstract = {Proteases are enzymes that cleave and hydrolyse the peptide bonds between two specific amino acid residues of target substrate proteins. Protease-controlled proteolysis plays a key role in the degradation and recycling of proteins, which is essential for various physiological processes. Thus, solving the substrate identification problem will have important implications for the precise understanding of functions and physiological roles of proteases, as well as for therapeutic target identification and pharmaceutical applicability. Consequently, there is a great demand for bioinformatics methods that can predict novel substrate cleavage events with high accuracy by utilizing both sequence and structural information. In this study, we present Procleave, a novel bioinformatics approach for predicting protease-specific substrates and specific cleavage sites by taking into account both their sequence and 3D structural information. Structural features of known cleavage sites were represented by discrete values using a LOWESS data-smoothing optimization method, which turned out to be critical for the performance of Procleave. The optimal approximations of all structural parameter values were encoded in a conditional random field (CRF) computational framework, alongside sequence and chemical group-based features. Here, we demonstrate the outstanding performance of Procleave through extensive benchmarking and independent tests. Procleave is capable of correctly identifying most cleavage sites in the case study. Importantly, when applied to the human structural proteome encompassing 17,628 protein structures, Procleave suggests a number of potential novel target substrates and their corresponding cleavage sites of different proteases. Procleave is implemented as a webserver and is freely accessible at http://procleave.erc.monash.edu/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Proteases are enzymes that cleave and hydrolyse the peptide bonds between two specific amino acid residues of target substrate proteins. Protease-controlled proteolysis plays a key role in the degradation and recycling of proteins, which is essential for various physiological processes. Thus, solving the substrate identification problem will have important implications for the precise understanding of functions and physiological roles of proteases, as well as for therapeutic target identification and pharmaceutical applicability. Consequently, there is a great demand for bioinformatics methods that can predict novel substrate cleavage events with high accuracy by utilizing both sequence and structural information. In this study, we present Procleave, a novel bioinformatics approach for predicting protease-specific substrates and specific cleavage sites by taking into account both their sequence and 3D structural information. Structural features of known cleavage sites were represented by discrete values using a LOWESS data-smoothing optimization method, which turned out to be critical for the performance of Procleave. The optimal approximations of all structural parameter values were encoded in a conditional random field (CRF) computational framework, alongside sequence and chemical group-based features. Here, we demonstrate the outstanding performance of Procleave through extensive benchmarking and independent tests. Procleave is capable of correctly identifying most cleavage sites in the case study. Importantly, when applied to the human structural proteome encompassing 17,628 protein structures, Procleave suggests a number of potential novel target substrates and their corresponding cleavage sites of different proteases. Procleave is implemented as a webserver and is freely accessible at http://procleave.erc.monash.edu/.
Zaidi, N. A.; Du, Y.; Webb, G. I.
On the Effectiveness of Discretizing Quantitative Attributes in Linear Classifiers
IEEE Access, vol. 8, pp. 198856-198871, 2020.
@article{9245528,
title = {On the Effectiveness of Discretizing Quantitative Attributes in Linear Classifiers},
author = {N. A. Zaidi and Y. Du and G. I. Webb},
doi = {10.1109/ACCESS.2020.3034955},
year = {2020},
date = {2020-01-01},
journal = {IEEE Access},
volume = {8},
pages = {198856-198871},
abstract = {Rule ensembles are designed to provide a useful trade-off between predictive accuracy and model interpretability. However, the myopic and random search components of current rule ensemble methods can compromise this goal: they often need more rules than necessary to reach a certain accuracy level or can even outright fail to accurately model a distribution that can actually be described well with a few rules. Here, we present a novel approach aiming to fit rule ensembles of maximal predictive power for a given ensemble size (and thus model comprehensibility). In particular, we present an efficient branch-and-bound algorithm that optimally solves the per-rule objective function of the popular second-order gradient boosting framework. Our main insight is that the boosting objective can be tightly bounded in linear time of the number of covered data points. Along with an additional novel pruning technique related to rule redundancy, this leads to a computationally feasible approach for boosting optimal rules that, as we demonstrate on a wide range of common benchmark problems, consistently outperforms the predictive performance of boosting greedy rules.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Rule ensembles are designed to provide a useful trade-off between predictive accuracy and model interpretability. However, the myopic and random search components of current rule ensemble methods can compromise this goal: they often need more rules than necessary to reach a certain accuracy level or can even outright fail to accurately model a distribution that can actually be described well with a few rules. Here, we present a novel approach aiming to fit rule ensembles of maximal predictive power for a given ensemble size (and thus model comprehensibility). In particular, we present an efficient branch-and-bound algorithm that optimally solves the per-rule objective function of the popular second-order gradient boosting framework. Our main insight is that the boosting objective can be tightly bounded in linear time of the number of covered data points. Along with an additional novel pruning technique related to rule redundancy, this leads to a computationally feasible approach for boosting optimal rules that, as we demonstrate on a wide range of common benchmark problems, consistently outperforms the predictive performance of boosting greedy rules.
Webb, Geoffrey I; Zhang, Zhongfei; Tseng, Vincent S.; Williams, Graham; Vlachos, Michalis; Cao, Longbing (Ed.)
IEEE 7th International Conference on Data Science and Advanced Analytics (DSAA)
IEEE Computer Society, 2020.
@proceedings{DSAA2020,
title = {IEEE 7th International Conference on Data Science and Advanced Analytics (DSAA)},
editor = {Geoffrey I Webb and Zhongfei Zhang and Vincent S. Tseng and Graham Williams and Michalis Vlachos and Longbing Cao},
doi = {10.1109/DSAA49011.2020},
year = {2020},
date = {2020-01-01},
publisher = {IEEE Computer Society},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Nguyen, K.; Le, T.; Nguyen, T.; Webb, G.; Phung, D.
Robust Variational Learning for Multiclass Kernel Models with Stein Refinement
IEEE Transactions on Knowledge & Data Engineering, vol. 34, no. 9, pp. 4425-4438, 2020.
@article{nguyen2020,
title = {Robust Variational Learning for Multiclass Kernel Models with Stein Refinement},
author = {K. Nguyen and T. Le and T. Nguyen and G. Webb and D. Phung},
doi = {10.1109/TKDE.2020.3041509},
year = {2020},
date = {2020-01-01},
journal = {IEEE Transactions on Knowledge \& Data Engineering},
volume = {34},
number = {9},
pages = {4425-4438},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA, USA},
abstract = {Kernel-based models have a strong generalization ability, but most of them, including SVM, are vulnerable to the curse of kernelization. Moreover, their predictive performances are sensitive to the hyperparameters tuning, which highly demands computational resources. These problems render kernel methods problematic when dealing with large-scale datasets. To this end, we first formulate the optimization problem in a kernel-based learning setting as a posterior inference problem, and then develop a rich family of Recurrent Neural Network-based variational inference techniques. Unlike existing literature, which stops at the variational distribution and uses it as the surrogate for the true posterior distribution, here we further leverage Stein Variational Gradient Descent to further bring the variational distribution closer to the true posterior, we refer to this step as Stein Refinement. Putting these altogether, we arrive at a robust and efficient variational learning method for multiclass kernel machines with extremely accurate approximation. Moreover, our formulation enables efficient learning of kernel parameters and hyperparameters which robustifies the proposed method against data uncertainties. The extensive experimental results show that our method, without tuning any parameter, obtains comparable performance to LIBSVM, a well-known implementation of SVM, and outperforms other baselines while being able to seamlessly scale with large-scale datasets.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Kernel-based models have a strong generalization ability, but most of them, including SVM, are vulnerable to the curse of kernelization. Moreover, their predictive performances are sensitive to the hyperparameters tuning, which highly demands computational resources. These problems render kernel methods problematic when dealing with large-scale datasets. To this end, we first formulate the optimization problem in a kernel-based learning setting as a posterior inference problem, and then develop a rich family of Recurrent Neural Network-based variational inference techniques. Unlike existing literature, which stops at the variational distribution and uses it as the surrogate for the true posterior distribution, here we further leverage Stein Variational Gradient Descent to further bring the variational distribution closer to the true posterior, we refer to this step as Stein Refinement. Putting these altogether, we arrive at a robust and efficient variational learning method for multiclass kernel machines with extremely accurate approximation. Moreover, our formulation enables efficient learning of kernel parameters and hyperparameters which robustifies the proposed method against data uncertainties. The extensive experimental results show that our method, without tuning any parameter, obtains comparable performance to LIBSVM, a well-known implementation of SVM, and outperforms other baselines while being able to seamlessly scale with large-scale datasets.
Fischer, Raphael; Piatkowski, Nico; Pelletier, Charlotte; Webb, Geoffrey I.; Petitjean, Francois; Morik, Katharina
No Cloud on the Horizon: Probabilistic Gap Filling in Satellite Image Series
IEEE 7th International Conference on Data Science and Advanced Analytics (DSAA), pp. 546-555, IEEE, 2020.
@inproceedings{Fischer2020,
title = {No Cloud on the Horizon: Probabilistic Gap Filling in Satellite Image Series},
author = {Raphael Fischer and Nico Piatkowski and Charlotte Pelletier and Geoffrey I. Webb and Francois Petitjean and Katharina Morik},
doi = {10.1109/dsaa49011.2020.00069},
year = {2020},
date = {2020-01-01},
booktitle = {IEEE 7th International Conference on Data Science and Advanced Analytics (DSAA)},
pages = {546-555},
publisher = {IEEE},
abstract = {Spatio-temporal data sets such as satellite image series are of utmost importance for understanding global developments like climate change or urbanization. However, incompleteness of data can greatly impact usability and knowledge discovery. In fact, there are many cases where not a single data point in the set is fully observed. For filling gaps, we introduce a novel approach that utilizes Markov random fields (MRFs). We extend the probabilistic framework to also consider empirical prior information, which allows to train even on highly incomplete data. Moreover, we devise a way to make discrete MRFs predict continuous values via state superposition. Experiments on real-world remote sensing imagery suffering from cloud cover show that the proposed approach outperforms state-of-the-art gap filling techniques.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Spatio-temporal data sets such as satellite image series are of utmost importance for understanding global developments like climate change or urbanization. However, incompleteness of data can greatly impact usability and knowledge discovery. In fact, there are many cases where not a single data point in the set is fully observed. For filling gaps, we introduce a novel approach that utilizes Markov random fields (MRFs). We extend the probabilistic framework to also consider empirical prior information, which allows to train even on highly incomplete data. Moreover, we devise a way to make discrete MRFs predict continuous values via state superposition. Experiments on real-world remote sensing imagery suffering from cloud cover show that the proposed approach outperforms state-of-the-art gap filling techniques.
Chen, Shenglei; Webb, Geoffrey I.; Liu, Linyuan; Ma, Xin
A novel selective naive Bayes algorithm
Knowledge-Based Systems, vol. 192, 2020.
@article{CHEN2019105361,
title = {A novel selective naive Bayes algorithm},
author = {Shenglei Chen and Geoffrey I. Webb and Linyuan Liu and Xin Ma},
doi = {10.1016/j.knosys.2019.105361},
year = {2020},
date = {2020-01-01},
journal = {Knowledge-Based Systems},
volume = {192},
abstract = {Naive Bayes is one of the most popular data mining algorithms. Its efficiency comes from the assumption of attribute independence, although this might be violated in many real-world data sets. Many efforts have been done to mitigate the assumption, among which attribute selection is an important approach. However, conventional efforts to perform attribute selection in naive Bayes suffer from heavy computational overhead. This paper proposes an efficient selective naive Bayes algorithm, which adopts only some of the attributes to construct selective naive Bayes models. These models are built in such a way that each one is a trivial extension of another. The most predictive selective naive Bayes model can be selected by the measures of incremental leave-one-out cross validation. As a result, attributes can be selected by efficient model selection. Empirical results demonstrate that the selective naive Bayes shows superior classification accuracy, yet at the same time maintains the simplicity and efficiency.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Naive Bayes is one of the most popular data mining algorithms. Its efficiency comes from the assumption of attribute independence, although this might be violated in many real-world data sets. Many efforts have been done to mitigate the assumption, among which attribute selection is an important approach. However, conventional efforts to perform attribute selection in naive Bayes suffer from heavy computational overhead. This paper proposes an efficient selective naive Bayes algorithm, which adopts only some of the attributes to construct selective naive Bayes models. These models are built in such a way that each one is a trivial extension of another. The most predictive selective naive Bayes model can be selected by the measures of incremental leave-one-out cross validation. As a result, attributes can be selected by efficient model selection. Empirical results demonstrate that the selective naive Bayes shows superior classification accuracy, yet at the same time maintains the simplicity and efficiency.
Chen, Zhen; Zhao, Pei; Li, Fuyi; Marquez-Lago, Tatiana T; Leier, Andre; Revote, Jerico; Zhu, Yan; Powell, David R; Akutsu, Tatsuya; Webb, Geoffrey I; Chou, Kuo-Chen; Smith, A Ian; Daly, Roger J; Li, Jian; Song, Jiangning
Briefings in Bioinformatics, vol. 21, no. 3, pp. 1047-1057, 2020, ISSN: 1477-4054.
@article{10.1093/bib/bbz041,
title = {iLearn: an integrated platform and meta-learner for feature engineering, machine-learning analysis and modeling of DNA, RNA and protein sequence data},
author = {Zhen Chen and Pei Zhao and Fuyi Li and Tatiana T Marquez-Lago and Andre Leier and Jerico Revote and Yan Zhu and David R Powell and Tatsuya Akutsu and Geoffrey I Webb and Kuo-Chen Chou and A Ian Smith and Roger J Daly and Jian Li and Jiangning Song},
doi = {10.1093/bib/bbz041},
issn = {1477-4054},
year = {2020},
date = {2020-01-01},
journal = {Briefings in Bioinformatics},
volume = {21},
number = {3},
pages = {1047-1057},
abstract = {With the explosive growth of biological sequences generated in the post-genomic era, one of the most challenging problems in bioinformatics and computational biology is to computationally characterize sequences, structures and functions in an efficient, accurate and high-throughput manner. A number of online web servers and stand-alone tools have been developed to address this to date; however, all these tools have their limitations and drawbacks in terms of their effectiveness, user-friendliness and capacity. Here, we present iLearn, a comprehensive and versatile Python-based toolkit, integrating the functionality of feature extraction, clustering, normalization, selection, dimensionality reduction, predictor construction, best descriptor/model selection, ensemble learning and results visualization for DNA, RNA and protein sequences. iLearn was designed for users that only want to upload their data set and select the functions they need calculated from it, while all necessary procedures and optimal settings are completed automatically by the software. iLearn includes a variety of descriptors for DNA, RNA and proteins, and four feature output formats are supported so as to facilitate direct output usage or communication with other computational tools. In total, iLearn encompasses 16 different types of feature clustering, selection, normalization and dimensionality reduction algorithms, and five commonly used machine-learning algorithms, thereby greatly facilitating feature analysis and predictor construction. iLearn is made freely available via an online web server and a stand-alone toolkit.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
With the explosive growth of biological sequences generated in the post-genomic era, one of the most challenging problems in bioinformatics and computational biology is to computationally characterize sequences, structures and functions in an efficient, accurate and high-throughput manner. A number of online web servers and stand-alone tools have been developed to address this to date; however, all these tools have their limitations and drawbacks in terms of their effectiveness, user-friendliness and capacity. Here, we present iLearn, a comprehensive and versatile Python-based toolkit, integrating the functionality of feature extraction, clustering, normalization, selection, dimensionality reduction, predictor construction, best descriptor/model selection, ensemble learning and results visualization for DNA, RNA and protein sequences. iLearn was designed for users that only want to upload their data set and select the functions they need calculated from it, while all necessary procedures and optimal settings are completed automatically by the software. iLearn includes a variety of descriptors for DNA, RNA and proteins, and four feature output formats are supported so as to facilitate direct output usage or communication with other computational tools. In total, iLearn encompasses 16 different types of feature clustering, selection, normalization and dimensionality reduction algorithms, and five commonly used machine-learning algorithms, thereby greatly facilitating feature analysis and predictor construction. iLearn is made freely available via an online web server and a stand-alone toolkit.
Lucas, Benjamin; Pelletier, Charlotte; Schmidt, Daniel; Webb, Geoffrey I; Petitjean, François
Unsupervised Domain Adaptation Techniques for Classification of Satellite Image Time Series
IEEE International Geoscience and Remote Sensing Symposium, pp. 1074–1077, IEEE 2020.
@inproceedings{lucas2020unsupervised,
title = {Unsupervised Domain Adaptation Techniques for Classification of Satellite Image Time Series},
author = {Benjamin Lucas and Charlotte Pelletier and Daniel Schmidt and Geoffrey I Webb and Fran\c{c}ois Petitjean},
doi = {10.1109/IGARSS39084.2020.9324339},
year = {2020},
date = {2020-01-01},
booktitle = {IEEE International Geoscience and Remote Sensing Symposium},
pages = {1074\textendash1077},
organization = {IEEE},
abstract = {Land cover maps are vitally important to many elements of environmental management. However the machine learning algorithms used to produce them require a substantive quantity of labelled training data to reach the best levels of accuracy. When researchers wish to map an area where no labelled training data are available, one potential solution is to use a classifier trained on another geographical area and adapting it to the target location-this is known as Unsupervised Domain Adaptation (DA). In this paper we undertake the first experiments using unsupervised DA methods for the classification of satellite image time series (SITS) data. Our experiments draw the interesting conclusion that existing methods provide no benefit when used on SITS data, and that this is likely due to the temporal nature of the data and the change in class distributions between the regions. This suggests that an unsupervised domain adaptation technique for SITS would be extremely beneficial for land cover mapping.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Land cover maps are vitally important to many elements of environmental management. However the machine learning algorithms used to produce them require a substantive quantity of labelled training data to reach the best levels of accuracy. When researchers wish to map an area where no labelled training data are available, one potential solution is to use a classifier trained on another geographical area and adapting it to the target location-this is known as Unsupervised Domain Adaptation (DA). In this paper we undertake the first experiments using unsupervised DA methods for the classification of satellite image time series (SITS) data. Our experiments draw the interesting conclusion that existing methods provide no benefit when used on SITS data, and that this is likely due to the temporal nature of the data and the change in class distributions between the regions. This suggests that an unsupervised domain adaptation technique for SITS would be extremely beneficial for land cover mapping.
Chen, Zhen; Zhao, Pei; Li, Fuyi; Wang, Yanan; Smith, A Ian; Webb, Geoffrey I; Akutsu, Tatsuya; Baggag, Abdelkader; Bensmail, Halima; Song, Jiangning
Briefings in Bioinformatics, vol. 21, no. 5, pp. 1676-1696, 2020, ISSN: 1477-4054.
@article{10.1093/bib/bbz112,
title = {Comprehensive review and assessment of computational methods for predicting RNA post-transcriptional modification sites from RNA sequences},
author = {Zhen Chen and Pei Zhao and Fuyi Li and Yanan Wang and A Ian Smith and Geoffrey I Webb and Tatsuya Akutsu and Abdelkader Baggag and Halima Bensmail and Jiangning Song},
doi = {10.1093/bib/bbz112},
issn = {1477-4054},
year = {2020},
date = {2020-01-01},
journal = {Briefings in Bioinformatics},
volume = {21},
number = {5},
pages = {1676-1696},
abstract = {RNA post-transcriptional modifications play a crucial role in a myriad of biological processes and cellular functions. To date, more than 160 RNA modifications have been discovered; therefore, accurate identification of RNA-modification sites is fundamental for a better understanding of RNA-mediated biological functions and mechanisms. However, due to limitations in experimental methods, systematic identification of different types of RNA-modification sites remains a major challenge. Recently, more than 20 computational methods have been developed to identify RNA-modification sites in tandem with high-throughput experimental methods, with most of these capable of predicting only single types of RNA-modification sites. These methods show high diversity in their dataset size, data quality, core algorithms, features extracted and feature selection techniques and evaluation strategies. Therefore, there is an urgent need to revisit these methods and summarize their methodologies, in order to improve and further develop computational techniques to identify and characterize RNA-modification sites from the large amounts of sequence data. With this goal in mind, first, we provide a comprehensive survey on a large collection of 27 state-of-the-art approaches for predicting N1-methyladenosine and N6-methyladenosine sites. We cover a variety of important aspects that are crucial for the development of successful predictors, including the dataset quality, operating algorithms, sequence and genomic features, feature selection, model performance evaluation and software utility. In addition, we also provide our thoughts on potential strategies to improve the model performance. Second, we propose a computational approach called DeepPromise based on deep learning techniques for simultaneous prediction of N1-methyladenosine and N6-methyladenosine. To extract the sequence context surrounding the modification sites, three feature encodings, including enhanced nucleic acid composition, one-hot encoding, and RNA embedding, were used as the input to seven consecutive layers of convolutional neural networks (CNNs), respectively. Moreover, DeepPromise further combined the prediction score of the CNN-based models and achieved around 43% higher area under receiver-operating curve (AUROC) for m1A site prediction and 6% higher AUROC for m6A site prediction, respectively, when compared with several existing state-of-the-art approaches on the independent test. In-depth analyses of characteristic sequence motifs identified from the convolution-layer filters indicated that nucleotide presentation at proximal positions surrounding the modification sites contributed most to the classification, whereas those at distal positions also affected classification but to different extents. To maximize user convenience, a web server was developed as an implementation of DeepPromise and made publicly available at http://DeepPromise.erc.monash.edu/, with the server accepting both RNA sequences and genomic sequences to allow prediction of two types of putative RNA-modification sites.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
RNA post-transcriptional modifications play a crucial role in a myriad of biological processes and cellular functions. To date, more than 160 RNA modifications have been discovered; therefore, accurate identification of RNA-modification sites is fundamental for a better understanding of RNA-mediated biological functions and mechanisms. However, due to limitations in experimental methods, systematic identification of different types of RNA-modification sites remains a major challenge. Recently, more than 20 computational methods have been developed to identify RNA-modification sites in tandem with high-throughput experimental methods, with most of these capable of predicting only single types of RNA-modification sites. These methods show high diversity in their dataset size, data quality, core algorithms, features extracted and feature selection techniques and evaluation strategies. Therefore, there is an urgent need to revisit these methods and summarize their methodologies, in order to improve and further develop computational techniques to identify and characterize RNA-modification sites from the large amounts of sequence data. With this goal in mind, first, we provide a comprehensive survey on a large collection of 27 state-of-the-art approaches for predicting N1-methyladenosine and N6-methyladenosine sites. We cover a variety of important aspects that are crucial for the development of successful predictors, including the dataset quality, operating algorithms, sequence and genomic features, feature selection, model performance evaluation and software utility. In addition, we also provide our thoughts on potential strategies to improve the model performance. Second, we propose a computational approach called DeepPromise based on deep learning techniques for simultaneous prediction of N1-methyladenosine and N6-methyladenosine. To extract the sequence context surrounding the modification sites, three feature encodings, including enhanced nucleic acid composition, one-hot encoding, and RNA embedding, were used as the input to seven consecutive layers of convolutional neural networks (CNNs), respectively. Moreover, DeepPromise further combined the prediction score of the CNN-based models and achieved around 43% higher area under receiver-operating curve (AUROC) for m1A site prediction and 6% higher AUROC for m6A site prediction, respectively, when compared with several existing state-of-the-art approaches on the independent test. In-depth analyses of characteristic sequence motifs identified from the convolution-layer filters indicated that nucleotide presentation at proximal positions surrounding the modification sites contributed most to the classification, whereas those at distal positions also affected classification but to different extents. To maximize user convenience, a web server was developed as an implementation of DeepPromise and made publicly available at http://DeepPromise.erc.monash.edu/, with the server accepting both RNA sequences and genomic sequences to allow prediction of two types of putative RNA-modification sites.
Miller, Lynn; Bolton, Mitzi; Boulton, Julie; Mintrom, Michael; Nicholson, Ann; Rüdiger, Christoph; Skinner, Rob; Raven, Rob; Webb, Geoffrey I
IEEE/ITU International Conference on Artificial Intelligence for Good (AI4G), pp. 180-185, IEEE 2020.
@inproceedings{miller2020ai,
title = {AI for monitoring the Sustainable Development Goals and supporting and promoting action and policy development},
author = {Lynn Miller and Mitzi Bolton and Julie Boulton and Michael Mintrom and Ann Nicholson and Christoph R\"{u}diger and Rob Skinner and Rob Raven and Geoffrey I Webb},
doi = {10.1109/AI4G50087.2020.9311014},
year = {2020},
date = {2020-01-01},
booktitle = {IEEE/ITU International Conference on Artificial Intelligence for Good (AI4G)},
pages = {180-185},
organization = {IEEE},
abstract = {The United Nations sustainable development goals (SDGs) were ratified with much enthusiasm by all UN member states in 2015. However, subsequent progress to meet these goals has been hampered by a lack of data available to measure the SDG indicators (SDIs), and a lack of evidence-based insights to inform effective policy responses. We outline an interdisciplinary program of research into the use of artificial intelligence techniques to support measurement of the SDIs, using both machine learning methods to model SDI measurements and explainable AI techniques to present the outputs in a human-friendly manner. As well as addressing the technical concerns, we will investigate the governance issues of what forms of evidence, methods of collecting that evidence and means of its communication will most usefully inform effective policy development. By addressing these fundamental challenges, we aim to provide policy makers with the evidence needed to take effective action towards realising the Sustainable Development Goals.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The United Nations sustainable development goals (SDGs) were ratified with much enthusiasm by all UN member states in 2015. However, subsequent progress to meet these goals has been hampered by a lack of data available to measure the SDG indicators (SDIs), and a lack of evidence-based insights to inform effective policy responses. We outline an interdisciplinary program of research into the use of artificial intelligence techniques to support measurement of the SDIs, using both machine learning methods to model SDI measurements and explainable AI techniques to present the outputs in a human-friendly manner. As well as addressing the technical concerns, we will investigate the governance issues of what forms of evidence, methods of collecting that evidence and means of its communication will most usefully inform effective policy development. By addressing these fundamental challenges, we aim to provide policy makers with the evidence needed to take effective action towards realising the Sustainable Development Goals.
Fawaz, Hassan Ismail; Lucas, Benjamin; Forestier, Germain; Pelletier, Charlotte; Schmidt, Daniel F.; Weber, Jonathan; Webb, Geoffrey I.; Idoumghar, Lhassane; Muller, Pierre-Alain; Petitjean, Francois
InceptionTime: Finding AlexNet for Time Series Classification
Data Mining and Knowledge Discovery, vol. 34, iss. 6, pp. 1936-1962, 2020.
@article{fawaz2019inceptiontime,
title = {InceptionTime: Finding AlexNet for Time Series Classification},
author = {Hassan Ismail Fawaz and Benjamin Lucas and Germain Forestier and Charlotte Pelletier and Daniel F. Schmidt and Jonathan Weber and Geoffrey I. Webb and Lhassane Idoumghar and Pierre-Alain Muller and Francois Petitjean},
url = {https://rdcu.be/b6TXh},
doi = {10.1007/s10618-020-00710-y},
year = {2020},
date = {2020-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {34},
issue = {6},
pages = {1936-1962},
abstract = {This paper brings deep learning at the forefront of research into time series classification (TSC). TSC is the area of machine learning tasked with the categorization (or labelling) of time series. The last few decades of work in this area have led to significant progress in the accuracy of classifiers, with the state of the art now represented by the HIVE-COTE algorithm. While extremely accurate, HIVE-COTE cannot be applied to many real-world datasets because of its high training time complexity in O(N^2 . T^4) for a dataset with N time series of length T. For example, it takes HIVE-COTE more than 8 days to learn from a small dataset with N = 1500 time series of short length T = 46. Meanwhile deep learning has received enormous attention because of its high accuracy and scalability. Recent approaches to deep learning for TSC have been scalable, but less accurate than HIVE-COTE. We introduce InceptionTime - an ensemble of deep Convolutional Neural Network models, inspired by the Inception-v4 architecture. Our experiments show that InceptionTime is on par with HIVE-COTE in terms of accuracy while being much more scalable: not only can it learn from 1500 time series in one hour but it can also learn from 8M time series in 13 h, a quantity of data that is fully out of reach of HIVE-COTE.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper brings deep learning at the forefront of research into time series classification (TSC). TSC is the area of machine learning tasked with the categorization (or labelling) of time series. The last few decades of work in this area have led to significant progress in the accuracy of classifiers, with the state of the art now represented by the HIVE-COTE algorithm. While extremely accurate, HIVE-COTE cannot be applied to many real-world datasets because of its high training time complexity in O(N^2 . T^4) for a dataset with N time series of length T. For example, it takes HIVE-COTE more than 8 days to learn from a small dataset with N = 1500 time series of short length T = 46. Meanwhile deep learning has received enormous attention because of its high accuracy and scalability. Recent approaches to deep learning for TSC have been scalable, but less accurate than HIVE-COTE. We introduce InceptionTime - an ensemble of deep Convolutional Neural Network models, inspired by the Inception-v4 architecture. Our experiments show that InceptionTime is on par with HIVE-COTE in terms of accuracy while being much more scalable: not only can it learn from 1500 time series in one hour but it can also learn from 8M time series in 13 h, a quantity of data that is fully out of reach of HIVE-COTE.
Goldenberg, Igor; Webb, Geoffrey I.
PCA-based drift and shift quantification framework for multidimensional data
Knowledge and Information Systems, vol. 62, pp. 2835-2854, 2020.
@article{Goldenberg2020,
title = {PCA-based drift and shift quantification framework for multidimensional data},
author = {Igor Goldenberg and Geoffrey I. Webb},
doi = {10.1007/s10115-020-01438-3},
year = {2020},
date = {2020-01-01},
journal = {Knowledge and Information Systems},
volume = {62},
pages = {2835-2854},
abstract = {Concept drift is a serious problem confronting machine learning systems in a dynamic and ever-changing world. In order to manage concept drift it may be useful to first quantify it by measuring the distance between distributions that generate data before and after a drift. There is a paucity of methods to do so in the case of multidimensional numeric data. This paper provides an in-depth analysis of the PCA-based change detection approach, identifies shortcomings of existing methods and shows how this approach can be used to measure a drift, not merely detect it.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Concept drift is a serious problem confronting machine learning systems in a dynamic and ever-changing world. In order to manage concept drift it may be useful to first quantify it by measuring the distance between distributions that generate data before and after a drift. There is a paucity of methods to do so in the case of multidimensional numeric data. This paper provides an in-depth analysis of the PCA-based change detection approach, identifies shortcomings of existing methods and shows how this approach can be used to measure a drift, not merely detect it.
Phung, Dinh; Webb, Geoffrey I.; Sammut, Claude (Ed.)
Encyclopedia of Machine Learning and Data Science
Springer US, 2020.
@book{Phung2020,
title = {Encyclopedia of Machine Learning and Data Science},
editor = {Dinh Phung and Geoffrey I. Webb and Claude Sammut},
doi = {10.1007/978-1-4899-7502-7},
year = {2020},
date = {2020-01-01},
publisher = {Springer US},
keywords = {},
pubstate = {published},
tppubtype = {book}
}
Li, Fuyi; Chen, Jinxiang; Leier, Andre; Marquez-Lago, Tatiana; Liu, Quanzhong; Wang, Yanze; Revote, Jerico; Smith, A Ian; Akutsu, Tatsuya; Webb, Geoffrey I; Kurgan, Lukasz; Song, Jiangning
Bioinformatics, vol. 36, no. 4, pp. 1057-1065, 2020, ISSN: 1367-4803.
@article{Li2020a,
title = {DeepCleave: a deep learning predictor for caspase and matrix metalloprotease substrates and cleavage sites},
author = {Fuyi Li and Jinxiang Chen and Andre Leier and Tatiana Marquez-Lago and Quanzhong Liu and Yanze Wang and Jerico Revote and A Ian Smith and Tatsuya Akutsu and Geoffrey I Webb and Lukasz Kurgan and Jiangning Song},
doi = {10.1093/bioinformatics/btz721},
issn = {1367-4803},
year = {2020},
date = {2020-01-01},
journal = {Bioinformatics},
volume = {36},
number = {4},
pages = {1057-1065},
abstract = {Proteases are enzymes that cleave target substrate proteins by catalyzing the hydrolysis of peptide bonds between specific amino acids. While the functional proteolysis regulated by proteases plays a central role in the "life and death" process of proteins, many of the corresponding substrates and their cleavage sites were not found yet. Availability of accurate predictors of the substrates and cleavage sites would facilitate understanding of proteases' functions and physiological roles. Deep learning is a promising approach for the development of accurate predictors of substrate cleavage events.We propose DeepCleave, the first deep learning-based predictor of protease-specific substrates and cleavage sites. DeepCleave uses protein substrate sequence data as input and employs convolutional neural networks with transfer learning to train accurate predictive models. High predictive performance of our models stems from the use of high-quality cleavage site features extracted from the substrate sequences through the deep learning process, and the application of transfer learning, multiple kernels and attention layer in the design of the deep network. Empirical tests against several related state-of-the-art methods demonstrate that DeepCleave outperforms these methods in predicting caspase and matrix metalloprotease substrate-cleavage sites.The DeepCleave webserver and source code are freely available at http://deepcleave.erc.monash.edu/.Supplementary data are available at Bioinformatics online.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Proteases are enzymes that cleave target substrate proteins by catalyzing the hydrolysis of peptide bonds between specific amino acids. While the functional proteolysis regulated by proteases plays a central role in the "life and death" process of proteins, many of the corresponding substrates and their cleavage sites were not found yet. Availability of accurate predictors of the substrates and cleavage sites would facilitate understanding of proteases' functions and physiological roles. Deep learning is a promising approach for the development of accurate predictors of substrate cleavage events.We propose DeepCleave, the first deep learning-based predictor of protease-specific substrates and cleavage sites. DeepCleave uses protein substrate sequence data as input and employs convolutional neural networks with transfer learning to train accurate predictive models. High predictive performance of our models stems from the use of high-quality cleavage site features extracted from the substrate sequences through the deep learning process, and the application of transfer learning, multiple kernels and attention layer in the design of the deep network. Empirical tests against several related state-of-the-art methods demonstrate that DeepCleave outperforms these methods in predicting caspase and matrix metalloprotease substrate-cleavage sites.The DeepCleave webserver and source code are freely available at http://deepcleave.erc.monash.edu/.Supplementary data are available at Bioinformatics online.
Li, Fuyi; Fan, Cunshuo; Marquez-Lago, Tatiana T; Leier, Andre; Revote, Jerico; Jia, Cangzhi; Zhu, Yan; Smith, A Ian; Webb, Geoffrey I; Liu, Quanzhong; Wei, Leyi; Li, Jian; Song, Jiangning
Briefings in Bioinformatics, vol. 21, no. 3, pp. 1069-1079, 2020, ISSN: 1477-4054.
@article{10.1093/bib/bbz050,
title = {PRISMOID: a comprehensive 3D structure database for post-translational modifications and mutations with functional impact},
author = {Fuyi Li and Cunshuo Fan and Tatiana T Marquez-Lago and Andre Leier and Jerico Revote and Cangzhi Jia and Yan Zhu and A Ian Smith and Geoffrey I Webb and Quanzhong Liu and Leyi Wei and Jian Li and Jiangning Song},
doi = {10.1093/bib/bbz050},
issn = {1477-4054},
year = {2020},
date = {2020-01-01},
journal = {Briefings in Bioinformatics},
volume = {21},
number = {3},
pages = {1069-1079},
abstract = {Post-translational modifications (PTMs) play very important roles in various cell signaling pathways and biological process. Due to PTMs' extremely important roles, many major PTMs have been studied, while the functional and mechanical characterization of major PTMs is well documented in several databases. However, most currently available databases mainly focus on protein sequences, while the real 3D structures of PTMs have been largely ignored. Therefore, studies of PTMs 3D structural signatures have been severely limited by the deficiency of the data. Here, we develop PRISMOID, a novel publicly available and free 3D structure database for a wide range of PTMs. PRISMOID represents an up-to-date and interactive online knowledge base with specific focus on 3D structural contexts of PTMs sites and mutations that occur on PTMs and in the close proximity of PTM sites with functional impact. The first version of PRISMOID encompasses 17145 non-redundant modification sites on 3919 related protein 3D structure entries pertaining to 37 different types of PTMs. Our entry web page is organized in a comprehensive manner, including detailed PTM annotation on the 3D structure and biological information in terms of mutations affecting PTMs, secondary structure features and per-residue solvent accessibility features of PTM sites, domain context, predicted natively disordered regions and sequence alignments. In addition, high-definition JavaScript packages are employed to enhance information visualization in PRISMOID. PRISMOID equips a variety of interactive and customizable search options and data browsing functions; these capabilities allow users to access data via keyword, ID and advanced options combination search in an efficient and user-friendly way. A download page is also provided to enable users to download the SQL file, computational structural features and PTM sites' data. We anticipate PRISMOID will swiftly become an invaluable online resource, assisting both biologists and bioinformaticians to conduct experiments and develop applications supporting discovery efforts in the sequence-structural-functional relationship of PTMs and providing important insight into mutations and PTM sites interaction mechanisms. The PRISMOID database is freely accessible at http://prismoid.erc.monash.edu/. The database and web interface are implemented in MySQL, JSP, JavaScript and HTML with all major browsers supported.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Post-translational modifications (PTMs) play very important roles in various cell signaling pathways and biological process. Due to PTMs' extremely important roles, many major PTMs have been studied, while the functional and mechanical characterization of major PTMs is well documented in several databases. However, most currently available databases mainly focus on protein sequences, while the real 3D structures of PTMs have been largely ignored. Therefore, studies of PTMs 3D structural signatures have been severely limited by the deficiency of the data. Here, we develop PRISMOID, a novel publicly available and free 3D structure database for a wide range of PTMs. PRISMOID represents an up-to-date and interactive online knowledge base with specific focus on 3D structural contexts of PTMs sites and mutations that occur on PTMs and in the close proximity of PTM sites with functional impact. The first version of PRISMOID encompasses 17145 non-redundant modification sites on 3919 related protein 3D structure entries pertaining to 37 different types of PTMs. Our entry web page is organized in a comprehensive manner, including detailed PTM annotation on the 3D structure and biological information in terms of mutations affecting PTMs, secondary structure features and per-residue solvent accessibility features of PTM sites, domain context, predicted natively disordered regions and sequence alignments. In addition, high-definition JavaScript packages are employed to enhance information visualization in PRISMOID. PRISMOID equips a variety of interactive and customizable search options and data browsing functions; these capabilities allow users to access data via keyword, ID and advanced options combination search in an efficient and user-friendly way. A download page is also provided to enable users to download the SQL file, computational structural features and PTM sites' data. We anticipate PRISMOID will swiftly become an invaluable online resource, assisting both biologists and bioinformaticians to conduct experiments and develop applications supporting discovery efforts in the sequence-structural-functional relationship of PTMs and providing important insight into mutations and PTM sites interaction mechanisms. The PRISMOID database is freely accessible at http://prismoid.erc.monash.edu/. The database and web interface are implemented in MySQL, JSP, JavaScript and HTML with all major browsers supported.
Wang, Xiaochuan; Li, Chen; Li, Fuyi; Sharma, Varun S.; Song, Jiangning; Webb, Geoffrey I.
BMC Bioinformatics, vol. 20, no. 1, 2019.
@article{Wang2019,
title = {SIMLIN: a bioinformatics tool for prediction of S-sulphenylation in the human proteome based on multi-stage ensemble-learning models},
author = {Xiaochuan Wang and Chen Li and Fuyi Li and Varun S. Sharma and Jiangning Song and Geoffrey I. Webb},
doi = {10.1186/s12859-019-3178-6},
year = {2019},
date = {2019-11-01},
journal = {BMC Bioinformatics},
volume = {20},
number = {1},
abstract = {S-sulphenylation is a ubiquitous protein post-translational modification (PTM) where an S-hydroxyl bond is formed via the reversible oxidation on the Sulfhydryl group of cysteine (C). Recent experimental studies have revealed that S-sulphenylation plays critical roles in many biological functions, such as protein regulation and cell signaling. State-of-the-art bioinformatic advances have facilitated high-throughput in silico screening of protein S-sulphenylation sites, thereby significantly reducing the time and labour costs traditionally required for the experimental investigation of S-sulphenylation.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
S-sulphenylation is a ubiquitous protein post-translational modification (PTM) where an S-hydroxyl bond is formed via the reversible oxidation on the Sulfhydryl group of cysteine (C). Recent experimental studies have revealed that S-sulphenylation plays critical roles in many biological functions, such as protein regulation and cell signaling. State-of-the-art bioinformatic advances have facilitated high-throughput in silico screening of protein S-sulphenylation sites, thereby significantly reducing the time and labour costs traditionally required for the experimental investigation of S-sulphenylation.
Pelletier, Charlotte; Webb, Geoffrey I.; Petitjean, Francois
Deep Learning for the Classification of Sentinel-2 Image Series
IEEE International Geoscience And Remote Sensing Symposium, 2019.
@inproceedings{PelletierEtAl19b,
title = {Deep Learning for the Classification of Sentinel-2 Image Series},
author = {Charlotte Pelletier and Geoffrey I. Webb and Francois Petitjean},
doi = {10.1109/IGARSS.2019.8900123},
year = {2019},
date = {2019-07-01},
booktitle = {IEEE International Geoscience And Remote Sensing Symposium},
abstract = {Satellite image time series (SITS) have proven to be essential for accurate and up-to-date land cover mapping over large areas. Most works about SITS have focused on the use of traditional classification algorithms such as Random Forests (RFs). Deep learning algorithms have been very successful for supervised tasks, in particular for data that exhibit a structure between attributes, such as space or time. In this work, we compare for the first time RFs to the two leading deep learning algorithms for handling temporal data: Recurrent Neural Networks (RNNs) and temporal Convolutional Neural Networks (TempCNNs). We carry out a large experiment using Sentinel-2 time series. We compare both accuracy and computational times to classify 10,980 km 2 over Australia. The results highlights the good performance of TemCNNs that obtain the highest accuracy. They also show that RNNs might be less suited for large scale study as they have higher runtime complexity.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Satellite image time series (SITS) have proven to be essential for accurate and up-to-date land cover mapping over large areas. Most works about SITS have focused on the use of traditional classification algorithms such as Random Forests (RFs). Deep learning algorithms have been very successful for supervised tasks, in particular for data that exhibit a structure between attributes, such as space or time. In this work, we compare for the first time RFs to the two leading deep learning algorithms for handling temporal data: Recurrent Neural Networks (RNNs) and temporal Convolutional Neural Networks (TempCNNs). We carry out a large experiment using Sentinel-2 time series. We compare both accuracy and computational times to classify 10,980 km 2 over Australia. The results highlights the good performance of TemCNNs that obtain the highest accuracy. They also show that RNNs might be less suited for large scale study as they have higher runtime complexity.
Li, Fuyi; Zhang, Yang; Purcell, Anthony W.; Webb, Geoffrey I.; Chou, Kuo-Chen; Lithgow, Trevor; Li, Chen; Song, Jiangning
Positive-unlabelled learning of glycosylation sites in the human proteome
BMC Bioinformatics, vol. 20, no. 1, pp. 112, 2019, ISSN: 1471-2105.
@article{Li2019,
title = {Positive-unlabelled learning of glycosylation sites in the human proteome},
author = {Fuyi Li and Yang Zhang and Anthony W. Purcell and Geoffrey I. Webb and Kuo-Chen Chou and Trevor Lithgow and Chen Li and Jiangning Song},
url = {https://rdcu.be/bpQBV},
doi = {10.1186/s12859-019-2700-1},
issn = {1471-2105},
year = {2019},
date = {2019-03-01},
journal = {BMC Bioinformatics},
volume = {20},
number = {1},
pages = {112},
abstract = {As an important type of post-translational modification (PTM), protein glycosylation plays a crucial role in protein stability and protein function. The abundance and ubiquity of protein glycosylation across three domains of life involving Eukarya, Bacteria and Archaea demonstrate its roles in regulating a variety of signalling and metabolic pathways. Mutations on and in the proximity of glycosylation sites are highly associated with human diseases. Accordingly, accurate prediction of glycosylation can complement laboratory-based methods and greatly benefit experimental efforts for characterization and understanding of functional roles of glycosylation. For this purpose, a number of supervised-learning approaches have been proposed to identify glycosylation sites, demonstrating a promising predictive performance. To train a conventional supervised-learning model, both reliable positive and negative samples are required. However, in practice, a large portion of negative samples (i.e. non-glycosylation sites) are mislabelled due to the limitation of current experimental technologies. Moreover, supervised algorithms often fail to take advantage of large volumes of unlabelled data, which can aid in model learning in conjunction with positive samples (i.e. experimentally verified glycosylation sites).},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
As an important type of post-translational modification (PTM), protein glycosylation plays a crucial role in protein stability and protein function. The abundance and ubiquity of protein glycosylation across three domains of life involving Eukarya, Bacteria and Archaea demonstrate its roles in regulating a variety of signalling and metabolic pathways. Mutations on and in the proximity of glycosylation sites are highly associated with human diseases. Accordingly, accurate prediction of glycosylation can complement laboratory-based methods and greatly benefit experimental efforts for characterization and understanding of functional roles of glycosylation. For this purpose, a number of supervised-learning approaches have been proposed to identify glycosylation sites, demonstrating a promising predictive performance. To train a conventional supervised-learning model, both reliable positive and negative samples are required. However, in practice, a large portion of negative samples (i.e. non-glycosylation sites) are mislabelled due to the limitation of current experimental technologies. Moreover, supervised algorithms often fail to take advantage of large volumes of unlabelled data, which can aid in model learning in conjunction with positive samples (i.e. experimentally verified glycosylation sites).
Hamalainen, Wilhelmiina; Webb, Geoffrey I.
A tutorial on statistically sound pattern discovery
Data Mining and Knowledge Discovery, vol. 33, no. 2, pp. 325-377, 2019, ISSN: 1573-756X.
@article{HamalainenWebb2019,
title = {A tutorial on statistically sound pattern discovery},
author = {Wilhelmiina Hamalainen and Geoffrey I. Webb},
url = {https://rdcu.be/bd2MI},
doi = {10.1007/s10618-018-0590-x},
issn = {1573-756X},
year = {2019},
date = {2019-03-01},
journal = {Data Mining and Knowledge Discovery},
volume = {33},
number = {2},
pages = {325-377},
abstract = {Statistically sound pattern discovery harnesses the rigour of statistical hypothesis testing to overcome many of the issues that have hampered standard data mining approaches to pattern discovery. Most importantly, application of appropriate statistical tests allows precise control over the risk of false discoveries\textemdashpatterns that are found in the sample data but do not hold in the wider population from which the sample was drawn. Statistical tests can also be applied to filter out patterns that are unlikely to be useful, removing uninformative variations of the key patterns in the data. This tutorial introduces the key statistical and data mining theory and techniques that underpin this fast developing field. We concentrate on two general classes of patterns: dependency rules that express statistical dependencies between condition and consequent parts and dependency sets that express mutual dependence between set elements. We clarify alternative interpretations of statistical dependence and introduce appropriate tests for evaluating statistical significance of patterns in different situations. We also introduce special techniques for controlling the likelihood of spurious discoveries when multitudes of patterns are evaluated. The paper is aimed at a wide variety of audiences. It provides the necessary statistical background and summary of the state-of-the-art for any data mining researcher or practitioner wishing to enter or understand statistically sound pattern discovery research or practice. It can serve as a general introduction to the field of statistically sound pattern discovery for any reader with a general background in data sciences.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Statistically sound pattern discovery harnesses the rigour of statistical hypothesis testing to overcome many of the issues that have hampered standard data mining approaches to pattern discovery. Most importantly, application of appropriate statistical tests allows precise control over the risk of false discoveries—patterns that are found in the sample data but do not hold in the wider population from which the sample was drawn. Statistical tests can also be applied to filter out patterns that are unlikely to be useful, removing uninformative variations of the key patterns in the data. This tutorial introduces the key statistical and data mining theory and techniques that underpin this fast developing field. We concentrate on two general classes of patterns: dependency rules that express statistical dependencies between condition and consequent parts and dependency sets that express mutual dependence between set elements. We clarify alternative interpretations of statistical dependence and introduce appropriate tests for evaluating statistical significance of patterns in different situations. We also introduce special techniques for controlling the likelihood of spurious discoveries when multitudes of patterns are evaluated. The paper is aimed at a wide variety of audiences. It provides the necessary statistical background and summary of the state-of-the-art for any data mining researcher or practitioner wishing to enter or understand statistically sound pattern discovery research or practice. It can serve as a general introduction to the field of statistically sound pattern discovery for any reader with a general background in data sciences.
Li, Fuyi; Wang, Yanan; Li, Chen; Marquez-Lago, Tatiana T; Leier, Andre; Rawlings, Neil D; Haffari, Gholamreza; Revote, Jerico; Akutsu, Tatsuya; Chou, Kuo-Chen; Purcell, Anthony W; Pike, Robert N; Webb, Geoffrey I; Smith, Ian A; Lithgow, Trevor; Daly, Roger J; Whisstock, James C; Song, Jiangning
Briefings in Bioinformatics, vol. 20, no. 6, pp. 2150-2166, 2019.
@article{Li18b,
title = {Twenty years of bioinformatics research for protease-specific substrate and cleavage site prediction: a comprehensive revisit and benchmarking of existing methods},
author = {Fuyi Li and Yanan Wang and Chen Li and Tatiana T Marquez-Lago and Andre Leier and Neil D Rawlings and Gholamreza Haffari and Jerico Revote and Tatsuya Akutsu and Kuo-Chen Chou and Anthony W Purcell and Robert N Pike and Geoffrey I Webb and Ian A Smith and Trevor Lithgow and Roger J Daly and James C Whisstock and Jiangning Song},
doi = {10.1093/bib/bby077},
year = {2019},
date = {2019-01-01},
journal = {Briefings in Bioinformatics},
volume = {20},
number = {6},
pages = {2150-2166},
abstract = {The roles of proteolytic cleavage have been intensively investigated and discussed during the past two decades. This irreversible chemical process has been frequently reported to influence a number of crucial biological processes (BPs), such as cell cycle, protein regulation and inflammation. A number of advanced studies have been published aiming at deciphering the mechanisms of proteolytic cleavage. Given its significance and the large number of functionally enriched substrates targeted by specific proteases, many computational approaches have been established for accurate prediction of protease-specific substrates and their cleavage sites. Consequently, there is an urgent need to systematically assess the state-of-the-art computational approaches for protease-specific cleavage site prediction to further advance the existing methodologies and to improve the prediction performance. With this goal in mind, in this article, we carefully evaluated a total of 19 computational methods (including 8 scoring function-based methods and 11 machine learning-based methods) in terms of their underlying algorithm, calculated features, performance evaluation and software usability. Then, extensive independent tests were performed to assess the robustness and scalability of the reviewed methods using our carefully prepared independent test data sets with 3641 cleavage sites (specific to 10 proteases). The comparative experimental results demonstrate that PROSPERous is the most accurate generic method for predicting eight protease-specific cleavage sites, while GPS-CCD and LabCaS outperformed other predictors for calpain-specific cleavage sites. Based on our review, we then outlined some potential ways to improve the prediction performance and ease the computational burden by applying ensemble learning, deep learning, positive unlabeled learning and parallel and distributed computing techniques. We anticipate that our study will serve as a practical and useful guide for interested readers to further advance next-generation bioinformatics tools for protease-specific cleavage site prediction.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The roles of proteolytic cleavage have been intensively investigated and discussed during the past two decades. This irreversible chemical process has been frequently reported to influence a number of crucial biological processes (BPs), such as cell cycle, protein regulation and inflammation. A number of advanced studies have been published aiming at deciphering the mechanisms of proteolytic cleavage. Given its significance and the large number of functionally enriched substrates targeted by specific proteases, many computational approaches have been established for accurate prediction of protease-specific substrates and their cleavage sites. Consequently, there is an urgent need to systematically assess the state-of-the-art computational approaches for protease-specific cleavage site prediction to further advance the existing methodologies and to improve the prediction performance. With this goal in mind, in this article, we carefully evaluated a total of 19 computational methods (including 8 scoring function-based methods and 11 machine learning-based methods) in terms of their underlying algorithm, calculated features, performance evaluation and software usability. Then, extensive independent tests were performed to assess the robustness and scalability of the reviewed methods using our carefully prepared independent test data sets with 3641 cleavage sites (specific to 10 proteases). The comparative experimental results demonstrate that PROSPERous is the most accurate generic method for predicting eight protease-specific cleavage sites, while GPS-CCD and LabCaS outperformed other predictors for calpain-specific cleavage sites. Based on our review, we then outlined some potential ways to improve the prediction performance and ease the computational burden by applying ensemble learning, deep learning, positive unlabeled learning and parallel and distributed computing techniques. We anticipate that our study will serve as a practical and useful guide for interested readers to further advance next-generation bioinformatics tools for protease-specific cleavage site prediction.
Lucas, B.; Pelletier, C.; Inglada, J.; Schmidt, D.; Webb, G. I.; Petitjean, F
Proceedings 10th International Workshop on the Analysis of Multitemporal Remote Sensing Images, MultiTemp 2019, IEEE, Institute of Electrical and Electronics Engineers, 2019.
@inproceedings{LucasEtAl2019b,
title = {Exploring Data Quantity Requirements for Domain Adaptation in the Classification of Satellite Image Time Series},
author = {B. Lucas and C. Pelletier and J. Inglada and D. Schmidt and G. I. Webb and F Petitjean},
doi = {10.1109/Multi-Temp.2019.8866898},
year = {2019},
date = {2019-01-01},
booktitle = {Proceedings 10th International Workshop on the Analysis of Multitemporal Remote Sensing Images, MultiTemp 2019},
publisher = {IEEE, Institute of Electrical and Electronics Engineers},
abstract = {Land cover maps are a vital input variable in all types of environmental research and management. However the modern state-of-The-Art machine learning techniques used to create them require substantial training data to produce optimal accuracy. Domain Adaptation is one technique researchers might use when labelled training data are unavailable or scarce. This paper looks at the result of training a convolutional neural network model on a region where data are available (source domain), and then adapting this model to another region (target domain) by retraining it on the available labelled data, and in particular how these results change with increasing data availability. Our experiments performing domain adaptation on satellite image time series, draw three interesting conclusions: (1) a model trained only on data from the source domain delivers 73.0% test accuracy on the target domain; (2) when all of the weights are retrained on the target data, over 16,000 instances were required to improve upon the accuracy of the source-only model; and (3) even if sufficient data is available in the target domain, using a model pretrained on a source domain will result in better overall test accuracy compared to a model trained on target domain data only-88.9% versus 84.7%.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Land cover maps are a vital input variable in all types of environmental research and management. However the modern state-of-The-Art machine learning techniques used to create them require substantial training data to produce optimal accuracy. Domain Adaptation is one technique researchers might use when labelled training data are unavailable or scarce. This paper looks at the result of training a convolutional neural network model on a region where data are available (source domain), and then adapting this model to another region (target domain) by retraining it on the available labelled data, and in particular how these results change with increasing data availability. Our experiments performing domain adaptation on satellite image time series, draw three interesting conclusions: (1) a model trained only on data from the source domain delivers 73.0% test accuracy on the target domain; (2) when all of the weights are retrained on the target data, over 16,000 instances were required to improve upon the accuracy of the source-only model; and (3) even if sufficient data is available in the target domain, using a model pretrained on a source domain will result in better overall test accuracy compared to a model trained on target domain data only-88.9% versus 84.7%.
Lucas, Benjamin; Shifaz, Ahmed; Pelletier, Charlotte; O'Neill, Lachlan; Zaidi, Nayyar; Goethals, Bart; Petitjean, Francois; Webb, Geoffrey I.
Proximity Forest: an effective and scalable distance-based classifier for time series
Data Mining and Knowledge Discovery, vol. 33, pp. 607-635, 2019, ISSN: 1573-756X.
@article{LucasEtAl2019,
title = {Proximity Forest: an effective and scalable distance-based classifier for time series},
author = {Benjamin Lucas and Ahmed Shifaz and Charlotte Pelletier and Lachlan O'Neill and Nayyar Zaidi and Bart Goethals and Francois Petitjean and Geoffrey I. Webb},
url = {https://rdcu.be/blB8E},
doi = {10.1007/s10618-019-00617-3},
issn = {1573-756X},
year = {2019},
date = {2019-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {33},
pages = {607-635},
abstract = {Research into the classification of time series has made enormous progress in the last decade. The UCR time series archive has played a significant role in challenging and guiding the development of new learners for time series classification. The largest dataset in the UCR archive holds 10,000 time series only; which may explain why the primary research focus has been on creating algorithms that have high accuracy on relatively small datasets. This paper introduces Proximity Forest, an algorithm that learns accurate models from datasets with millions of time series, and classifies a time series in milliseconds. The models are ensembles of highly randomized Proximity Trees. Whereas conventional decision trees branch on attribute values (and usually perform poorly on time series), Proximity Trees branch on the proximity of time series to one exemplar time series or another; allowing us to leverage the decades of work into developing relevant measures for time series. Proximity Forest gains both efficiency and accuracy by stochastic selection of both exemplars and similarity measures. Our work is motivated by recent time series applications that provide orders of magnitude more time series than the UCR benchmarks. Our experiments demonstrate that Proximity Forest is highly competitive on the UCR archive: it ranks among the most accurate classifiers while being significantly faster. We demonstrate on a 1M time series Earth observation dataset that Proximity Forest retains this accuracy on datasets that are many orders of magnitude greater than those in the UCR repository, while learning its models at least 100,000 times faster than current state-of-the-art models Elastic Ensemble and COTE.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Research into the classification of time series has made enormous progress in the last decade. The UCR time series archive has played a significant role in challenging and guiding the development of new learners for time series classification. The largest dataset in the UCR archive holds 10,000 time series only; which may explain why the primary research focus has been on creating algorithms that have high accuracy on relatively small datasets. This paper introduces Proximity Forest, an algorithm that learns accurate models from datasets with millions of time series, and classifies a time series in milliseconds. The models are ensembles of highly randomized Proximity Trees. Whereas conventional decision trees branch on attribute values (and usually perform poorly on time series), Proximity Trees branch on the proximity of time series to one exemplar time series or another; allowing us to leverage the decades of work into developing relevant measures for time series. Proximity Forest gains both efficiency and accuracy by stochastic selection of both exemplars and similarity measures. Our work is motivated by recent time series applications that provide orders of magnitude more time series than the UCR benchmarks. Our experiments demonstrate that Proximity Forest is highly competitive on the UCR archive: it ranks among the most accurate classifiers while being significantly faster. We demonstrate on a 1M time series Earth observation dataset that Proximity Forest retains this accuracy on datasets that are many orders of magnitude greater than those in the UCR repository, while learning its models at least 100,000 times faster than current state-of-the-art models Elastic Ensemble and COTE.
Pelletier, C.; Ji, Z.; Hagolle, O.; Morse-McNabb, E.; Sheffield, K.; Webb, G. I.; Petitjean, F.
Using Sentinel-2 Image Time Series to map the State of Victoria, Australia
Proceedings 10th International Workshop on the Analysis of Multitemporal Remote Sensing Images, MultiTemp 2019, 2019.
@inproceedings{PelletierEtAl19c,
title = {Using Sentinel-2 Image Time Series to map the State of Victoria, Australia},
author = {C. Pelletier and Z. Ji and O. Hagolle and E. Morse-McNabb and K. Sheffield and G. I. Webb and F. Petitjean},
doi = {10.1109/Multi-Temp.2019.8866921},
year = {2019},
date = {2019-01-01},
booktitle = {Proceedings 10th International Workshop on the Analysis of Multitemporal Remote Sensing Images, MultiTemp 2019},
abstract = {Sentinel-2 satellites are now acquiring images of the entire Earth every five days from 10 to 60 m spatial resolution. The supervised classification of this new optical image time series allows the operational production of accurate land cover maps over large areas. In this paper, we investigate the use of one year of Sentinel-2 data to map the state of Victoria in Australia. In particular, we produce two land cover maps using the most established and advanced algorithms in time series classification: Random Forest (RF) and Temporal Convolutional Neural Network (TempCNN). To our knowledge, these are the first land cover maps at 10 m spatial resolution for an Australian state.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Sentinel-2 satellites are now acquiring images of the entire Earth every five days from 10 to 60 m spatial resolution. The supervised classification of this new optical image time series allows the operational production of accurate land cover maps over large areas. In this paper, we investigate the use of one year of Sentinel-2 data to map the state of Victoria in Australia. In particular, we produce two land cover maps using the most established and advanced algorithms in time series classification: Random Forest (RF) and Temporal Convolutional Neural Network (TempCNN). To our knowledge, these are the first land cover maps at 10 m spatial resolution for an Australian state.
Pelletier, Charlotte; Webb, Geoffrey I.; Petitjean, Francois
Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series
Remote Sensing, vol. 11, no. 5, 2019, ISSN: 2072-4292.
@article{PelletierEtAl19,
title = {Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series},
author = {Charlotte Pelletier and Geoffrey I. Webb and Francois Petitjean},
doi = {10.3390/rs11050523},
issn = {2072-4292},
year = {2019},
date = {2019-01-01},
journal = {Remote Sensing},
volume = {11},
number = {5},
abstract = {Latest remote sensing sensors are capable of acquiring high spatial and spectral Satellite Image Time Series (SITS) of the world. These image series are a key component of classification systems that aim at obtaining up-to-date and accurate land cover maps of the Earth\’s surfaces. More specifically, current SITS combine high temporal, spectral and spatial resolutions, which makes it possible to closely monitor vegetation dynamics. Although traditional classification algorithms, such as Random Forest (RF), have been successfully applied to create land cover maps from SITS, these algorithms do not make the most of the temporal domain. This paper proposes a comprehensive study of Temporal Convolutional Neural Networks (TempCNNs), a deep learning approach which applies convolutions in the temporal dimension in order to automatically learn temporal (and spectral) features. The goal of this paper is to quantitatively and qualitatively evaluate the contribution of TempCNNs for SITS classification, as compared to RF and Recurrent Neural Networks (RNNs) \—a standard deep learning approach that is particularly suited to temporal data. We carry out experiments on Formosat-2 scene with 46 images and one million labelled time series. The experimental results show that TempCNNs are more accurate than the current state of the art for SITS classification. We provide some general guidelines on the network architecture, common regularization mechanisms, and hyper-parameter values such as batch size; we also draw out some differences with standard results in computer vision (e.g., about pooling layers). Finally, we assess the visual quality of the land cover maps produced by TempCNNs.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Latest remote sensing sensors are capable of acquiring high spatial and spectral Satellite Image Time Series (SITS) of the world. These image series are a key component of classification systems that aim at obtaining up-to-date and accurate land cover maps of the Earth’s surfaces. More specifically, current SITS combine high temporal, spectral and spatial resolutions, which makes it possible to closely monitor vegetation dynamics. Although traditional classification algorithms, such as Random Forest (RF), have been successfully applied to create land cover maps from SITS, these algorithms do not make the most of the temporal domain. This paper proposes a comprehensive study of Temporal Convolutional Neural Networks (TempCNNs), a deep learning approach which applies convolutions in the temporal dimension in order to automatically learn temporal (and spectral) features. The goal of this paper is to quantitatively and qualitatively evaluate the contribution of TempCNNs for SITS classification, as compared to RF and Recurrent Neural Networks (RNNs) —a standard deep learning approach that is particularly suited to temporal data. We carry out experiments on Formosat-2 scene with 46 images and one million labelled time series. The experimental results show that TempCNNs are more accurate than the current state of the art for SITS classification. We provide some general guidelines on the network architecture, common regularization mechanisms, and hyper-parameter values such as batch size; we also draw out some differences with standard results in computer vision (e.g., about pooling layers). Finally, we assess the visual quality of the land cover maps produced by TempCNNs.
Song, Jiangning; Wang, Yanan; Li, Fuyi; Akutsu, Tatsuya; Rawlings, Neil D; Webb, Geoffrey I; Chou, Kuo-Chen
Briefings in Bioinformatics, vol. 20, no. 2, pp. 638-658, 2019.
@article{doi:10.1093/bib/bby028,
title = {iProt-Sub: a comprehensive package for accurately mapping and predicting protease-specific substrates and cleavage sites},
author = {Jiangning Song and Yanan Wang and Fuyi Li and Tatsuya Akutsu and Neil D Rawlings and Geoffrey I Webb and Kuo-Chen Chou},
doi = {10.1093/bib/bby028},
year = {2019},
date = {2019-01-01},
journal = {Briefings in Bioinformatics},
volume = {20},
number = {2},
pages = {638-658},
abstract = {Regulation of proteolysis plays a critical role in a myriad of important cellular processes. The key to better understanding the mechanisms that control this process is to identify the specific substrates that each protease targets. To address this, we have developed iProt-Sub, a powerful bioinformatics tool for the accurate prediction of protease-specific substrates and their cleavage sites. Importantly, iProt-Sub represents a significantly advanced version of its successful predecessor, PROSPER. It provides optimized cleavage site prediction models with better prediction performance and coverage for more species-specific proteases (4 major protease families and 38 different proteases). iProt-Sub integrates heterogeneous sequence and structural features and uses a two-step feature selection procedure to further remove redundant and irrelevant features in an effort to improve the cleavage site prediction accuracy. Features used by iProt-Sub are encoded by 11 different sequence encoding schemes, including local amino acid sequence profile, secondary structure, solvent accessibility and native disorder, which will allow a more accurate representation of the protease specificity of approximately 38 proteases and training of the prediction models. Benchmarking experiments using cross-validation and independent tests showed that iProt-Sub is able to achieve a better performance than several existing generic tools. We anticipate that iProt-Sub will be a powerful tool for proteome-wide prediction of protease-specific substrates and their cleavage sites, and will facilitate hypothesis-driven functional interrogation of protease-specific substrate cleavage and proteolytic events.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Regulation of proteolysis plays a critical role in a myriad of important cellular processes. The key to better understanding the mechanisms that control this process is to identify the specific substrates that each protease targets. To address this, we have developed iProt-Sub, a powerful bioinformatics tool for the accurate prediction of protease-specific substrates and their cleavage sites. Importantly, iProt-Sub represents a significantly advanced version of its successful predecessor, PROSPER. It provides optimized cleavage site prediction models with better prediction performance and coverage for more species-specific proteases (4 major protease families and 38 different proteases). iProt-Sub integrates heterogeneous sequence and structural features and uses a two-step feature selection procedure to further remove redundant and irrelevant features in an effort to improve the cleavage site prediction accuracy. Features used by iProt-Sub are encoded by 11 different sequence encoding schemes, including local amino acid sequence profile, secondary structure, solvent accessibility and native disorder, which will allow a more accurate representation of the protease specificity of approximately 38 proteases and training of the prediction models. Benchmarking experiments using cross-validation and independent tests showed that iProt-Sub is able to achieve a better performance than several existing generic tools. We anticipate that iProt-Sub will be a powerful tool for proteome-wide prediction of protease-specific substrates and their cleavage sites, and will facilitate hypothesis-driven functional interrogation of protease-specific substrate cleavage and proteolytic events.
Tan, Chang Wei; Petitjean, Francois; Webb, Geoffrey I.
Elastic bands across the path: A new framework and methods to lower bound DTW
Proceedings of the 2019 SIAM International Conference on Data Mining, pp. 522-530, 2019.
@inproceedings{TanEtAl19,
title = {Elastic bands across the path: A new framework and methods to lower bound DTW},
author = {Chang Wei Tan and Francois Petitjean and Geoffrey I. Webb},
url = {https://arxiv.org/abs/1808.09617},
year = {2019},
date = {2019-01-01},
booktitle = {Proceedings of the 2019 SIAM International Conference on Data Mining},
pages = {522-530},
abstract = {There has been renewed recent interest in developing effective lower bounds for Dynamic Time Warping (DTW) distance between time series. These have many applications in time series indexing, clustering, forecasting, regression and classification. One of the key time series classification algorithms, the nearest neighbor algorithm with DTW distance (NN-DTW) is very expensive to compute, due to the quadratic complexity of DTW. Lower bound search can speed up NN-DTW substantially. An effective and tight lower bound quickly prunes off unpromising nearest neighbor candidates from the search space and minimises the number of the costly DTW computations. The speed up provided by lower bound search becomes increasingly critical as training set size increases. Different lower bounds provide different trade-offs between computation time and tightness. Most existing lower bounds interact with DTW warping window sizes. They are very tight and effective at smaller warping window sizes, but become looser as the warping window increases, thus reducing the pruning effectiveness for NN-DTW. In this work, we present a new class of lower bounds that are tighter than the popular Keogh lower bound, while requiring similar computation time. Our new lower bounds take advantage of the DTW boundary condition, monotonicity and continuity constraints to create a tighter lower bound. Of particular significance, they remain relatively tight even for large windows. A single parameter to these new lower bounds controls the speed-tightness trade-off. We demonstrate that these new lower bounds provide an exceptional balance between computation time and tightness for the NN-DTW time series classification task, resulting in greatly improved efficiency for NN-DTW lower bound search.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
There has been renewed recent interest in developing effective lower bounds for Dynamic Time Warping (DTW) distance between time series. These have many applications in time series indexing, clustering, forecasting, regression and classification. One of the key time series classification algorithms, the nearest neighbor algorithm with DTW distance (NN-DTW) is very expensive to compute, due to the quadratic complexity of DTW. Lower bound search can speed up NN-DTW substantially. An effective and tight lower bound quickly prunes off unpromising nearest neighbor candidates from the search space and minimises the number of the costly DTW computations. The speed up provided by lower bound search becomes increasingly critical as training set size increases. Different lower bounds provide different trade-offs between computation time and tightness. Most existing lower bounds interact with DTW warping window sizes. They are very tight and effective at smaller warping window sizes, but become looser as the warping window increases, thus reducing the pruning effectiveness for NN-DTW. In this work, we present a new class of lower bounds that are tighter than the popular Keogh lower bound, while requiring similar computation time. Our new lower bounds take advantage of the DTW boundary condition, monotonicity and continuity constraints to create a tighter lower bound. Of particular significance, they remain relatively tight even for large windows. A single parameter to these new lower bounds controls the speed-tightness trade-off. We demonstrate that these new lower bounds provide an exceptional balance between computation time and tightness for the NN-DTW time series classification task, resulting in greatly improved efficiency for NN-DTW lower bound search.
Wang, Jiawei; Yang, Bingjiao; An, Yi; Marquez-Lago, Tatiana; Leier, Andre; Wilksch, Jonathan; Hong, Qingyang; Zhang, Yang; Hayashida, Morihiro; Akutsu, Tatsuya; Webb, Geoffrey I; Strugnell, Richard A; Song, Jiangning; Lithgow, Trevor
Briefings in Bioinformatics, vol. 20, no. 3, pp. 931-951, 2019.
@article{doi:10.1093/bib/bbx164,
title = {Systematic analysis and prediction of type IV secreted effector proteins by machine learning approaches},
author = {Jiawei Wang and Bingjiao Yang and Yi An and Tatiana Marquez-Lago and Andre Leier and Jonathan Wilksch and Qingyang Hong and Yang Zhang and Morihiro Hayashida and Tatsuya Akutsu and Geoffrey I Webb and Richard A Strugnell and Jiangning Song and Trevor Lithgow},
doi = {10.1093/bib/bbx164},
year = {2019},
date = {2019-01-01},
journal = {Briefings in Bioinformatics},
volume = {20},
number = {3},
pages = {931-951},
abstract = {In the course of infecting their hosts, pathogenic bacteria secrete numerous effectors, namely, bacterial proteins that pervert host cell biology. Many Gram-negative bacteria, including context-dependent human pathogens, use a type IV secretion system (T4SS) to translocate effectors directly into the cytosol of host cells. Various type IV secreted effectors (T4SEs) have been experimentally validated to play crucial roles in virulence by manipulating host cell gene expression and other processes. Consequently, the identification of novel effector proteins is an important step in increasing our understanding of host-pathogen interactions and bacterial pathogenesis. Here, we train and compare six machine learning models, namely, Naive Bayes (NB), K-nearest neighbor (KNN), logistic regression (LR), random forest (RF), support vector machines (SVMs) and multilayer perceptron (MLP), for the identification of T4SEs using 10 types of selected features and 5-fold cross-validation. Our study shows that: (1) including different but complementary features generally enhance the predictive performance of T4SEs; (2) ensemble models, obtained by integrating individual single-feature models, exhibit a significantly improved predictive performance and (3) the 'majority voting strategy' led to a more stable and accurate classification performance when applied to predicting an ensemble learning model with distinct single features. We further developed a new method to effectively predict T4SEs, Bastion4 (Bacterial secretion effector predictor for T4SS), and we show our ensemble classifier clearly outperforms two recent prediction tools. In summary, we developed a state-of-the-art T4SE predictor by conducting a comprehensive performance evaluation of different machine learning algorithms along with a detailed analysis of single- and multi-feature selections.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In the course of infecting their hosts, pathogenic bacteria secrete numerous effectors, namely, bacterial proteins that pervert host cell biology. Many Gram-negative bacteria, including context-dependent human pathogens, use a type IV secretion system (T4SS) to translocate effectors directly into the cytosol of host cells. Various type IV secreted effectors (T4SEs) have been experimentally validated to play crucial roles in virulence by manipulating host cell gene expression and other processes. Consequently, the identification of novel effector proteins is an important step in increasing our understanding of host-pathogen interactions and bacterial pathogenesis. Here, we train and compare six machine learning models, namely, Naive Bayes (NB), K-nearest neighbor (KNN), logistic regression (LR), random forest (RF), support vector machines (SVMs) and multilayer perceptron (MLP), for the identification of T4SEs using 10 types of selected features and 5-fold cross-validation. Our study shows that: (1) including different but complementary features generally enhance the predictive performance of T4SEs; (2) ensemble models, obtained by integrating individual single-feature models, exhibit a significantly improved predictive performance and (3) the 'majority voting strategy' led to a more stable and accurate classification performance when applied to predicting an ensemble learning model with distinct single features. We further developed a new method to effectively predict T4SEs, Bastion4 (Bacterial secretion effector predictor for T4SS), and we show our ensemble classifier clearly outperforms two recent prediction tools. In summary, we developed a state-of-the-art T4SE predictor by conducting a comprehensive performance evaluation of different machine learning algorithms along with a detailed analysis of single- and multi-feature selections.
Yu, Hualong; Webb, Geoffrey I
Adaptive Online Extreme Learning Machine by Regulating Forgetting Factor by Concept Drift Map
Neurocomputing, vol. 343, pp. 141-153, 2019.
@article{yuetal2019,
title = {Adaptive Online Extreme Learning Machine by Regulating Forgetting Factor by Concept Drift Map},
author = {Hualong Yu and Geoffrey I Webb},
doi = {10.1016/j.neucom.2018.11.098},
year = {2019},
date = {2019-01-01},
journal = {Neurocomputing},
volume = {343},
pages = {141-153},
publisher = {Elsevier},
abstract = {In online-learning, the data is incrementally received and the distributions from which it is drawn may keep changing over time. This phenomenon is widely known as concept drift. Such changes may affect the generalization of a learned model to future data. This problem may be exacerbated by the form of the drift itself changing over time. Quantitative measures to describe and analyze the concept drift have been proposed in previous work. A description composed from these measures is called a concept drift map. We believe that these maps could be useful for guiding how much knowledge in the old model should be forgotten. Therefore, this paper presents an adaptive online learning model that uses a concept drift map to regulate the forgetting factor of an extreme learning machine. Specifically, when a batch of new instances are labeled, the distribution of each class on each attribute is firstly estimated, and then it is compared with the distribution estimated in the previous batch to calculate the magnitude of concept drift, which is further used to regulate the forgetting factor and to update the learning model. Therefore, the novelty of this paper lies in that a quantitative distance metric between two distributions constructed on continuous attribute space is presented to construct concept drift map which can be further associated with the forgetting factor to make the learning model adapt the concept drift. Experimental results on several benchmark stream data sets show the proposed model is generally superior to several previous algorithms when classifying a variety of data streams subject to drift, indicating its effectiveness and feasibility.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In online-learning, the data is incrementally received and the distributions from which it is drawn may keep changing over time. This phenomenon is widely known as concept drift. Such changes may affect the generalization of a learned model to future data. This problem may be exacerbated by the form of the drift itself changing over time. Quantitative measures to describe and analyze the concept drift have been proposed in previous work. A description composed from these measures is called a concept drift map. We believe that these maps could be useful for guiding how much knowledge in the old model should be forgotten. Therefore, this paper presents an adaptive online learning model that uses a concept drift map to regulate the forgetting factor of an extreme learning machine. Specifically, when a batch of new instances are labeled, the distribution of each class on each attribute is firstly estimated, and then it is compared with the distribution estimated in the previous batch to calculate the magnitude of concept drift, which is further used to regulate the forgetting factor and to update the learning model. Therefore, the novelty of this paper lies in that a quantitative distance metric between two distributions constructed on continuous attribute space is presented to construct concept drift map which can be further associated with the forgetting factor to make the learning model adapt the concept drift. Experimental results on several benchmark stream data sets show the proposed model is generally superior to several previous algorithms when classifying a variety of data streams subject to drift, indicating its effectiveness and feasibility.
Zhang, Yanju; Xie, Ruopeng; Wang, Jiawei; Leier, Andre; Marquez-Lago, Tatiana T.; Akutsu, Tatsuya; Webb, Geoffrey I.; Chou, Kuo-Chen; Song, Jiangning
Briefings in Bioinformatics, vol. 20, no. 6, pp. 2185-2199, 2019.
@article{ZhangEtAl18,
title = {Computational analysis and prediction of lysine malonylation sites by exploiting informative features in an integrative machine-learning framework},
author = {Yanju Zhang and Ruopeng Xie and Jiawei Wang and Andre Leier and Tatiana T. Marquez-Lago and Tatsuya Akutsu and Geoffrey I. Webb and Kuo-Chen Chou and Jiangning Song},
doi = {10.1093/bib/bby079},
year = {2019},
date = {2019-01-01},
journal = {Briefings in Bioinformatics},
volume = {20},
number = {6},
pages = {2185-2199},
abstract = {As a newly discovered post-translational modification (PTM), lysine malonylation (Kmal) regulates a myriad of cellular processes from prokaryotes to eukaryotes and has important implications in human diseases. Despite its functional significance, computational methods to accurately identify malonylation sites are still lacking and urgently needed. In particular, there is currently no comprehensive analysis and assessment of different features and machine learning (ML) methods that are required for constructing the necessary prediction models. Here, we review, analyze and compare 11 different feature encoding methods, with the goal of extracting key patterns and characteristics from residue sequences of Kmal sites. We identify optimized feature sets, with which four commonly used ML methods (random forest, support vector machines, K-nearest neighbor and logistic regression) and one recently proposed [Light Gradient Boosting Machine (LightGBM)] are trained on data from three species, namely, Escherichia coli, Mus musculus and Homo sapiens, and compared using randomized 10-fold cross-validation tests. We show that integration of the single method-based models through ensemble learning further improves the prediction performance and model robustness on the independent test. When compared to the existing state-of-the-art predictor, MaloPred, the optimal ensemble models were more accurate for all three species (AUC: 0.930, 0.923 and 0.944 for E. coli, M. musculus and H. sapiens, respectively). Using the ensemble models, we developed an accessible online predictor, kmal-sp, available at http://kmalsp.erc.monash.edu/. We hope that this comprehensive survey and the proposed strategy for building more accurate models can serve as a useful guide for inspiring future developments of computational methods for PTM site prediction, expedite the discovery of new malonylation and other PTM types and facilitate hypothesis-driven experimental validation of novel malonylated substrates and malonylation sites.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
As a newly discovered post-translational modification (PTM), lysine malonylation (Kmal) regulates a myriad of cellular processes from prokaryotes to eukaryotes and has important implications in human diseases. Despite its functional significance, computational methods to accurately identify malonylation sites are still lacking and urgently needed. In particular, there is currently no comprehensive analysis and assessment of different features and machine learning (ML) methods that are required for constructing the necessary prediction models. Here, we review, analyze and compare 11 different feature encoding methods, with the goal of extracting key patterns and characteristics from residue sequences of Kmal sites. We identify optimized feature sets, with which four commonly used ML methods (random forest, support vector machines, K-nearest neighbor and logistic regression) and one recently proposed [Light Gradient Boosting Machine (LightGBM)] are trained on data from three species, namely, Escherichia coli, Mus musculus and Homo sapiens, and compared using randomized 10-fold cross-validation tests. We show that integration of the single method-based models through ensemble learning further improves the prediction performance and model robustness on the independent test. When compared to the existing state-of-the-art predictor, MaloPred, the optimal ensemble models were more accurate for all three species (AUC: 0.930, 0.923 and 0.944 for E. coli, M. musculus and H. sapiens, respectively). Using the ensemble models, we developed an accessible online predictor, kmal-sp, available at http://kmalsp.erc.monash.edu/. We hope that this comprehensive survey and the proposed strategy for building more accurate models can serve as a useful guide for inspiring future developments of computational methods for PTM site prediction, expedite the discovery of new malonylation and other PTM types and facilitate hypothesis-driven experimental validation of novel malonylated substrates and malonylation sites.
Chen, Zhen; Li, Lei; Xu, Dakang; Chou, Kuo-Chen; Liu, Xuhan; Smith, Alexander Ian; Li, Fuyi; Song, Jiangning; Li, Chen; Leier, Andre; Marquez-Lago, Tatiana; Akutsu, Tatsuya; Webb, Geoffrey I
Briefings in Bioinformatics, vol. 20, no. 6, pp. 2267-2290, 2019.
@article{ChenEtAl118b,
title = {Large-scale comparative assessment of computational predictors for lysine post-translational modification sites},
author = {Zhen Chen and Lei Li and Dakang Xu and Kuo-Chen Chou and Xuhan Liu and Alexander Ian Smith and Fuyi Li and Jiangning Song and Chen Li and Andre Leier and Tatiana Marquez-Lago and Tatsuya Akutsu and Geoffrey I Webb},
doi = {10.1093/bib/bby089},
year = {2019},
date = {2019-01-01},
journal = {Briefings in Bioinformatics},
volume = {20},
number = {6},
pages = {2267-2290},
abstract = {Lysine post-translational modifications (PTMs) play a crucial role in regulating diverse functions and biological processes of proteins. However, because of the large volumes of sequencing data generated from genome-sequencing projects, systematic identification of different types of lysine PTM substrates and PTM sites in the entire proteome remains a major challenge. In recent years, a number of computational methods for lysine PTM identification have been developed. These methods show high diversity in their core algorithms, features extracted and feature selection techniques and evaluation strategies. There is therefore an urgent need to revisit these methods and summarize their methodologies, to improve and further develop computational techniques to identify and characterize lysine PTMs from the large amounts of sequence data. With this goal in mind, we first provide a comprehensive survey on a large collection of 49 state-of-the-art approaches for lysine PTM prediction. We cover a variety of important aspects that are crucial for the development of successful predictors, including operating algorithms, sequence and structural features, feature selection, model performance evaluation and software utility. We further provide our thoughts on potential strategies to improve the model performance. Second, in order to examine the feasibility of using deep learning for lysine PTM prediction, we propose a novel computational framework, termed MUscADEL (Multiple Scalable Accurate Deep Learner for lysine PTMs), using deep, bidirectional, long short-term memory recurrent neural networks for accurate and systematic mapping of eight major types of lysine PTMs in the human and mouse proteomes. Extensive benchmarking tests show that MUscADEL outperforms current methods for lysine PTM characterization, demonstrating the potential and power of deep learning techniques in protein PTM prediction. The web server of MUscADEL, together with all the data sets assembled in this study, is freely available at http://muscadel.erc.monash.edu/. We anticipate this comprehensive review and the application of deep learning will provide practical guide and useful insights into PTM prediction and inspire future bioinformatics studies in the related fields.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Lysine post-translational modifications (PTMs) play a crucial role in regulating diverse functions and biological processes of proteins. However, because of the large volumes of sequencing data generated from genome-sequencing projects, systematic identification of different types of lysine PTM substrates and PTM sites in the entire proteome remains a major challenge. In recent years, a number of computational methods for lysine PTM identification have been developed. These methods show high diversity in their core algorithms, features extracted and feature selection techniques and evaluation strategies. There is therefore an urgent need to revisit these methods and summarize their methodologies, to improve and further develop computational techniques to identify and characterize lysine PTMs from the large amounts of sequence data. With this goal in mind, we first provide a comprehensive survey on a large collection of 49 state-of-the-art approaches for lysine PTM prediction. We cover a variety of important aspects that are crucial for the development of successful predictors, including operating algorithms, sequence and structural features, feature selection, model performance evaluation and software utility. We further provide our thoughts on potential strategies to improve the model performance. Second, in order to examine the feasibility of using deep learning for lysine PTM prediction, we propose a novel computational framework, termed MUscADEL (Multiple Scalable Accurate Deep Learner for lysine PTMs), using deep, bidirectional, long short-term memory recurrent neural networks for accurate and systematic mapping of eight major types of lysine PTMs in the human and mouse proteomes. Extensive benchmarking tests show that MUscADEL outperforms current methods for lysine PTM characterization, demonstrating the potential and power of deep learning techniques in protein PTM prediction. The web server of MUscADEL, together with all the data sets assembled in this study, is freely available at http://muscadel.erc.monash.edu/. We anticipate this comprehensive review and the application of deep learning will provide practical guide and useful insights into PTM prediction and inspire future bioinformatics studies in the related fields.
Goldenberg, Igor; Webb, Geoffrey I.
Survey of distance measures for quantifying concept drift and shift in numeric data
Knowledge and Information Systems, vol. 60, no. 2, pp. 591-615, 2019, ISSN: 0219-3116.
@article{Goldenberg2018,
title = {Survey of distance measures for quantifying concept drift and shift in numeric data},
author = {Igor Goldenberg and Geoffrey I. Webb},
doi = {10.1007/s10115-018-1257-z},
issn = {0219-3116},
year = {2019},
date = {2019-01-01},
journal = {Knowledge and Information Systems},
volume = {60},
number = {2},
pages = {591-615},
abstract = {Deployed machine learning systems are necessarily learned from historical data and are often applied to current data. When the world changes, the learned models can lose fidelity. Such changes to the statistical properties of data over time are known as concept drift. Similarly, models are often learned in one context, but need to be applied in another. This is called concept shift. Quantifying the magnitude of drift or shift, especially in the context of covariate drift or shift, or unsupervised learning, requires use of measures of distance between distributions. In this paper, we survey such distance measures with respect to their suitability for estimating drift and shift magnitude between samples of numeric data.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deployed machine learning systems are necessarily learned from historical data and are often applied to current data. When the world changes, the learned models can lose fidelity. Such changes to the statistical properties of data over time are known as concept drift. Similarly, models are often learned in one context, but need to be applied in another. This is called concept shift. Quantifying the magnitude of drift or shift, especially in the context of covariate drift or shift, or unsupervised learning, requires use of measures of distance between distributions. In this paper, we survey such distance measures with respect to their suitability for estimating drift and shift magnitude between samples of numeric data.
Manapragada, Chaitanya; Webb, Geoffrey I.; Salehi, Mahsa
Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1953–1962, ACM, London, United Kingdom, 2018, ISBN: 978-1-4503-5552-0.
@inproceedings{ManapragadaEtAl18,
title = {Extremely Fast Decision Tree},
author = {Chaitanya Manapragada and Geoffrey I. Webb and Mahsa Salehi},
doi = {10.1145/3219819.3220005},
isbn = {978-1-4503-5552-0},
year = {2018},
date = {2018-01-01},
booktitle = {Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
pages = {1953\textendash1962},
publisher = {ACM},
address = {London, United Kingdom},
series = {KDD '18},
abstract = {We introduce a novel incremental decision tree learning algorithm, Hoeffding Anytime Tree, that is statistically more efficient than the current state-of-the-art, Hoeffding Tree. We demonstrate that an implementation of Hoeffding Anytime Tree\textemdash"Extremely Fast Decision Tree", a minor modification to the MOA implementation of Hoeffding Tree\textemdashobtains significantly superior prequential accuracy on most of the largest classification datasets from the UCI repository. Hoeffding Anytime Tree produces the asymptotic batch tree in the limit, is naturally resilient to concept drift, and can be used as a higher accuracy replacement for Hoeffding Tree in most scenarios, at a small additional computational cost.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We introduce a novel incremental decision tree learning algorithm, Hoeffding Anytime Tree, that is statistically more efficient than the current state-of-the-art, Hoeffding Tree. We demonstrate that an implementation of Hoeffding Anytime Tree—"Extremely Fast Decision Tree", a minor modification to the MOA implementation of Hoeffding Tree—obtains significantly superior prequential accuracy on most of the largest classification datasets from the UCI repository. Hoeffding Anytime Tree produces the asymptotic batch tree in the limit, is naturally resilient to concept drift, and can be used as a higher accuracy replacement for Hoeffding Tree in most scenarios, at a small additional computational cost.
Nguyen, Khanh; Le, Trung; Nguyen, Tu Dinh; Phung, Dinh; Webb, Geoffrey I.
Robust Bayesian Kernel Machine via Stein Variational Gradient Descent for Big Data
Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 2003–2011, ACM, London, United Kingdom, 2018, ISBN: 978-1-4503-5552-0.
@inproceedings{nguyenetal18,
title = {Robust Bayesian Kernel Machine via Stein Variational Gradient Descent for Big Data},
author = {Khanh Nguyen and Trung Le and Tu Dinh Nguyen and Dinh Phung and Geoffrey I. Webb},
url = {http://doi.acm.org/10.1145/3219819.3220015},
doi = {10.1145/3219819.3220015},
isbn = {978-1-4503-5552-0},
year = {2018},
date = {2018-01-01},
booktitle = {Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
pages = {2003\textendash2011},
publisher = {ACM},
address = {London, United Kingdom},
series = {KDD '18},
abstract = {Kernel methods are powerful supervised machine learning models for their strong generalization ability, especially on limited data to effectively generalize on unseen data. However, most kernel methods, including the state-of-the-art LIBSVM, are vulnerable to the curse of kernelization, making them infeasible to apply to large-scale datasets. This issue is exacerbated when kernel methods are used in conjunction with a grid search to tune their kernel parameters and hyperparameters which brings in the question of model robustness when applied to real datasets. In this paper, we propose a robust Bayesian Kernel Machine (BKM) - a Bayesian kernel machine that exploits the strengths of both the Bayesian modelling and kernel methods. A key challenge for such a formulation is the need for an efficient learning algorithm. To this end, we successfully extended the recent Stein variational theory for Bayesian inference for our proposed model, resulting in fast and efficient learning and prediction algorithms. Importantly our proposed BKM is resilient to the curse of kernelization, hence making it applicable to large-scale datasets and robust to parameter tuning, avoiding the associated expense and potential pitfalls with current practice of parameter tuning. Our extensive experimental results on 12 benchmark datasets show that our BKM without tuning any parameter can achieve comparable predictive performance with the state-of-the-art LIBSVM and significantly outperforms other baselines, while obtaining significantly speedup in terms of the total training time compared with its rivals.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Kernel methods are powerful supervised machine learning models for their strong generalization ability, especially on limited data to effectively generalize on unseen data. However, most kernel methods, including the state-of-the-art LIBSVM, are vulnerable to the curse of kernelization, making them infeasible to apply to large-scale datasets. This issue is exacerbated when kernel methods are used in conjunction with a grid search to tune their kernel parameters and hyperparameters which brings in the question of model robustness when applied to real datasets. In this paper, we propose a robust Bayesian Kernel Machine (BKM) - a Bayesian kernel machine that exploits the strengths of both the Bayesian modelling and kernel methods. A key challenge for such a formulation is the need for an efficient learning algorithm. To this end, we successfully extended the recent Stein variational theory for Bayesian inference for our proposed model, resulting in fast and efficient learning and prediction algorithms. Importantly our proposed BKM is resilient to the curse of kernelization, hence making it applicable to large-scale datasets and robust to parameter tuning, avoiding the associated expense and potential pitfalls with current practice of parameter tuning. Our extensive experimental results on 12 benchmark datasets show that our BKM without tuning any parameter can achieve comparable predictive performance with the state-of-the-art LIBSVM and significantly outperforms other baselines, while obtaining significantly speedup in terms of the total training time compared with its rivals.
Petitjean, Francois; Buntine, Wray; Webb, Geoffrey I.; Zaidi, Nayyar
Machine Learning, vol. 107, no. 8-10, pp. 1303-1331, 2018, ISSN: 1573-0565.
@article{Petitjean2018,
title = {Accurate parameter estimation for Bayesian network classifiers using hierarchical Dirichlet processes},
author = {Francois Petitjean and Wray Buntine and Geoffrey I. Webb and Nayyar Zaidi},
url = {https://rdcu.be/OX3g},
doi = {10.1007/s10994-018-5718-0},
issn = {1573-0565},
year = {2018},
date = {2018-01-01},
journal = {Machine Learning},
volume = {107},
number = {8-10},
pages = {1303-1331},
abstract = {This paper introduces a novel parameter estimation method for the probability tables of Bayesian network classifiers (BNCs), using hierarchical Dirichlet processes (HDPs). The main result of this paper is to show that improved parameter estimation allows BNCs to outperform leading learning methods such as random forest for both 0\textendash1 loss and RMSE, albeit just on categorical datasets. As data assets become larger, entering the hyped world of ``big'', efficient accurate classification requires three main elements: (1) classifiers with low-bias that can capture the fine-detail of large datasets (2) out-of-core learners that can learn from data without having to hold it all in main memory and (3) models that can classify new data very efficiently. The latest BNCs satisfy these requirements. Their bias can be controlled easily by increasing the number of parents of the nodes in the graph. Their structure can be learned out of core with a limited number of passes over the data. However, as the bias is made lower to accurately model classification tasks, so is the accuracy of their parameters' estimates, as each parameter is estimated from ever decreasing quantities of data. In this paper, we introduce the use of HDPs for accurate BNC parameter estimation even with lower bias. We conduct an extensive set of experiments on 68 standard datasets and demonstrate that our resulting classifiers perform very competitively with random forest in terms of prediction, while keeping the out-of-core capability and superior classification time.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper introduces a novel parameter estimation method for the probability tables of Bayesian network classifiers (BNCs), using hierarchical Dirichlet processes (HDPs). The main result of this paper is to show that improved parameter estimation allows BNCs to outperform leading learning methods such as random forest for both 0–1 loss and RMSE, albeit just on categorical datasets. As data assets become larger, entering the hyped world of ``big'', efficient accurate classification requires three main elements: (1) classifiers with low-bias that can capture the fine-detail of large datasets (2) out-of-core learners that can learn from data without having to hold it all in main memory and (3) models that can classify new data very efficiently. The latest BNCs satisfy these requirements. Their bias can be controlled easily by increasing the number of parents of the nodes in the graph. Their structure can be learned out of core with a limited number of passes over the data. However, as the bias is made lower to accurately model classification tasks, so is the accuracy of their parameters' estimates, as each parameter is estimated from ever decreasing quantities of data. In this paper, we introduce the use of HDPs for accurate BNC parameter estimation even with lower bias. We conduct an extensive set of experiments on 68 standard datasets and demonstrate that our resulting classifiers perform very competitively with random forest in terms of prediction, while keeping the out-of-core capability and superior classification time.
Shi, Wenzhong; Zhang, Anshu; Webb, Geoffrey I.
Mining significant crisp-fuzzy spatial association rules
International Journal of Geographical Information Science, vol. 32, no. 6, pp. 1247-1270, 2018.
@article{doi:10.1080/13658816.2018.1434525,
title = {Mining significant crisp-fuzzy spatial association rules},
author = {Wenzhong Shi and Anshu Zhang and Geoffrey I. Webb},
url = {http://www.tandfonline.com/eprint/aMdSMrAGuEHsHWSzIuqm/full},
doi = {10.1080/13658816.2018.1434525},
year = {2018},
date = {2018-01-01},
journal = {International Journal of Geographical Information Science},
volume = {32},
number = {6},
pages = {1247-1270},
publisher = {Taylor \& Francis},
abstract = {Spatial association rule mining (SARM) is an important data mining task for understanding implicit and sophisticated interactions in spatial data. The usefulness of SARM results, represented as sets of rules, depends on their reliability: the abundance of rules, control over the risk of spurious rules, and accuracy of rule interestingness measure (RIM) values. This study presents crisp-fuzzy SARM, a novel SARM method that can enhance the reliability of resultant rules. The method firstly prunes dubious rules using statistically sound tests and crisp supports for the patterns involved, and then evaluates RIMs of accepted rules using fuzzy supports. For the RIM evaluation stage, the study also proposes a Gaussian-curve-based fuzzy data discretization model for SARM with improved design for spatial semantics. The proposed techniques were evaluated by both synthetic and real-world data. The synthetic data was generated with predesigned rules and RIM values, thus the reliability of SARM results could be confidently and quantitatively evaluated. The proposed techniques showed high efficacy in enhancing the reliability of SARM results in all three aspects. The abundance of resultant rules was improved by 50% or more compared with using conventional fuzzy SARM. Minimal risk of spurious rules was guaranteed by statistically sound tests. The probability that the entire result contained any spurious rules was below 1%. The RIM values also avoided large positive errors committed by crisp SARM, which typically exceeded 50% for representative RIMs. The real-world case study on New York City points of interest reconfirms the improved reliability of crisp-fuzzy SARM results, and demonstrates that such improvement is critical for practical spatial data analytics and decision support.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Spatial association rule mining (SARM) is an important data mining task for understanding implicit and sophisticated interactions in spatial data. The usefulness of SARM results, represented as sets of rules, depends on their reliability: the abundance of rules, control over the risk of spurious rules, and accuracy of rule interestingness measure (RIM) values. This study presents crisp-fuzzy SARM, a novel SARM method that can enhance the reliability of resultant rules. The method firstly prunes dubious rules using statistically sound tests and crisp supports for the patterns involved, and then evaluates RIMs of accepted rules using fuzzy supports. For the RIM evaluation stage, the study also proposes a Gaussian-curve-based fuzzy data discretization model for SARM with improved design for spatial semantics. The proposed techniques were evaluated by both synthetic and real-world data. The synthetic data was generated with predesigned rules and RIM values, thus the reliability of SARM results could be confidently and quantitatively evaluated. The proposed techniques showed high efficacy in enhancing the reliability of SARM results in all three aspects. The abundance of resultant rules was improved by 50% or more compared with using conventional fuzzy SARM. Minimal risk of spurious rules was guaranteed by statistically sound tests. The probability that the entire result contained any spurious rules was below 1%. The RIM values also avoided large positive errors committed by crisp SARM, which typically exceeded 50% for representative RIMs. The real-world case study on New York City points of interest reconfirms the improved reliability of crisp-fuzzy SARM results, and demonstrates that such improvement is critical for practical spatial data analytics and decision support.
Song, J.; Li, F.; Takemoto, K.; Haffari, G.; Akutsu, T.; Chou, K. C.; Webb, G. I.
Journal of Theoretical Biology, vol. 443, pp. 125-137, 2018.
@article{SongEtAl18,
title = {PREvaIL, an integrative approach for inferring catalytic residues using sequence, structural, and network features in a machine-learning framework},
author = {J. Song and F. Li and K. Takemoto and G. Haffari and T. Akutsu and K. C. Chou and G. I. Webb},
url = {https://authors.elsevier.com/c/1WWQY57ilzyRc},
doi = {10.1016/j.jtbi.2018.01.023},
year = {2018},
date = {2018-01-01},
journal = {Journal of Theoretical Biology},
volume = {443},
pages = {125-137},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Tan, Chang Wei; Herrmann, Matthieu; Forestier, Germain; Webb, Geoffrey I.; Petitjean, Francois
Efficient search of the best warping window for Dynamic Time Warping
Proceedings of the 2018 SIAM International Conference on Data Mining, pp. 459-467, 2018.
@inproceedings{TanEtAl18,
title = {Efficient search of the best warping window for Dynamic Time Warping},
author = {Chang Wei Tan and Matthieu Herrmann and Germain Forestier and Geoffrey I. Webb and Francois Petitjean},
year = {2018},
date = {2018-01-01},
booktitle = {Proceedings of the 2018 SIAM International Conference on Data Mining},
pages = {459-467},
abstract = {Time series classification maps time series to labels. The nearest neighbour algorithm (NN) using the Dynamic Time Warping (DTW) similarity measure is a leading algorithm for this task and a component of the current best ensemble classifiers for time series. However, NN-DTW is only a winning combination when its meta-parameter - its warping window - is learned from the training data. The warping window (WW) intuitively controls the amount of distortion allowed when comparing a pair of time series. With a training database of N time series of lengths L, a naive approach to learning the WW requires Omega(N^2 . L^3) operations. This often translates in NN-DTW requiring days for training on datasets containing a few thousand time series only. In this paper, we introduce FastWWSearch: an efficient and exact method to learn WW. We show on 86 datasets that our method is always faster than the state of the art, with at least one order of magnitude and up to 1000x speed-up.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Time series classification maps time series to labels. The nearest neighbour algorithm (NN) using the Dynamic Time Warping (DTW) similarity measure is a leading algorithm for this task and a component of the current best ensemble classifiers for time series. However, NN-DTW is only a winning combination when its meta-parameter - its warping window - is learned from the training data. The warping window (WW) intuitively controls the amount of distortion allowed when comparing a pair of time series. With a training database of N time series of lengths L, a naive approach to learning the WW requires Omega(N^2 . L^3) operations. This often translates in NN-DTW requiring days for training on datasets containing a few thousand time series only. In this paper, we introduce FastWWSearch: an efficient and exact method to learn WW. We show on 86 datasets that our method is always faster than the state of the art, with at least one order of magnitude and up to 1000x speed-up.
Tan, Chang Wei; Webb, Geoffrey I.; Petitjean, Francois; Reichl, Paul
Tamping Effectiveness Prediction Using Supervised Machine Learning Techniques
Proceedings of the First International Conference on Rail Transportation (ICRT-17), ACSE, 2018.
@inproceedings{TanEtAl18b,
title = {Tamping Effectiveness Prediction Using Supervised Machine Learning Techniques},
author = {Chang Wei Tan and Geoffrey I. Webb and Francois Petitjean and Paul Reichl},
doi = {10.1061/9780784481257.101},
year = {2018},
date = {2018-01-01},
booktitle = {Proceedings of the First International Conference on Rail Transportation (ICRT-17)},
publisher = {ACSE},
abstract = {Railway maintenance planning is critical in maintaining track assets. Tamping is a common railway maintenance procedure and is often used when geometrical issues are first identified. Tamping repacks ballast particles under sleepers to restore the correct geometrical position of ballasted tracks. However, historical data shows that tamping is not always effective in restoring track to a satisfactory condition. Furthermore, ineffective, or unnecessary tamping tends to reduce the lifetime of existing track. An intuitive way of preventing ineffective tamping is to predict the likely tamping effectiveness. This work aims to predict the likely tamping effectiveness ahead of time using supervised machine learning techniques. Supervised machine learning techniques predict an outcome using labelled training data. In this case, the training database consists of multivariate sensor data from instrumented revenue vehicles (IRVs). The data between the previous and current tamping dates are used. This forms a time series database labelled with the tamping effectiveness of each track location based on the responses recorded from the IRVs before and after tamping. The labelled time series database is then used to train a time series classifier for prediction. This work uses the state of the art time series classification algorithm, k-nearest neighbour (k-NN) extended to the case of multivariate time series. k-NN is a non-parametric algorithm that does not make assumptions on the underlying model of the training data. With a sufficiently large training database, non-parametric algorithms can outperform parametric algorithms. Using k-NN, the tamping effectiveness of a potential tamping location that is not in the training database, or locations in the next tamping cycle, is predicted using the expected tamping effectiveness from a location in the training database that is the most similar to the target. This allows the algorithm to effectively to identify locations where tamping is likely to be ineffective. This work achieves high accuracy in the prediction of tamping effectiveness even at 12 weeks before tamping. It is hoped that the methodology will help in assisting decision making for maintenance planning activities.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Railway maintenance planning is critical in maintaining track assets. Tamping is a common railway maintenance procedure and is often used when geometrical issues are first identified. Tamping repacks ballast particles under sleepers to restore the correct geometrical position of ballasted tracks. However, historical data shows that tamping is not always effective in restoring track to a satisfactory condition. Furthermore, ineffective, or unnecessary tamping tends to reduce the lifetime of existing track. An intuitive way of preventing ineffective tamping is to predict the likely tamping effectiveness. This work aims to predict the likely tamping effectiveness ahead of time using supervised machine learning techniques. Supervised machine learning techniques predict an outcome using labelled training data. In this case, the training database consists of multivariate sensor data from instrumented revenue vehicles (IRVs). The data between the previous and current tamping dates are used. This forms a time series database labelled with the tamping effectiveness of each track location based on the responses recorded from the IRVs before and after tamping. The labelled time series database is then used to train a time series classifier for prediction. This work uses the state of the art time series classification algorithm, k-nearest neighbour (k-NN) extended to the case of multivariate time series. k-NN is a non-parametric algorithm that does not make assumptions on the underlying model of the training data. With a sufficiently large training database, non-parametric algorithms can outperform parametric algorithms. Using k-NN, the tamping effectiveness of a potential tamping location that is not in the training database, or locations in the next tamping cycle, is predicted using the expected tamping effectiveness from a location in the training database that is the most similar to the target. This allows the algorithm to effectively to identify locations where tamping is likely to be ineffective. This work achieves high accuracy in the prediction of tamping effectiveness even at 12 weeks before tamping. It is hoped that the methodology will help in assisting decision making for maintenance planning activities.
Wang, Huilin; Feng, Liubin; Webb, Geoffrey I; Kurgan, Lukasz; Song, Jiangning; Lin, Donghai
Critical evaluation of bioinformatics tools for the prediction of protein crystallization propensity
Briefings in Bioinformatics, vol. 19, no. 5, pp. 838-852, 2018.
@article{WangEtAl18,
title = {Critical evaluation of bioinformatics tools for the prediction of protein crystallization propensity},
author = {Huilin Wang and Liubin Feng and Geoffrey I Webb and Lukasz Kurgan and Jiangning Song and Donghai Lin},
doi = {10.1093/bib/bbx018},
year = {2018},
date = {2018-01-01},
journal = {Briefings in Bioinformatics},
volume = {19},
number = {5},
pages = {838-852},
abstract = {X-ray crystallography is the main tool for structural determination of proteins. Yet, the underlying crystallization process is costly, has a high attrition rate and involves a series of trial-and-error attempts to obtain diffraction-quality crystals. The Structural Genomics Consortium aims to systematically solve representative structures of major protein-fold classes using primarily high-throughput X-ray crystallography. The attrition rate of these efforts can be improved by selection of proteins that are potentially easier to be crystallized. In this context, bioinformatics approaches have been developed to predict crystallization propensities based on protein sequences. These approaches are used to facilitate prioritization of the most promising target proteins, search for alternative structural orthologues of the target proteins and suggest designs of constructs capable of potentially enhancing the likelihood of successful crystallization. We reviewed and compared nine predictors of protein crystallization propensity. Moreover, we demonstrated that integrating selected outputs from multiple predictors as candidate input features to build the predictive model results in a significantly higher predictive performance when compared to using these predictors individually. Furthermore, we also introduced a new and accurate predictor of protein crystallization propensity, Crysf, which uses functional features extracted from UniProt as inputs. This comprehensive review will assist structural biologists in selecting the most appropriate predictor, and is also beneficial for bioinformaticians to develop a new generation of predictive algorithms.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
X-ray crystallography is the main tool for structural determination of proteins. Yet, the underlying crystallization process is costly, has a high attrition rate and involves a series of trial-and-error attempts to obtain diffraction-quality crystals. The Structural Genomics Consortium aims to systematically solve representative structures of major protein-fold classes using primarily high-throughput X-ray crystallography. The attrition rate of these efforts can be improved by selection of proteins that are potentially easier to be crystallized. In this context, bioinformatics approaches have been developed to predict crystallization propensities based on protein sequences. These approaches are used to facilitate prioritization of the most promising target proteins, search for alternative structural orthologues of the target proteins and suggest designs of constructs capable of potentially enhancing the likelihood of successful crystallization. We reviewed and compared nine predictors of protein crystallization propensity. Moreover, we demonstrated that integrating selected outputs from multiple predictors as candidate input features to build the predictive model results in a significantly higher predictive performance when compared to using these predictors individually. Furthermore, we also introduced a new and accurate predictor of protein crystallization propensity, Crysf, which uses functional features extracted from UniProt as inputs. This comprehensive review will assist structural biologists in selecting the most appropriate predictor, and is also beneficial for bioinformaticians to develop a new generation of predictive algorithms.
Webb, Geoffrey I; Lee, Loong Kuan; Goethals, Bart; Petitjean, Francois
Analyzing concept drift and shift from sample data
Data Mining and Knowledge Discovery, vol. 32, no. 5, pp. 1179-1199, 2018.
@article{WebbEtAl18,
title = {Analyzing concept drift and shift from sample data},
author = {Geoffrey I Webb and Loong Kuan Lee and Bart Goethals and Francois Petitjean},
url = {http://rdcu.be/IUTI},
doi = {10.1007/s10618-018-0554-1},
year = {2018},
date = {2018-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {32},
number = {5},
pages = {1179-1199},
abstract = {Concept drift and shift are major issues that greatly affect the accuracy and reliability of many real-world applications of machine learning. We propose a new data mining task, concept drift mapping - the description and analysis of instances of concept drift or shift. We argue that concept drift mapping is an essential prerequisite for tackling concept drift and shift. We propose tools for this purpose, arguing for the importance of quantitative descriptions of drift and shift in marginal distributions. We present quantitative concept drift mapping techniques, along with methods for visualizing their results. We illustrate their effectiveness for real-world applications across energy-pricing, vegetation monitoring and airline scheduling.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Concept drift and shift are major issues that greatly affect the accuracy and reliability of many real-world applications of machine learning. We propose a new data mining task, concept drift mapping - the description and analysis of instances of concept drift or shift. We argue that concept drift mapping is an essential prerequisite for tackling concept drift and shift. We propose tools for this purpose, arguing for the importance of quantitative descriptions of drift and shift in marginal distributions. We present quantitative concept drift mapping techniques, along with methods for visualizing their results. We illustrate their effectiveness for real-world applications across energy-pricing, vegetation monitoring and airline scheduling.
Zaidi, Nayyar A.; Petitjean, Francois; Webb, Geoffrey I.
Proceedings of the 2018 SIAM International Conference on Data Mining, pp. 459-467, 2018.
@inproceedings{Zaidi2018,
title = {Efficient and Effective Accelerated Hierarchical Higher-Order Logistic Regression for Large Data Quantities},
author = {Nayyar A. Zaidi and Francois Petitjean and Geoffrey I. Webb},
doi = {10.1137/1.9781611975321.52},
year = {2018},
date = {2018-01-01},
booktitle = {Proceedings of the 2018 SIAM International Conference on Data Mining},
pages = {459-467},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Zaidi, Nayyar A.; Webb, Geoffrey I.; Petitjean, Francois; Forestier, Germain
On the Inter-Relationships among Drift Rate, Forgetting Rate, Bias/Variance Profile and Error
arxiv, pp. 1801.09354, 2018.
@article{ZaidiEtAl18b,
title = {On the Inter-Relationships among Drift Rate,
Forgetting Rate, Bias/Variance Profile and Error},
author = {Nayyar A. Zaidi and Geoffrey I. Webb and Francois Petitjean and Germain Forestier},
url = {https://arxiv.org/abs/1801.09354},
year = {2018},
date = {2018-01-01},
journal = {arxiv},
pages = {1801.09354},
abstract = {We propose two general and falsifiable hypotheses about expectations on generalization error when learning in the context of concept drift. One posits that as drift rate increases, the forgetting rate that minimizes generalization error will also increase and vice versa. The other posits that as a learner's forgetting rate increases, the bias/variance profile that minimizes generalization error will have lower variance and vice versa. These hypotheses lead to the concept of the sweet path, a path through the 3-d space of alternative drift rates, forgetting rates and bias/variance profiles on which generalization error will be minimized, such that slow drift is coupled with low forgetting and low bias, while rapid drift is coupled with fast forgetting and low variance. We present experiments that support the existence of such a sweet path. We also demonstrate that simple learners that select appropriate forgetting rates and bias/variance profiles are highly competitive with the state-of-the-art in incremental learners for concept drift on real-world drift problems.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We propose two general and falsifiable hypotheses about expectations on generalization error when learning in the context of concept drift. One posits that as drift rate increases, the forgetting rate that minimizes generalization error will also increase and vice versa. The other posits that as a learner's forgetting rate increases, the bias/variance profile that minimizes generalization error will have lower variance and vice versa. These hypotheses lead to the concept of the sweet path, a path through the 3-d space of alternative drift rates, forgetting rates and bias/variance profiles on which generalization error will be minimized, such that slow drift is coupled with low forgetting and low bias, while rapid drift is coupled with fast forgetting and low variance. We present experiments that support the existence of such a sweet path. We also demonstrate that simple learners that select appropriate forgetting rates and bias/variance profiles are highly competitive with the state-of-the-art in incremental learners for concept drift on real-world drift problems.
Phung, Dinh; Tseng, Vincent S.; Webb, Geoffrey I.; Ho, Bao; Ganji, Mohadeseh; Rashidi, Lida (Ed.)
Springer, vol. 10939, 2018.
@proceedings{phung2018advances,
title = {Advances in Knowledge Discovery and Data Mining: 22nd Pacific-Asia Conference, PAKDD 2018, Melbourne, VIC, Australia, June 3-6, 2018, Proceedings},
editor = {Dinh Phung and Vincent S. Tseng and Geoffrey I. Webb and Bao Ho and Mohadeseh Ganji and Lida Rashidi},
url = {http://link.springer.com/openurl.asp?genre=issue\&issn=0302-9743\&volume=10937},
year = {2018},
date = {2018-01-01},
volume = {10939},
publisher = {Springer},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
An, Yi; Wang, Jiawei; Li, Chen; Leier, Andre; Marquez-Lago, Tatiana; Wilksch, Jonathan; Zhang, Yang; Webb, Geoffrey I.; Song, Jiangning; Lithgow, Trevor
Briefings in Bioinformatics, vol. 19, no. 1, pp. 148-161, 2018.
@article{AnEtAl2016,
title = {Comprehensive assessment and performance improvement of effector protein predictors for bacterial secretion systems III, IV and VI},
author = {Yi An and Jiawei Wang and Chen Li and Andre Leier and Tatiana Marquez-Lago and Jonathan Wilksch and Yang Zhang and Geoffrey I. Webb and Jiangning Song and Trevor Lithgow},
doi = {10.1093/bib/bbw100},
year = {2018},
date = {2018-01-01},
journal = {Briefings in Bioinformatics},
volume = {19},
number = {1},
pages = {148-161},
abstract = {Bacterial effector proteins secreted by various protein secretion systems play crucial roles in host-pathogen interactions. In this context, computational tools capable of accurately predicting effector proteins of the various types of bacterial secretion systems are highly desirable. Existing computational approaches use different machine learning (ML) techniques and heterogeneous features derived from protein sequences and/or structural information. These predictors differ not only in terms of the used ML methods but also with respect to the used curated data sets, the features selection and their prediction performance. Here, we provide a comprehensive survey and benchmarking of currently available tools for the prediction of effector proteins of bacterial types III, IV and VI secretion systems (T3SS, T4SS and T6SS, respectively). We review core algorithms, feature selection techniques, tool availability and applicability and evaluate the prediction performance based on carefully curated independent test data sets. In an effort to improve predictive performance, we constructed three ensemble models based on ML algorithms by integrating the output of all individual predictors reviewed. Our benchmarks demonstrate that these ensemble models outperform all the reviewed tools for the prediction of effector proteins of T3SS and T4SS. The webserver of the proposed ensemble methods for T3SS and T4SS effector protein prediction is freely available at http://tbooster.erc.monash.edu/index.jsp. We anticipate that this survey will serve as a useful guide for interested users and that the new ensemble predictors will stimulate research into host-pathogen relationships and inspiration for the development of new bioinformatics tools for predicting effector proteins of T3SS, T4SS and T6SS.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Bacterial effector proteins secreted by various protein secretion systems play crucial roles in host-pathogen interactions. In this context, computational tools capable of accurately predicting effector proteins of the various types of bacterial secretion systems are highly desirable. Existing computational approaches use different machine learning (ML) techniques and heterogeneous features derived from protein sequences and/or structural information. These predictors differ not only in terms of the used ML methods but also with respect to the used curated data sets, the features selection and their prediction performance. Here, we provide a comprehensive survey and benchmarking of currently available tools for the prediction of effector proteins of bacterial types III, IV and VI secretion systems (T3SS, T4SS and T6SS, respectively). We review core algorithms, feature selection techniques, tool availability and applicability and evaluate the prediction performance based on carefully curated independent test data sets. In an effort to improve predictive performance, we constructed three ensemble models based on ML algorithms by integrating the output of all individual predictors reviewed. Our benchmarks demonstrate that these ensemble models outperform all the reviewed tools for the prediction of effector proteins of T3SS and T4SS. The webserver of the proposed ensemble methods for T3SS and T4SS effector protein prediction is freely available at http://tbooster.erc.monash.edu/index.jsp. We anticipate that this survey will serve as a useful guide for interested users and that the new ensemble predictors will stimulate research into host-pathogen relationships and inspiration for the development of new bioinformatics tools for predicting effector proteins of T3SS, T4SS and T6SS.
Chen, Zhen; Zhao, Pei; Li, Fuyi; Leier, Andre; Marquez-Lago, Tatiana T; Wang, Yanan; Webb, Geoffrey I; Smith, A Ian; Daly, Roger J; Chou, Kuo-Chen; Song, Jiangning
Bioinformatics, pp. 2499-2502, 2018.
@article{ChenEtAl18,
title = {iFeature: A python package and web server for features extraction and selection from protein and peptide sequences},
author = {Zhen Chen and Pei Zhao and Fuyi Li and Andre Leier and Tatiana T Marquez-Lago and Yanan Wang and Geoffrey I Webb and A Ian Smith and Roger J Daly and Kuo-Chen Chou and Jiangning Song},
doi = {10.1093/bioinformatics/bty140},
year = {2018},
date = {2018-01-01},
journal = {Bioinformatics},
pages = {2499-2502},
abstract = {Structural and physiochemical descriptors extracted from sequence data have been widely used to represent sequences and predict structural, functional, expression and interaction profiles of proteins and peptides as well as DNAs/RNAs. Here, we present iFeature, a versatile Python-based toolkit for generating various numerical feature representation schemes for both protein and peptide sequences. iFeature is capable of calculating and extracting a comprehensive spectrum of 18 major sequence encoding schemes that encompass 53 different types of feature descriptors. It also allows users to extract specific amino acid properties from the AAindex database. Furthermore, iFeature integrates 12 different types of commonly used feature clustering, selection and dimensionality reduction algorithms, greatly facilitating training, analysis and benchmarking of machine-learning models. The functionality of iFeature is made freely available via an online web server and a stand-alone toolkit.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Structural and physiochemical descriptors extracted from sequence data have been widely used to represent sequences and predict structural, functional, expression and interaction profiles of proteins and peptides as well as DNAs/RNAs. Here, we present iFeature, a versatile Python-based toolkit for generating various numerical feature representation schemes for both protein and peptide sequences. iFeature is capable of calculating and extracting a comprehensive spectrum of 18 major sequence encoding schemes that encompass 53 different types of feature descriptors. It also allows users to extract specific amino acid properties from the AAindex database. Furthermore, iFeature integrates 12 different types of commonly used feature clustering, selection and dimensionality reduction algorithms, greatly facilitating training, analysis and benchmarking of machine-learning models. The functionality of iFeature is made freely available via an online web server and a stand-alone toolkit.
Li, Chen; Clark, Liah V. T.; Zhang, Rory; Porebski, Benjamin T.; McCoey, Julia M.; Borg, Natalie A.; Webb, Geoffrey I.; Kass, Itamar; Buckle, Malcolm; Song, Jiangning; Woolfson, Adrian; Buckle, Ashley M.
Structural Capacitance in Protein Evolution and Human Diseases
Journal of Molecular Biology, vol. 430, no. 18, pp. 3200-3217, 2018, ISSN: 0022-2836.
@article{Li2018,
title = {Structural Capacitance in Protein Evolution and Human Diseases},
author = {Chen Li and Liah V. T. Clark and Rory Zhang and Benjamin T. Porebski and Julia M. McCoey and Natalie A. Borg and Geoffrey I. Webb and Itamar Kass and Malcolm Buckle and Jiangning Song and Adrian Woolfson and Ashley M. Buckle},
doi = {10.1016/j.jmb.2018.06.051},
issn = {0022-2836},
year = {2018},
date = {2018-01-01},
journal = {Journal of Molecular Biology},
volume = {430},
number = {18},
pages = {3200-3217},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Tan, CW; Webb, G. I.; Petitjean, F; Reichl, P
Machine learning approaches for tamping effectiveness prediction
Proceedings of the International Heavy Haul Association Conference, Cape Town, South Africa, 2017.
@inproceedings{tan2017machine,
title = {Machine learning approaches for tamping effectiveness prediction},
author = {CW Tan and G. I. Webb and F Petitjean and P Reichl},
year = {2017},
date = {2017-01-01},
booktitle = {Proceedings of the International Heavy Haul Association Conference},
address = {Cape Town, South Africa},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Song, Jiangning; Li, Chen; Zheng, Cheng; Revote, Jerico; Zhang, Ziding; Webb, Geoffrey I.
Current Bioinformatics, vol. 12, no. 6, pp. 480-489, 2017, ISSN: 1574-8936/2212-392X.
@article{SongEtAl16,
title = {MetalExplorer, a Bioinformatics Tool for the Improved Prediction of Eight Types of Metal-binding Sites Using a Random Forest Algorithm with Two-step Feature Selection},
author = {Jiangning Song and Chen Li and Cheng Zheng and Jerico Revote and Ziding Zhang and Geoffrey I. Webb},
doi = {10.2174/2468422806666160618091522},
issn = {1574-8936/2212-392X},
year = {2017},
date = {2017-01-01},
journal = {Current Bioinformatics},
volume = {12},
number = {6},
pages = {480-489},
abstract = {Metalloproteins are highly involved in many biological processes,
including catalysis, recognition, transport, transcription, and signal
transduction. The metal ions they bind usually play enzymatic or structural
roles in mediating these diverse functional roles. Thus, the systematic
analysis and prediction of metal-binding sites using sequence and/or
structural information are crucial for understanding their
sequence-structure-function relationships. In this study, we propose
MetalExplorer (http://metalexplorer.erc.monash.edu.au/), a new machine
learning-based method for predicting eight different types of metal-binding
sites (Ca, Co, Cu, Fe, Ni, Mg, Mn, and Zn) in proteins. Our approach
combines heterogeneous sequence-, structure-, and residue contact
network-based features. The predictive performance of MetalExplorer was
tested by cross-validation and independent tests using non-redundant
datasets of known structures. This method applies a two-step feature
selection approach based on the maximum relevance minimum redundancy and
forward feature selection to identify the most informative features that
contribute to the prediction performance. With a precision of 60%,
MetalExplorer achieved high recall values, which ranged from 59% to 88% for
the eight metal ion types in fivefold cross-validation tests. Moreover, the
common and type-specific features in the optimal subsets of all metal ions
were characterized in terms of their contributions to the overall
performance. In terms of both benchmark and independent datasets at the 60%
precision control level, MetalExplorer compared favorably with an existing
metalloprotein prediction tool, SitePredict. Thus, MetalExplorer is expected
to be a powerful tool for the accurate prediction of potential metal-binding
sites and it should facilitate the functional analysis and rational design
of novel metalloproteins.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Metalloproteins are highly involved in many biological processes,
including catalysis, recognition, transport, transcription, and signal
transduction. The metal ions they bind usually play enzymatic or structural
roles in mediating these diverse functional roles. Thus, the systematic
analysis and prediction of metal-binding sites using sequence and/or
structural information are crucial for understanding their
sequence-structure-function relationships. In this study, we propose
MetalExplorer (http://metalexplorer.erc.monash.edu.au/), a new machine
learning-based method for predicting eight different types of metal-binding
sites (Ca, Co, Cu, Fe, Ni, Mg, Mn, and Zn) in proteins. Our approach
combines heterogeneous sequence-, structure-, and residue contact
network-based features. The predictive performance of MetalExplorer was
tested by cross-validation and independent tests using non-redundant
datasets of known structures. This method applies a two-step feature
selection approach based on the maximum relevance minimum redundancy and
forward feature selection to identify the most informative features that
contribute to the prediction performance. With a precision of 60%,
MetalExplorer achieved high recall values, which ranged from 59% to 88% for
the eight metal ion types in fivefold cross-validation tests. Moreover, the
common and type-specific features in the optimal subsets of all metal ions
were characterized in terms of their contributions to the overall
performance. In terms of both benchmark and independent datasets at the 60%
precision control level, MetalExplorer compared favorably with an existing
metalloprotein prediction tool, SitePredict. Thus, MetalExplorer is expected
to be a powerful tool for the accurate prediction of potential metal-binding
sites and it should facilitate the functional analysis and rational design
of novel metalloproteins.
including catalysis, recognition, transport, transcription, and signal
transduction. The metal ions they bind usually play enzymatic or structural
roles in mediating these diverse functional roles. Thus, the systematic
analysis and prediction of metal-binding sites using sequence and/or
structural information are crucial for understanding their
sequence-structure-function relationships. In this study, we propose
MetalExplorer (http://metalexplorer.erc.monash.edu.au/), a new machine
learning-based method for predicting eight different types of metal-binding
sites (Ca, Co, Cu, Fe, Ni, Mg, Mn, and Zn) in proteins. Our approach
combines heterogeneous sequence-, structure-, and residue contact
network-based features. The predictive performance of MetalExplorer was
tested by cross-validation and independent tests using non-redundant
datasets of known structures. This method applies a two-step feature
selection approach based on the maximum relevance minimum redundancy and
forward feature selection to identify the most informative features that
contribute to the prediction performance. With a precision of 60%,
MetalExplorer achieved high recall values, which ranged from 59% to 88% for
the eight metal ion types in fivefold cross-validation tests. Moreover, the
common and type-specific features in the optimal subsets of all metal ions
were characterized in terms of their contributions to the overall
performance. In terms of both benchmark and independent datasets at the 60%
precision control level, MetalExplorer compared favorably with an existing
metalloprotein prediction tool, SitePredict. Thus, MetalExplorer is expected
to be a powerful tool for the accurate prediction of potential metal-binding
sites and it should facilitate the functional analysis and rational design
of novel metalloproteins.
Song, Jiangning; Li, Fuyi; Leier, Andre; Marquez-Lago, Tatiana T; Akutsu, Tatsuya; Haffari, Gholamreza; Chou, Kuo-Chen; Webb, Geoffrey I; Pike, Robert N
Bioinformatics, vol. 34, no. 4, pp. 684-687, 2017.
@article{Song2017a,
title = {PROSPERous: high-throughput prediction of substrate cleavage sites for 90 proteases with improved accuracy},
author = {Jiangning Song and Fuyi Li and Andre Leier and Tatiana T Marquez-Lago and Tatsuya Akutsu and Gholamreza Haffari and Kuo-Chen Chou and Geoffrey I Webb and Robert N Pike},
doi = {10.1093/bioinformatics/btx670},
year = {2017},
date = {2017-01-01},
journal = {Bioinformatics},
volume = {34},
number = {4},
pages = {684-687},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Song, Jiangning; Wang, Huilin; Wang, Jiawei; Leier, Andre; Marquez-Lago, Tatiana; Yang, Bingjiao; Zhang, Ziding; Akutsu, Tatsuya; Webb, Geoffrey I.; Daly, Roger J.
Scientific Reports, vol. 7, no. 1, 2017.
@article{Song2017,
title = {PhosphoPredict: A bioinformatics tool for prediction of human kinase-specific phosphorylation substrates and sites by integrating heterogeneous feature selection},
author = {Jiangning Song and Huilin Wang and Jiawei Wang and Andre Leier and Tatiana Marquez-Lago and Bingjiao Yang and Ziding Zhang and Tatsuya Akutsu and Geoffrey I. Webb and Roger J. Daly},
doi = {10.1038/s41598-017-07199-4},
year = {2017},
date = {2017-01-01},
journal = {Scientific Reports},
volume = {7},
number = {1},
abstract = {Protein phosphorylation is a major form of post-translational modification (PTM) that regulates diverse cellular processes. In silico methods for phosphorylation site prediction can provide a useful and complementary strategy for complete phosphoproteome annotation. Here, we present a novel bioinformatics tool, PhosphoPredict, that combines protein sequence and functional features to predict kinase-specific substrates and their associated phosphorylation sites for 12 human kinases and kinase families, including ATM, CDKs, GSK-3, MAPKs, PKA, PKB, PKC, and SRC. To elucidate critical determinants, we identified feature subsets that were most informative and relevant for predicting substrate specificity for each individual kinase family. Extensive benchmarking experiments based on both five-fold cross-validation and independent tests indicated that the performance of PhosphoPredict is competitive with that of several other popular prediction tools, including KinasePhos, PPSP, GPS, and Musite. We found that combining protein functional and sequence features significantly improves phosphorylation site prediction performance across all kinases. Application of PhosphoPredict to the entire human proteome identified 150 to 800 potential phosphorylation substrates for each of the 12 kinases or kinase families. PhosphoPredict significantly extends the bioinformatics portfolio for kinase function analysis and will facilitate high-throughput identification of kinase-specific phosphorylation sites, thereby contributing to both basic and translational research programs.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Protein phosphorylation is a major form of post-translational modification (PTM) that regulates diverse cellular processes. In silico methods for phosphorylation site prediction can provide a useful and complementary strategy for complete phosphoproteome annotation. Here, we present a novel bioinformatics tool, PhosphoPredict, that combines protein sequence and functional features to predict kinase-specific substrates and their associated phosphorylation sites for 12 human kinases and kinase families, including ATM, CDKs, GSK-3, MAPKs, PKA, PKB, PKC, and SRC. To elucidate critical determinants, we identified feature subsets that were most informative and relevant for predicting substrate specificity for each individual kinase family. Extensive benchmarking experiments based on both five-fold cross-validation and independent tests indicated that the performance of PhosphoPredict is competitive with that of several other popular prediction tools, including KinasePhos, PPSP, GPS, and Musite. We found that combining protein functional and sequence features significantly improves phosphorylation site prediction performance across all kinases. Application of PhosphoPredict to the entire human proteome identified 150 to 800 potential phosphorylation substrates for each of the 12 kinases or kinase families. PhosphoPredict significantly extends the bioinformatics portfolio for kinase function analysis and will facilitate high-throughput identification of kinase-specific phosphorylation sites, thereby contributing to both basic and translational research programs.
Tan, Chang Wei; Webb, Geoffrey I.; Petitjean, Francois
Indexing and classifying gigabytes of time series under time warping
Proceedings of the 2017 SIAM International Conference on Data Mining, pp. 282-290, SIAM 2017.
@inproceedings{TanEtAl17a,
title = {Indexing and classifying gigabytes of time series under time warping},
author = {Chang Wei Tan and Geoffrey I. Webb and Francois Petitjean},
doi = {10.1137/1.9781611974973.32},
year = {2017},
date = {2017-01-01},
booktitle = {Proceedings of the 2017 SIAM International Conference on Data Mining},
pages = {282-290},
organization = {SIAM},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Wang, Jiawei; Yang, Bingjiao; Revote, Jerico; Leier, Andre; Marquez-Lago, Tatiana T.; Webb, Geoffrey I.; Song, Jiangning; Chou, Kuo-Chen; Lithgow, Trevor
Bioinformatics, vol. 33, no. 17, pp. 2756-2758, 2017.
@article{WangJEtAl17,
title = {POSSUM: a bioinformatics toolkit for generating numerical sequence feature descriptors based on PSSM profiles},
author = {Jiawei Wang and Bingjiao Yang and Jerico Revote and Andre Leier and Tatiana T. Marquez-Lago and Geoffrey I. Webb and Jiangning Song and Kuo-Chen Chou and Trevor Lithgow},
doi = {10.1093/bioinformatics/btx302},
year = {2017},
date = {2017-01-01},
journal = {Bioinformatics},
volume = {33},
number = {17},
pages = {2756-2758},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Wang, Yanan; Song, Jiangning; Marquez-Lago, Tatiana T.; Leier, Andre; Li, Chen; Lithgow, Trevor; Webb, Geoffrey I.; Shen, Hong-Bin
Knowledge-transfer learning for prediction of matrix metalloprotease substrate-cleavage sites
Scientific Reports, vol. 7, 2017.
@article{WangYEtAl17,
title = {Knowledge-transfer learning for prediction of matrix metalloprotease substrate-cleavage sites},
author = {Yanan Wang and Jiangning Song and Tatiana T. Marquez-Lago and Andre Leier and Chen Li and Trevor Lithgow and Geoffrey I. Webb and Hong-Bin Shen},
doi = {10.1038/s41598-017-06219-7},
year = {2017},
date = {2017-01-01},
journal = {Scientific Reports},
volume = {7},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Zaidi, Nayyar A.; Webb, Geoffrey I.
A Fast Trust-Region Newton Method for Softmax Logistic Regression
Proceedings of the 2017 SIAM International Conference on Data Mining, pp. 705-713, SIAM 2017.
@inproceedings{ZaidiWebb17,
title = {A Fast Trust-Region Newton Method for Softmax Logistic Regression},
author = {Nayyar A. Zaidi and Geoffrey I. Webb},
doi = {10.1137/1.9781611974973.79},
year = {2017},
date = {2017-01-01},
booktitle = {Proceedings of the 2017 SIAM International Conference on Data Mining},
pages = {705-713},
organization = {SIAM},
abstract = {With the emergence of big data, there has been a growing
interest in optimization routines that lead to faster convergence of Logistic Regression (LR). Among many optimization methods such as Gradient Descent, Quasi-Newton, Conjugate Gradient, etc., the Trust-region based truncated Newton method (TRON) algorithm has been shown to converge
the fastest. The TRON algorithm also forms an important
component of the highly efficient and widely used liblinear
package. It has been shown that the WANBIA-C trick of
scaling with the log of the naive Bayes conditional probabilities can greatly accelerate the convergence of LR trained using (first-order) Gradient Descent and (approximate second-order) Quasi-Newton optimization. In this work we study
the applicability of the WANBIA-C trick to TRON. We first
devise a TRON algorithm optimizing the softmax objective
function and then demonstrate that WANBIA-C style preconditioning can be beneficial for TRON, leading to an ex-
tremely fast (batch) LR algorithm. Second, we present a
comparative analysis of one-vs-all LR and softmax LR in
terms of the 0-1 Loss, Bias, Variance, RMSE, Log-Loss,
Training and Classification time, and show that softmax LR
leads to significantly better RMSE and Log-Loss. We evaluate our proposed approach on 51 benchmark datasets.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
With the emergence of big data, there has been a growing
interest in optimization routines that lead to faster convergence of Logistic Regression (LR). Among many optimization methods such as Gradient Descent, Quasi-Newton, Conjugate Gradient, etc., the Trust-region based truncated Newton method (TRON) algorithm has been shown to converge
the fastest. The TRON algorithm also forms an important
component of the highly efficient and widely used liblinear
package. It has been shown that the WANBIA-C trick of
scaling with the log of the naive Bayes conditional probabilities can greatly accelerate the convergence of LR trained using (first-order) Gradient Descent and (approximate second-order) Quasi-Newton optimization. In this work we study
the applicability of the WANBIA-C trick to TRON. We first
devise a TRON algorithm optimizing the softmax objective
function and then demonstrate that WANBIA-C style preconditioning can be beneficial for TRON, leading to an ex-
tremely fast (batch) LR algorithm. Second, we present a
comparative analysis of one-vs-all LR and softmax LR in
terms of the 0-1 Loss, Bias, Variance, RMSE, Log-Loss,
Training and Classification time, and show that softmax LR
leads to significantly better RMSE and Log-Loss. We evaluate our proposed approach on 51 benchmark datasets.
interest in optimization routines that lead to faster convergence of Logistic Regression (LR). Among many optimization methods such as Gradient Descent, Quasi-Newton, Conjugate Gradient, etc., the Trust-region based truncated Newton method (TRON) algorithm has been shown to converge
the fastest. The TRON algorithm also forms an important
component of the highly efficient and widely used liblinear
package. It has been shown that the WANBIA-C trick of
scaling with the log of the naive Bayes conditional probabilities can greatly accelerate the convergence of LR trained using (first-order) Gradient Descent and (approximate second-order) Quasi-Newton optimization. In this work we study
the applicability of the WANBIA-C trick to TRON. We first
devise a TRON algorithm optimizing the softmax objective
function and then demonstrate that WANBIA-C style preconditioning can be beneficial for TRON, leading to an ex-
tremely fast (batch) LR algorithm. Second, we present a
comparative analysis of one-vs-all LR and softmax LR in
terms of the 0-1 Loss, Bias, Variance, RMSE, Log-Loss,
Training and Classification time, and show that softmax LR
leads to significantly better RMSE and Log-Loss. We evaluate our proposed approach on 51 benchmark datasets.
Zaidi, N.; Webb, Geoffrey I.; Carman, M.; Petitjean, F.; Buntine, W.; Hynes, H.; Sterck, H. De
Efficient Parameter Learning of Bayesian Network Classifiers
Machine Learning, vol. 106, no. 9-10, pp. 1289-1329, 2017.
@article{ZaidiEtAl17,
title = {Efficient Parameter Learning of Bayesian Network Classifiers},
author = {N. Zaidi and Geoffrey I. Webb and M. Carman and F. Petitjean and W. Buntine and H. Hynes and H. De Sterck},
url = {http://rdcu.be/oP1t},
doi = {10.1007/s10994-016-5619-z},
year = {2017},
date = {2017-01-01},
journal = {Machine Learning},
volume = {106},
number = {9-10},
pages = {1289-1329},
abstract = {Recent advances have demonstrated substantial benefits from learning with both generative and discriminative parameters. On the one hand, generative approaches address the estimation of the parameters of the joint distribution P(y,x), which for most network types is very computationally efficient (a notable exception to this are Markov networks) and on the other hand, discriminative approaches address the estimation of the parameters of the posterior distribution and, are more effective for classification, since they fit P(y|x) directly. However, discriminative approaches are less computationally efficient as the normalization factor in the conditional log-likelihood precludes the derivation of closed-form estimation of parameters. This paper introduces a new discriminative parameter learning method for Bayesian network classifiers that combines in an elegant fashion parameters learned using both generative and discriminative methods. The proposed method is discriminative in nature, but uses estimates of generative probabilities to speed-up the optimization process. A second contribution is to propose a simple framework to characterize the parameter learning task for Bayesian network classifiers. We conduct an extensive set of experiments on 72 standard datasets and demonstrate that our proposed discriminative parameterization provides an efficient alternative to other state-of-the-art parameterizations.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Recent advances have demonstrated substantial benefits from learning with both generative and discriminative parameters. On the one hand, generative approaches address the estimation of the parameters of the joint distribution P(y,x), which for most network types is very computationally efficient (a notable exception to this are Markov networks) and on the other hand, discriminative approaches address the estimation of the parameters of the posterior distribution and, are more effective for classification, since they fit P(y|x) directly. However, discriminative approaches are less computationally efficient as the normalization factor in the conditional log-likelihood precludes the derivation of closed-form estimation of parameters. This paper introduces a new discriminative parameter learning method for Bayesian network classifiers that combines in an elegant fashion parameters learned using both generative and discriminative methods. The proposed method is discriminative in nature, but uses estimates of generative probabilities to speed-up the optimization process. A second contribution is to propose a simple framework to characterize the parameter learning task for Bayesian network classifiers. We conduct an extensive set of experiments on 72 standard datasets and demonstrate that our proposed discriminative parameterization provides an efficient alternative to other state-of-the-art parameterizations.
Sammut, C.; Webb, G. I. (Ed.)
Encyclopedia of Machine Learning and Data Mining
Springer, Berlin, 2017.
@book{SammutWebb17,
title = {Encyclopedia of Machine Learning and Data Mining},
editor = {C. Sammut and G. I. Webb},
doi = {10.1007/978-1-4899-7502-7},
year = {2017},
date = {2017-01-01},
publisher = {Springer},
address = {Berlin},
keywords = {},
pubstate = {published},
tppubtype = {book}
}
An, Yi; Wang, Jiawei; Li, Chen; Revote, Jerico; Zhang, Yang; Naderer, Thomas; Hayashida, Mirohiro; Akutsu, Tatsuya; Webb, Geoffrey I.; Lithgow, Trevor; Song, Jiangning
Scientific Reports, vol. 7, 2017.
@article{AnEtAl17,
title = {SecretEPDB: a comprehensive web-based resource for secreted effector proteins of the bacterial types III, IV and VI secretion systems},
author = {Yi An and Jiawei Wang and Chen Li and Jerico Revote and Yang Zhang and Thomas Naderer and Mirohiro Hayashida and Tatsuya Akutsu and Geoffrey I. Webb and Trevor Lithgow and Jiangning Song},
url = {http://rdcu.be/oJ9I},
doi = {10.1038/srep41031},
year = {2017},
date = {2017-01-01},
journal = {Scientific Reports},
volume = {7},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ananda-Rajah, Michelle R.; Bergmeir, Christoph; Petitjean, Francois; Slavin, Monica A.; Thursky, Karin A.; Webb, Geoffrey I.
JCO Clinical Cancer Informatics, no. 1, pp. 1-10, 2017.
@article{Ananda-RajahEtAl17,
title = {Toward Electronic Surveillance of Invasive Mold Diseases in Hematology-Oncology Patients: An Expert System Combining Natural Language Processing of Chest Computed Tomography Reports, Microbiology, and Antifungal Drug Data},
author = {Michelle R. Ananda-Rajah and Christoph Bergmeir and Francois Petitjean and Monica A. Slavin and Karin A. Thursky and Geoffrey I. Webb},
doi = {10.1200/CCI.17.00011},
year = {2017},
date = {2017-01-01},
journal = {JCO Clinical Cancer Informatics},
number = {1},
pages = {1-10},
abstract = {Prospective epidemiologic surveillance of invasive mold disease (IMD) in hematology patients is hampered by the absence of a reliable laboratory prompt. This study develops an expert system for electronic surveillance of IMD that combines probabilities using natural language processing (NLP) of computed tomography (CT) reports with microbiology and antifungal drug data to improve prediction of IMD.MethodsMicrobiology indicators and antifungal drug dispensing data were extracted from hospital information systems at three tertiary hospitals for 123 hematology-oncology patients. Of this group, 64 case patients had 26 probable/proven IMD according to international definitions, and 59 patients were uninfected controls. Derived probabilities from NLP combined with medical expertise identified patients at high likelihood of IMD, with remaining patients processed by a machine-learning classifier trained on all available features. Results Compared with the baseline text classifier, the expert system that incorporated the best performing algorithm (naive Bayes) improved specificity from 50.8% (95% CI, 37.5% to 64.1%) to 74.6% (95% CI, 61.6% to 85.0%), reducing false positives by 48% from 29 to 15; improved sensitivity slightly from 96.9% (95% CI, 89.2% to 99.6%) to 98.4% (95% CI, 91.6% to 100%); and improved receiver operating characteristic area from 73.9% (95% CI, 67.1% to 80.6%) to 92.8% (95% CI, 88% to 97.5%). Conclusion An expert system that uses multiple sources of data (CT reports, microbiology, antifungal drug dispensing) is a promising approach to continuous prospective surveillance of IMD in the hospital, and demonstrates reduced false notifications (positives) compared with NLP of CT reports alone. Our expert system could provide decision support for IMD surveillance, which is critical to antifungal stewardship and improving supportive care in cancer.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Prospective epidemiologic surveillance of invasive mold disease (IMD) in hematology patients is hampered by the absence of a reliable laboratory prompt. This study develops an expert system for electronic surveillance of IMD that combines probabilities using natural language processing (NLP) of computed tomography (CT) reports with microbiology and antifungal drug data to improve prediction of IMD.MethodsMicrobiology indicators and antifungal drug dispensing data were extracted from hospital information systems at three tertiary hospitals for 123 hematology-oncology patients. Of this group, 64 case patients had 26 probable/proven IMD according to international definitions, and 59 patients were uninfected controls. Derived probabilities from NLP combined with medical expertise identified patients at high likelihood of IMD, with remaining patients processed by a machine-learning classifier trained on all available features. Results Compared with the baseline text classifier, the expert system that incorporated the best performing algorithm (naive Bayes) improved specificity from 50.8% (95% CI, 37.5% to 64.1%) to 74.6% (95% CI, 61.6% to 85.0%), reducing false positives by 48% from 29 to 15; improved sensitivity slightly from 96.9% (95% CI, 89.2% to 99.6%) to 98.4% (95% CI, 91.6% to 100%); and improved receiver operating characteristic area from 73.9% (95% CI, 67.1% to 80.6%) to 92.8% (95% CI, 88% to 97.5%). Conclusion An expert system that uses multiple sources of data (CT reports, microbiology, antifungal drug dispensing) is a promising approach to continuous prospective surveillance of IMD in the hospital, and demonstrates reduced false notifications (positives) compared with NLP of CT reports alone. Our expert system could provide decision support for IMD surveillance, which is critical to antifungal stewardship and improving supportive care in cancer.
Bergmeir, Christoph; Bilgrami, Irma; Bain, Christopher; Webb, Geoffrey I; Orosz, Judit; Pilcher, David
PLoS ONE, vol. 12, no. 12, 2017.
@article{BergmeirEtAl2017,
title = {Designing a more efficient, effective and safe Medical Emergency Team (MET) service using data analysis},
author = {Christoph Bergmeir and Irma Bilgrami and Christopher Bain and Geoffrey I Webb and Judit Orosz and David Pilcher},
doi = {10.1371/journal.pone.0188688},
year = {2017},
date = {2017-01-01},
journal = {PLoS ONE},
volume = {12},
number = {12},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Chen, S.; Martinez, A.; Webb, G.; Wang, L.
Sample-based Attribute Selective AnDE for Large Data
IEEE Transactions on Knowledge and Data Engineering, vol. 29, no. 1, pp. 172-185, 2017, ISSN: 1041-4347.
@article{ChenEtAl16b,
title = {Sample-based Attribute Selective AnDE for Large Data},
author = {S. Chen and A. Martinez and G. Webb and L. Wang},
doi = {10.1109/TKDE.2016.2608881},
issn = {1041-4347},
year = {2017},
date = {2017-01-01},
journal = {IEEE Transactions on Knowledge and Data Engineering},
volume = {29},
number = {1},
pages = {172-185},
abstract = {More and more applications come with large data sets in the past decade. However, existing algorithms cannot guarantee to scale well on large data. Averaged n-Dependence Estimators (AnDE) allows for flexible learning from out-of-core data, by varying the value of n (number of super parents). Hence AnDE is especially appropriate for large data learning. In this paper, we propose a sample-based attribute selection technique for AnDE. It needs one more pass through the training data, in which a multitude of approximate AnDE models are built and efficiently assessed by leave-one-out cross validation. The use of a sample reduces the training time. Experiments on 15 large data sets demonstrate that the proposed technique significantly reduces AnDE's error at the cost of a modest increase in training time. This efficient and scalable out-of-core approach delivers superior or comparable performance to typical in-core Bayesian network classifiers.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
More and more applications come with large data sets in the past decade. However, existing algorithms cannot guarantee to scale well on large data. Averaged n-Dependence Estimators (AnDE) allows for flexible learning from out-of-core data, by varying the value of n (number of super parents). Hence AnDE is especially appropriate for large data learning. In this paper, we propose a sample-based attribute selection technique for AnDE. It needs one more pass through the training data, in which a multitude of approximate AnDE models are built and efficiently assessed by leave-one-out cross validation. The use of a sample reduces the training time. Experiments on 15 large data sets demonstrate that the proposed technique significantly reduces AnDE's error at the cost of a modest increase in training time. This efficient and scalable out-of-core approach delivers superior or comparable performance to typical in-core Bayesian network classifiers.
Chen, Shenglei; Martinez, Ana M.; Webb, Geoffrey I.; Wang, Limin
Selective AnDE for large data learning: a low-bias memory constrained approach
Knowledge and Information Systems, vol. 50, no. 2, pp. 475-503, 2017, ISSN: 0219-3116.
@article{ChenEtAl16,
title = {Selective AnDE for large data learning: a low-bias memory constrained approach},
author = {Shenglei Chen and Ana M. Martinez and Geoffrey I. Webb and Limin Wang},
doi = {10.1007/s10115-016-0937-9},
issn = {0219-3116},
year = {2017},
date = {2017-01-01},
journal = {Knowledge and Information Systems},
volume = {50},
number = {2},
pages = {475-503},
abstract = {Learning from data that are too big to fit into memory poses great challenges to currently available learning approaches. Averaged n-Dependence Estimators (AnDE) allows for a flexible learning from out-of-core data, by varying the value of n (number of super parents). Hence, AnDE is especially appropriate for learning from large quantities of data. Memory requirement in AnDE, however, increases combinatorially with the number of attributes and the parameter n. In large data learning, number of attributes is often large and we also expect high n to achieve low-bias classification. In order to achieve the lower bias of AnDE with higher n but with less memory requirement, we propose a memory constrained selective AnDE algorithm, in which two passes of learning through training examples are involved. The first pass performs attribute selection on super parents according to available memory, whereas the second one learns an AnDE model with parents only on the selected attributes. Extensive experiments show that the new selective AnDE has considerably lower bias and prediction error relative to A $$n'$$ n ´x DE, where $$n' = n-1$$ n ´x = n - 1 , while maintaining the same space complexity and similar time complexity. The proposed algorithm works well on categorical data. Numerical data sets need to be discretized first.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Learning from data that are too big to fit into memory poses great challenges to currently available learning approaches. Averaged n-Dependence Estimators (AnDE) allows for a flexible learning from out-of-core data, by varying the value of n (number of super parents). Hence, AnDE is especially appropriate for learning from large quantities of data. Memory requirement in AnDE, however, increases combinatorially with the number of attributes and the parameter n. In large data learning, number of attributes is often large and we also expect high n to achieve low-bias classification. In order to achieve the lower bias of AnDE with higher n but with less memory requirement, we propose a memory constrained selective AnDE algorithm, in which two passes of learning through training examples are involved. The first pass performs attribute selection on super parents according to available memory, whereas the second one learns an AnDE model with parents only on the selected attributes. Extensive experiments show that the new selective AnDE has considerably lower bias and prediction error relative to A $$n'$$ n ´x DE, where $$n' = n-1$$ n ´x = n - 1 , while maintaining the same space complexity and similar time complexity. The proposed algorithm works well on categorical data. Numerical data sets need to be discretized first.
Fernando, Thilak L.; Webb, Geoffrey I.
Data Mining and Knowledge Discovery, vol. 31, no. 1, pp. 264-286, 2017.
@article{FernandoWebb16,
title = {SimUSF: an efficient and effective similarity measure that is invariant to violations of the interval scale assumption},
author = {Thilak L. Fernando and Geoffrey I. Webb},
doi = {10.1007/s10618-016-0463-0},
year = {2017},
date = {2017-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {31},
number = {1},
pages = {264-286},
abstract = {Similarity measures are central to many machine learning algorithms. There are many different similarity measures, each catering for different applications and data requirements. Most similarity measures used with numerical data assume that the attributes are interval scale. In the interval scale, it is assumed that a unit difference has the same meaning irrespective of the magnitudes of the values separated. When this assumption is violated, accuracy may be reduced. Our experiments show that removing the interval scale assumption by transforming data to ranks can improve the accuracy of distance-based similarity measures on some tasks. However the rank transform has high time and storage overheads. In this paper, we introduce an efficient similarity measure which does not consider the magnitudes of inter-instance distances. We compare the new similarity measure with popular similarity measures in two applications: DBScan clustering and content based multimedia information retrieval with real world datasets and different transform functions. The results show that the proposed similarity measure provides good performance on a range of tasks and is invariant to violations of the interval scale assumption.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Similarity measures are central to many machine learning algorithms. There are many different similarity measures, each catering for different applications and data requirements. Most similarity measures used with numerical data assume that the attributes are interval scale. In the interval scale, it is assumed that a unit difference has the same meaning irrespective of the magnitudes of the values separated. When this assumption is violated, accuracy may be reduced. Our experiments show that removing the interval scale assumption by transforming data to ranks can improve the accuracy of distance-based similarity measures on some tasks. However the rank transform has high time and storage overheads. In this paper, we introduce an efficient similarity measure which does not consider the magnitudes of inter-instance distances. We compare the new similarity measure with popular similarity measures in two applications: DBScan clustering and content based multimedia information retrieval with real world datasets and different transform functions. The results show that the proposed similarity measure provides good performance on a range of tasks and is invariant to violations of the interval scale assumption.
Forestier, Germain; Petitjean, Francois; Dau, Hoang Anh; Webb, Geoffrey I; Keogh, Eamonn
Generating synthetic time series to augment sparse datasets
IEEE International Conference on Data Mining (ICDM-17), pp. 865-870, 2017.
@inproceedings{ForestierEtAl17,
title = {Generating synthetic time series to augment sparse datasets},
author = {Germain Forestier and Francois Petitjean and Hoang Anh Dau and Geoffrey I Webb and Eamonn Keogh},
year = {2017},
date = {2017-01-01},
booktitle = {IEEE International Conference on Data Mining (ICDM-17)},
pages = {865-870},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Hamalainen, Wilhelmiina; Webb, Geoffrey I
Specious rules: an efficient and effective unifying method for removing misleading and uninformative patterns in association rule mining
Proceedings of the 2017 SIAM International Conference on Data Mining, pp. 309-317, SIAM 2017.
@inproceedings{HamalainenWebb17,
title = {Specious rules: an efficient and effective unifying method for removing misleading and uninformative patterns in association rule mining},
author = {Wilhelmiina Hamalainen and Geoffrey I Webb},
year = {2017},
date = {2017-01-01},
booktitle = {Proceedings of the 2017 SIAM International Conference on Data Mining},
pages = {309-317},
organization = {SIAM},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Li, Fuyi; Li, Chen; Revote, Jerico; Zhang, Yang; Webb, Geoffrey I.; Li, Jian; Song, Jiangning; Lithgow, Trevor
Scientific Reports, vol. 6, 2016.
@article{LiEtAl16,
title = {GlycoMinestruct: a new bioinformatics tool for highly accurate mapping of the human N-linked and O-linked glycoproteomes by incorporating structural features},
author = {Fuyi Li and Chen Li and Jerico Revote and Yang Zhang and Geoffrey I. Webb and Jian Li and Jiangning Song and Trevor Lithgow},
doi = {10.1038/srep34595},
year = {2016},
date = {2016-10-01},
journal = {Scientific Reports},
volume = {6},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Martinez, Ana M.; Webb, Geoffrey I.; Chen, Shenglei; Zaidi, Nayyar A.
Scalable Learning of Bayesian Network Classifiers
Journal of Machine Learning Research, vol. 17, no. 44, pp. 1-35, 2016.
@article{MartinezEtAl16,
title = {Scalable Learning of Bayesian Network Classifiers},
author = {Ana M. Martinez and Geoffrey I. Webb and Shenglei Chen and Nayyar A. Zaidi},
url = {http://jmlr.org/papers/v17/martinez16a.html},
year = {2016},
date = {2016-01-01},
journal = {Journal of Machine Learning Research},
volume = {17},
number = {44},
pages = {1-35},
abstract = {Ever increasing data quantity makes ever more urgent the need for highly scalable learners that have good classification performance. Therefore, an out-of-core learner with excellent time and space complexity, along with high expressivity (that is, capacity to learn very complex multivariate probability distributions) is extremely desirable. This paper presents such a learner. We propose an extension to the k-dependence Bayesian classifier (KDB) that discriminatively selects a sub- model of a full KDB classifier. It requires only one additional pass through the training data, making it a three-pass learner. Our extensive experimental evaluation on 16 large data sets reveals that this out-of-core algorithm achieves competitive classification performance, and substantially better training and classification time than state-of-the-art in-core learners such as random forest and linear and non-linear logistic regression.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ever increasing data quantity makes ever more urgent the need for highly scalable learners that have good classification performance. Therefore, an out-of-core learner with excellent time and space complexity, along with high expressivity (that is, capacity to learn very complex multivariate probability distributions) is extremely desirable. This paper presents such a learner. We propose an extension to the k-dependence Bayesian classifier (KDB) that discriminatively selects a sub- model of a full KDB classifier. It requires only one additional pass through the training data, making it a three-pass learner. Our extensive experimental evaluation on 16 large data sets reveals that this out-of-core algorithm achieves competitive classification performance, and substantially better training and classification time than state-of-the-art in-core learners such as random forest and linear and non-linear logistic regression.
Petitjean, F.; Forestier, G.; Webb, G. I.; Nicholson, A. E.; Chen, Y.; Keogh, E.
Knowledge and Information Systems, vol. 47, no. 1, pp. 1-26, 2016.
@article{PetitjeanEtAl16a,
title = {Faster and more accurate classification of time series by exploiting a novel dynamic time warping averaging algorithm},
author = {F. Petitjean and G. Forestier and G. I. Webb and A. E. Nicholson and Y. Chen and E. Keogh},
doi = {10.1007/s10115-015-0878-8},
year = {2016},
date = {2016-01-01},
journal = {Knowledge and Information Systems},
volume = {47},
number = {1},
pages = {1-26},
abstract = {A concerted research effort over the past two decades has heralded significant improvements in both the efficiency and effectiveness of time series classification. The consensus that has emerged in the community is that the best solution is a surprisingly simple one. In virtually all domains, the most accurate classifier is the nearest neighbor algorithm with dynamic time warping as the distance measure. The time complexity of dynamic time warping means that successful deployments on resource-constrained devices remain elusive. Moreover, the recent explosion of interest in wearable computing devices, which typically have limited computational resources, has greatly increased the need for very efficient classification algorithms. A classic technique to obtain the benefits of the nearest neighbor algorithm, without inheriting its undesirable time and space complexity, is to use the nearest centroid algorithm. Unfortunately, the unique properties of (most) time series data mean that the centroid typically does not resemble any of the instances, an unintuitive and underappreciated fact. In this paper we demonstrate that we can exploit a recent result by Petitjean et al. to allow meaningful averaging of 'warped' time series, which then allows us to create super-efficient nearest 'centroid' classifiers that are at least as accurate as their more computationally challenged nearest neighbor relatives. We demonstrate empirically the utility of our approach by comparing it to all the appropriate strawmen algorithms on the ubiquitous UCR Benchmarks and with a case study in supporting insect classification on resource-constrained sensors.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
A concerted research effort over the past two decades has heralded significant improvements in both the efficiency and effectiveness of time series classification. The consensus that has emerged in the community is that the best solution is a surprisingly simple one. In virtually all domains, the most accurate classifier is the nearest neighbor algorithm with dynamic time warping as the distance measure. The time complexity of dynamic time warping means that successful deployments on resource-constrained devices remain elusive. Moreover, the recent explosion of interest in wearable computing devices, which typically have limited computational resources, has greatly increased the need for very efficient classification algorithms. A classic technique to obtain the benefits of the nearest neighbor algorithm, without inheriting its undesirable time and space complexity, is to use the nearest centroid algorithm. Unfortunately, the unique properties of (most) time series data mean that the centroid typically does not resemble any of the instances, an unintuitive and underappreciated fact. In this paper we demonstrate that we can exploit a recent result by Petitjean et al. to allow meaningful averaging of 'warped' time series, which then allows us to create super-efficient nearest 'centroid' classifiers that are at least as accurate as their more computationally challenged nearest neighbor relatives. We demonstrate empirically the utility of our approach by comparing it to all the appropriate strawmen algorithms on the ubiquitous UCR Benchmarks and with a case study in supporting insect classification on resource-constrained sensors.
Petitjean, Francois; Li, Tao; Tatti, Nikolaj; Webb, Geoffrey I.
Skopus: Mining top-k sequential patterns under leverage
Data Mining and Knowledge Discovery, vol. 30, no. 5, pp. 1086-1111, 2016, ISSN: 1573-756X.
@article{PetitjeanEtAl16b,
title = {Skopus: Mining top-k sequential patterns under leverage},
author = {Francois Petitjean and Tao Li and Nikolaj Tatti and Geoffrey I. Webb},
url = {http://rdcu.be/tsDo},
doi = {10.1007/s10618-016-0467-9},
issn = {1573-756X},
year = {2016},
date = {2016-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {30},
number = {5},
pages = {1086-1111},
abstract = {This paper presents a framework for exact discovery of the top-k sequential patterns under Leverage. It combines (1) a novel definition of the expected support for a sequential pattern\textemdasha concept on which most interestingness measures directly rely\textemdashwith (2) Skopus: a new branch-and-bound algorithm for the exact discovery of top-k sequential patterns under a given measure of interest. Our interestingness measure employs the partition approach. A pattern is interesting to the extent that it is more frequent than can be explained by assuming independence between any of the pairs of patterns from which it can be composed. The larger the support compared to the expectation under independence, the more interesting is the pattern. We build on these two elements to exactly extract the k sequential patterns with highest leverage, consistent with our definition of expected support. We conduct experiments on both synthetic data with known patterns and real-world datasets; both experiments confirm the consistency and relevance of our approach with regard to the state of the art.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper presents a framework for exact discovery of the top-k sequential patterns under Leverage. It combines (1) a novel definition of the expected support for a sequential pattern—a concept on which most interestingness measures directly rely—with (2) Skopus: a new branch-and-bound algorithm for the exact discovery of top-k sequential patterns under a given measure of interest. Our interestingness measure employs the partition approach. A pattern is interesting to the extent that it is more frequent than can be explained by assuming independence between any of the pairs of patterns from which it can be composed. The larger the support compared to the expectation under independence, the more interesting is the pattern. We build on these two elements to exactly extract the k sequential patterns with highest leverage, consistent with our definition of expected support. We conduct experiments on both synthetic data with known patterns and real-world datasets; both experiments confirm the consistency and relevance of our approach with regard to the state of the art.
Petitjean, F.; Webb, G. I.
Scalable Learning of Graphical Models
Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD-16, pp. 2131-2132, ACM Press, 2016.
@inproceedings{PetitjeanWebbTut16,
title = {Scalable Learning of Graphical Models},
author = {F. Petitjean and G. I. Webb},
url = {http://dl.acm.org/authorize?N19101},
year = {2016},
date = {2016-01-01},
booktitle = {Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD-16},
pages = {2131-2132},
publisher = {ACM Press},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Porebski, Benjamin T.; Keleher, Shani; Hollins, Jeffrey J.; Nickson, Adrian A.; Marijanovic, Emilia M.; Borg, Natalie A.; Costa, Mauricio G. S.; Pearce, Mary A.; Dai, Weiwen; Zhu, Liguang; Irving, James A.; Hoke, David E.; Kass, Itamar; Whisstock, James C.; Bottomley, Stephen P.; Webb, Geoffrey I.; McGowan, Sheena; Buckle, Ashley M.
Smoothing a rugged protein folding landscape by sequence-based redesign
Scientific Reports, vol. 6, 2016.
@article{Porebski2016,
title = {Smoothing a rugged protein folding landscape by sequence-based redesign},
author = {Benjamin T. Porebski and Shani Keleher and Jeffrey J. Hollins and Adrian A. Nickson and Emilia M. Marijanovic and Natalie A. Borg and Mauricio G. S. Costa and Mary A. Pearce and Weiwen Dai and Liguang Zhu and James A. Irving and David E. Hoke and Itamar Kass and James C. Whisstock and Stephen P. Bottomley and Geoffrey I. Webb and Sheena McGowan and Ashley M. Buckle},
url = {http://dx.doi.org/10.1038/srep33958},
doi = {10.1038/srep33958},
year = {2016},
date = {2016-01-01},
journal = {Scientific Reports},
volume = {6},
abstract = {The rugged folding landscapes of functional proteins puts them at risk of misfolding and aggregation. Serine protease inhibitors, or serpins, are paradigms for this delicate balance between function and misfolding. Serpins exist in a metastable state that undergoes a major conformational change in order to inhibit proteases. However, conformational labiality of the native serpin fold renders them susceptible to misfolding, which underlies misfolding diseases such as alpha1-antitrypsin deficiency. To investigate how serpins balance function and folding, we used consensus design to create conserpin, a synthetic serpin that folds reversibly, is functional, thermostable, and polymerization resistant. Characterization of its structure, folding and dynamics suggest that consensus design has remodeled the folding landscape to reconcile competing requirements for stability and function. This approach may offer general benefits for engineering functional proteins that have risky folding landscapes, including the removal of aggregation-prone intermediates, and modifying scaffolds for use as protein therapeutics.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The rugged folding landscapes of functional proteins puts them at risk of misfolding and aggregation. Serine protease inhibitors, or serpins, are paradigms for this delicate balance between function and misfolding. Serpins exist in a metastable state that undergoes a major conformational change in order to inhibit proteases. However, conformational labiality of the native serpin fold renders them susceptible to misfolding, which underlies misfolding diseases such as alpha1-antitrypsin deficiency. To investigate how serpins balance function and folding, we used consensus design to create conserpin, a synthetic serpin that folds reversibly, is functional, thermostable, and polymerization resistant. Characterization of its structure, folding and dynamics suggest that consensus design has remodeled the folding landscape to reconcile competing requirements for stability and function. This approach may offer general benefits for engineering functional proteins that have risky folding landscapes, including the removal of aggregation-prone intermediates, and modifying scaffolds for use as protein therapeutics.
Wang, H.; Feng, L.; Zhang, Z.; Webb, G. I.; Lin, D.; Song, J.
Crysalis: an integrated server for computational analysis and design of protein crystallization
Scientific Reports, vol. 6, 2016.
@article{WangEtAl16,
title = {Crysalis: an integrated server for computational analysis and design of protein crystallization},
author = {H. Wang and L. Feng and Z. Zhang and G. I. Webb and D. Lin and J. Song},
doi = {10.1038/srep21383},
year = {2016},
date = {2016-01-01},
journal = {Scientific Reports},
volume = {6},
abstract = {The failure of multi-step experimental procedures to yield diffraction-quality crystals is a major bottleneck in protein structure determination. Accordingly, several bioinformatics methods have been successfully developed and employed to select crystallizable proteins. Unfortunately, the majority of existing in silico methods only allow the prediction of crystallization propensity, seldom enabling computational design of protein mutants that can be targeted for enhancing protein crystallizability. Here, we present Crysalis, an integrated crystallization analysis tool that builds on support-vector regression (SVR) models to facilitate computational protein crystallization prediction, analysis, and design. More specifically, the functionality of this new tool includes: (1) rapid selection of target crystallizable proteins at the proteome level, (2) identification of site non-optimality for protein crystallization and systematic analysis of all potential single-point mutations that might enhance protein crystallization propensity, and (3) annotation of target protein based on predicted structural properties. We applied the design mode of Crysalis to identify site non-optimality for protein crystallization on a proteome-scale, focusing on proteins currently classified as non-crystallizable. Our results revealed that site non-optimality is based on biases related to residues, predicted structures, physicochemical properties, and sequence loci, which provides in-depth understanding of the features influencing protein crystallization. Crysalis is freely available at http://nmrcen.xmu.edu.cn/crysalis/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The failure of multi-step experimental procedures to yield diffraction-quality crystals is a major bottleneck in protein structure determination. Accordingly, several bioinformatics methods have been successfully developed and employed to select crystallizable proteins. Unfortunately, the majority of existing in silico methods only allow the prediction of crystallization propensity, seldom enabling computational design of protein mutants that can be targeted for enhancing protein crystallizability. Here, we present Crysalis, an integrated crystallization analysis tool that builds on support-vector regression (SVR) models to facilitate computational protein crystallization prediction, analysis, and design. More specifically, the functionality of this new tool includes: (1) rapid selection of target crystallizable proteins at the proteome level, (2) identification of site non-optimality for protein crystallization and systematic analysis of all potential single-point mutations that might enhance protein crystallization propensity, and (3) annotation of target protein based on predicted structural properties. We applied the design mode of Crysalis to identify site non-optimality for protein crystallization on a proteome-scale, focusing on proteins currently classified as non-crystallizable. Our results revealed that site non-optimality is based on biases related to residues, predicted structures, physicochemical properties, and sequence loci, which provides in-depth understanding of the features influencing protein crystallization. Crysalis is freely available at http://nmrcen.xmu.edu.cn/crysalis/.
Webb, G. I.; Hyde, R.; Cao, H.; Nguyen, H. L.; Petitjean, F.
Data Mining and Knowledge Discovery, vol. 30, no. 4, pp. 964-994, 2016.
@article{WebbEtAl16,
title = {Characterizing Concept Drift},
author = {G. I. Webb and R. Hyde and H. Cao and H. L. Nguyen and F. Petitjean},
url = {https://rdcu.be/7vMN},
doi = {10.1007/s10618-015-0448-4},
year = {2016},
date = {2016-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {30},
number = {4},
pages = {964-994},
abstract = {Most machine learning models are static, but the world is dynamic, and increasing online deployment of learned models gives increasing urgency to the development of efficient and effective mechanisms to address learning in the context of non-stationary distributions, or as it is commonly called concept drift. However, the key issue of characterizing the different types of drift that can occur has not previously been subjected to rigorous definition and analysis. In particular, while some qualitative drift categorizations have been proposed, few have been formally defined, and the quantitative descriptions required for precise and objective understanding of learner performance have not existed. We present the first comprehensive framework for quantitative analysis of drift. This supports the development of the first comprehensive set of formal definitions of types of concept drift. The formal definitions clarify ambiguities and identify gaps in previous definitions, giving rise to a new comprehensive taxonomy of concept drift types and a solid foundation for research into mechanisms to detect and address concept drift.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Most machine learning models are static, but the world is dynamic, and increasing online deployment of learned models gives increasing urgency to the development of efficient and effective mechanisms to address learning in the context of non-stationary distributions, or as it is commonly called concept drift. However, the key issue of characterizing the different types of drift that can occur has not previously been subjected to rigorous definition and analysis. In particular, while some qualitative drift categorizations have been proposed, few have been formally defined, and the quantitative descriptions required for precise and objective understanding of learner performance have not existed. We present the first comprehensive framework for quantitative analysis of drift. This supports the development of the first comprehensive set of formal definitions of types of concept drift. The formal definitions clarify ambiguities and identify gaps in previous definitions, giving rise to a new comprehensive taxonomy of concept drift types and a solid foundation for research into mechanisms to detect and address concept drift.
Webb, Geoffrey I.; Petitjean, Francois
A multiple test correction for streams and cascades of statistical hypothesis tests
Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD-16, pp. 1255-1264, ACM Press, 2016.
@inproceedings{WebbPetitjean16,
title = {A multiple test correction for streams and cascades of statistical hypothesis tests},
author = {Geoffrey I. Webb and Francois Petitjean},
url = {http://dl.acm.org/authorize?N19100},
doi = {10.1145/2939672.2939775},
year = {2016},
date = {2016-01-01},
booktitle = {Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD-16},
pages = {1255-1264},
publisher = {ACM Press},
abstract = {Statistical hypothesis testing is a popular and powerful tool for inferring knowledge from data. For every such test performed, there is always a non-zero probability of making a false discovery, i.e.~rejecting a null hypothesis in error. Familywise error rate (FWER) is the probability of making at least one false discovery during an inference process. The expected FWER grows exponentially with the number of hypothesis tests that are performed, almost guaranteeing that an error will be committed if the number of tests is big enough and the risk is not managed; a problem known as the multiple testing problem. State-of-the-art methods for controlling FWER in multiple comparison settings require that the set of hypotheses be pre-determined. This greatly hinders statistical testing for many modern applications of statistical inference, such as model selection, because neither the set of hypotheses that will be tested, nor even the number of hypotheses, can be known in advance.
This paper introduces Subfamilywise Multiple Testing, a multiple-testing correction that can be used in applications for which there are repeated pools of null hypotheses from each of which a single null hypothesis is to be rejected and neither the specific hypotheses nor their number are known until the final rejection decision is completed.
To demonstrate the importance and relevance of this work to current machine learning problems, we further refine the theory to the problem of model selection and show how to use Subfamilywise Multiple Testing for learning graphical models.
We assess its ability to discover graphical models on more than 7,000 datasets, studying the ability of Subfamilywise Multiple Testing to outperform the state of the art on data with varying size and dimensionality, as well as with varying density and power of the present correlations. Subfamilywise Multiple Testing provides a significant improvement in statistical efficiency, often requiring only half as much data to discover the same model, while strictly controlling FWER.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Statistical hypothesis testing is a popular and powerful tool for inferring knowledge from data. For every such test performed, there is always a non-zero probability of making a false discovery, i.e.~rejecting a null hypothesis in error. Familywise error rate (FWER) is the probability of making at least one false discovery during an inference process. The expected FWER grows exponentially with the number of hypothesis tests that are performed, almost guaranteeing that an error will be committed if the number of tests is big enough and the risk is not managed; a problem known as the multiple testing problem. State-of-the-art methods for controlling FWER in multiple comparison settings require that the set of hypotheses be pre-determined. This greatly hinders statistical testing for many modern applications of statistical inference, such as model selection, because neither the set of hypotheses that will be tested, nor even the number of hypotheses, can be known in advance.
This paper introduces Subfamilywise Multiple Testing, a multiple-testing correction that can be used in applications for which there are repeated pools of null hypotheses from each of which a single null hypothesis is to be rejected and neither the specific hypotheses nor their number are known until the final rejection decision is completed.
To demonstrate the importance and relevance of this work to current machine learning problems, we further refine the theory to the problem of model selection and show how to use Subfamilywise Multiple Testing for learning graphical models.
We assess its ability to discover graphical models on more than 7,000 datasets, studying the ability of Subfamilywise Multiple Testing to outperform the state of the art on data with varying size and dimensionality, as well as with varying density and power of the present correlations. Subfamilywise Multiple Testing provides a significant improvement in statistical efficiency, often requiring only half as much data to discover the same model, while strictly controlling FWER.
This paper introduces Subfamilywise Multiple Testing, a multiple-testing correction that can be used in applications for which there are repeated pools of null hypotheses from each of which a single null hypothesis is to be rejected and neither the specific hypotheses nor their number are known until the final rejection decision is completed.
To demonstrate the importance and relevance of this work to current machine learning problems, we further refine the theory to the problem of model selection and show how to use Subfamilywise Multiple Testing for learning graphical models.
We assess its ability to discover graphical models on more than 7,000 datasets, studying the ability of Subfamilywise Multiple Testing to outperform the state of the art on data with varying size and dimensionality, as well as with varying density and power of the present correlations. Subfamilywise Multiple Testing provides a significant improvement in statistical efficiency, often requiring only half as much data to discover the same model, while strictly controlling FWER.
Zaidi, Nayyar A.; Petitjean, Francois; Webb, Geoffrey I.
Preconditioning an Artificial Neural Network Using Naive Bayes
Bailey, James; Khan, Latifur; Washio, Takashi; Dobbie, Gill; Huang, Zhexue Joshua; Wang, Ruili (Ed.): Proceedings of the 20th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2016, pp. 341-353, Springer International Publishing, 2016, ISBN: 978-3-319-31753-3.
@inproceedings{ZaidiEtAl16,
title = {Preconditioning an Artificial Neural Network Using Naive Bayes},
author = {Nayyar A. Zaidi and Francois Petitjean and Geoffrey I. Webb},
editor = {James Bailey and Latifur Khan and Takashi Washio and Gill Dobbie and Zhexue Joshua Huang and Ruili Wang},
url = {http://dx.doi.org/10.1007/978-3-319-31753-3_28},
doi = {10.1007/978-3-319-31753-3_28},
isbn = {978-3-319-31753-3},
year = {2016},
date = {2016-01-01},
booktitle = {Proceedings of the 20th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining, PAKDD 2016},
pages = {341-353},
publisher = {Springer International Publishing},
abstract = {Logistic Regression (LR) is a workhorse of the statistics community and a state-of-the-art machine learning classifier. It learns a linear model from inputs to outputs trained by optimizing the Conditional Log-Likelihood (CLL) of the data. Recently, it has been shown that preconditioning LR using a Naive Bayes (NB) model speeds up LR learning many-fold. One can, however, train a linear model by optimizing the mean-square-error (MSE) instead of CLL. This leads to an Artificial Neural Network (ANN) with no hidden layer. In this work, we study the effect of NB preconditioning on such an ANN classifier. Optimizing MSE instead of CLL may lead to a lower bias classifier and hence result in better performance on big datasets. We show that this NB preconditioning can speed-up convergence significantly. We also show that optimizing a linear model with MSE leads to a lower bias classifier than optimizing with CLL. We also compare the performance to state-of-the-art classifier Random Forest.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Logistic Regression (LR) is a workhorse of the statistics community and a state-of-the-art machine learning classifier. It learns a linear model from inputs to outputs trained by optimizing the Conditional Log-Likelihood (CLL) of the data. Recently, it has been shown that preconditioning LR using a Naive Bayes (NB) model speeds up LR learning many-fold. One can, however, train a linear model by optimizing the mean-square-error (MSE) instead of CLL. This leads to an Artificial Neural Network (ANN) with no hidden layer. In this work, we study the effect of NB preconditioning on such an ANN classifier. Optimizing MSE instead of CLL may lead to a lower bias classifier and hence result in better performance on big datasets. We show that this NB preconditioning can speed-up convergence significantly. We also show that optimizing a linear model with MSE leads to a lower bias classifier than optimizing with CLL. We also compare the performance to state-of-the-art classifier Random Forest.
Zaidi, Nayyar A.; Webb, Geoffrey I.; Carman, Mark J.; Petitjean, Francois; Cerquides, Jesus
ALRn: Accelerated higher-order logistic regression
Machine Learning, vol. 104, no. 2, pp. 151-194, 2016, ISSN: 1573-0565.
@article{ZaidiEtAl16b,
title = {ALRn: Accelerated higher-order logistic regression},
author = {Nayyar A. Zaidi and Geoffrey I. Webb and Mark J. Carman and Francois Petitjean and Jesus Cerquides},
doi = {10.1007/s10994-016-5574-8},
issn = {1573-0565},
year = {2016},
date = {2016-01-01},
journal = {Machine Learning},
volume = {104},
number = {2},
pages = {151-194},
abstract = {This paper introduces Accelerated Logistic Regression: a hybrid generative-discriminative approach to training Logistic Regression with high-order features. We present two main results: (1) that our combined generative-discriminative approach significantly improves the efficiency of Logistic Regression and (2) that incorporating higher order features (i.e. features that are the Cartesian products of the original features) reduces the bias of Logistic Regression, which in turn significantly reduces its error on large datasets. We assess the efficacy of Accelerated Logistic Regression by conducting an extensive set of experiments on 75 standard datasets. We demonstrate its competitiveness, particularly on large datasets, by comparing against state-of-the-art classifiers including Random Forest and Averaged n-Dependence Estimators.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper introduces Accelerated Logistic Regression: a hybrid generative-discriminative approach to training Logistic Regression with high-order features. We present two main results: (1) that our combined generative-discriminative approach significantly improves the efficiency of Logistic Regression and (2) that incorporating higher order features (i.e. features that are the Cartesian products of the original features) reduces the bias of Logistic Regression, which in turn significantly reduces its error on large datasets. We assess the efficacy of Accelerated Logistic Regression by conducting an extensive set of experiments on 75 standard datasets. We demonstrate its competitiveness, particularly on large datasets, by comparing against state-of-the-art classifiers including Random Forest and Averaged n-Dependence Estimators.
Zhang, Anshu; Shi, Wenzhong; Webb, Geoffrey I.
Mining significant association rules from uncertain data
Data Mining and Knowledge Discovery, vol. 30, no. 4, pp. 928-963, 2016.
@article{ZhangEtAl16,
title = {Mining significant association rules from uncertain data},
author = {Anshu Zhang and Wenzhong Shi and Geoffrey I. Webb},
doi = {10.1007/s10618-015-0446-6},
year = {2016},
date = {2016-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {30},
number = {4},
pages = {928-963},
publisher = {Springer},
abstract = {In association rule mining, the trade-off between avoiding harmful spurious rules and preserving authentic ones is an ever critical barrier to obtaining reliable and useful results. The statistically sound technique for evaluating statistical significance of association rules is superior in preventing spurious rules, yet can also cause severe loss of true rules in presence of data error. This study presents a new and improved method for statistical test on association rules with uncertain erroneous data. An original mathematical model was established to describe data error propagation through computational procedures of the statistical test. Based on the error model, a scheme combining analytic and simulative processes was designed to correct the statistical test for distortions caused by data error. Experiments on both synthetic and real-world data show that the method significantly recovers the loss in true rules (reduces type-2 error) due to data error occurring in original statistically sound method. Meanwhile, the new method maintains effective control over the familywise error rate, which is the distinctive advantage of the original statistically sound technique. Furthermore, the method is robust against inaccurate data error probability information and situations not fulfilling the commonly accepted assumption on independent error probabilities of different data items. The method is particularly effective for rules which were most practically meaningful yet sensitive to data error. The method proves promising in enhancing values of association rule mining results and helping users make correct decisions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In association rule mining, the trade-off between avoiding harmful spurious rules and preserving authentic ones is an ever critical barrier to obtaining reliable and useful results. The statistically sound technique for evaluating statistical significance of association rules is superior in preventing spurious rules, yet can also cause severe loss of true rules in presence of data error. This study presents a new and improved method for statistical test on association rules with uncertain erroneous data. An original mathematical model was established to describe data error propagation through computational procedures of the statistical test. Based on the error model, a scheme combining analytic and simulative processes was designed to correct the statistical test for distortions caused by data error. Experiments on both synthetic and real-world data show that the method significantly recovers the loss in true rules (reduces type-2 error) due to data error occurring in original statistically sound method. Meanwhile, the new method maintains effective control over the familywise error rate, which is the distinctive advantage of the original statistically sound technique. Furthermore, the method is robust against inaccurate data error probability information and situations not fulfilling the commonly accepted assumption on independent error probabilities of different data items. The method is particularly effective for rules which were most practically meaningful yet sensitive to data error. The method proves promising in enhancing values of association rule mining results and helping users make correct decisions.
Chang, C. C. H.; Li, C.; Webb, G. I.; Tey, B.; Song, J.
Scientific Reports, vol. 6, 2016.
@article{ChangEtAl2016,
title = {Periscope: quantitative prediction of soluble protein expression in the periplasm of Escherichia coli},
author = {C. C. H. Chang and C. Li and G. I. Webb and B. Tey and J. Song},
doi = {10.1038/srep21844},
year = {2016},
date = {2016-01-01},
journal = {Scientific Reports},
volume = {6},
abstract = {Periplasmic expression of soluble proteins in Escherichia coli not only offers a much-simplified downstream purification process, but also enhances the probability of obtaining correctly folded and biologically active proteins. Different combinations of signal peptides and target proteins lead to different soluble protein expression levels, ranging from negligible to several grams per litre. Accurate algorithms for rational selection of promising candidates can serve as a powerful tool to complement with current trial-and-error approaches. Accordingly, proteomics studies can be conducted with greater efficiency and cost-effectiveness. Here, we developed a predictor with a two-stage architecture, to predict the real-valued expression level of target protein in the periplasm. The output of the first-stage support vector machine (SVM) classifier determines which second-stage support vector regression (SVR) classifier to be used. When tested on an independent test dataset, the predictor achieved an overall prediction accuracy of 78% and a Pearson's correlation coefficient (PCC) of 0.77. We further illustrate the relative importance of various features with respect to different models. The results indicate that the occurrence of dipeptide glutamine and aspartic acid is the most important feature for the classification model. Finally, we provide access to the implemented predictor through the Periscope webserver, freely accessible at http://lightning.med.monash.edu/periscope/.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Periplasmic expression of soluble proteins in Escherichia coli not only offers a much-simplified downstream purification process, but also enhances the probability of obtaining correctly folded and biologically active proteins. Different combinations of signal peptides and target proteins lead to different soluble protein expression levels, ranging from negligible to several grams per litre. Accurate algorithms for rational selection of promising candidates can serve as a powerful tool to complement with current trial-and-error approaches. Accordingly, proteomics studies can be conducted with greater efficiency and cost-effectiveness. Here, we developed a predictor with a two-stage architecture, to predict the real-valued expression level of target protein in the periplasm. The output of the first-stage support vector machine (SVM) classifier determines which second-stage support vector regression (SVR) classifier to be used. When tested on an independent test dataset, the predictor achieved an overall prediction accuracy of 78% and a Pearson's correlation coefficient (PCC) of 0.77. We further illustrate the relative importance of various features with respect to different models. The results indicate that the occurrence of dipeptide glutamine and aspartic acid is the most important feature for the classification model. Finally, we provide access to the implemented predictor through the Periscope webserver, freely accessible at http://lightning.med.monash.edu/periscope/.
Li, F.; Li, C.; Wang, M.; Webb, G. I.; Zhang, Y.; Whisstock, J. C.; Song, J.
Bioinformatics, vol. 31, no. 9, pp. 1411-1419, 2015.
@article{LiEtAl15,
title = {GlycoMine: a machine learning-based approach for predicting N-, C- and O-linked glycosylation in the human proteome},
author = {F. Li and C. Li and M. Wang and G. I. Webb and Y. Zhang and J. C. Whisstock and J. Song},
doi = {10.1093/bioinformatics/btu852},
year = {2015},
date = {2015-01-01},
journal = {Bioinformatics},
volume = {31},
number = {9},
pages = {1411-1419},
abstract = {Motivation: Glycosylation is a ubiquitous type of protein post-translational modification (PTM) in eukaryotic cells, which plays vital roles in various biological processes (BPs) such as cellular communication, ligand recognition and subcellular recognition. It is estimated that \>50% of the entire human proteome is glycosylated. However, it is still a significant challenge to identify glycosylation sites, which requires expensive/laborious experimental research. Thus, bioinformatics approaches that can predict the glycan occupancy at specific sequons in protein sequences would be useful for understanding and utilizing this important PTM.
Results: In this study, we present a novel bioinformatics tool called GlycoMine, which is a comprehensive tool for the systematic in silico identification of C-linked, N-linked, and O-linked glycosylation sites in the human proteome. GlycoMine was developed using the random forest algorithm and evaluated based on a well-prepared up-to-date benchmark dataset that encompasses all three types of glycosylation sites, which was curated from multiple public resources. Heterogeneous sequences and functional features were derived from various sources, and subjected to further two-step feature selection to characterize a condensed subset of optimal features that contributed most to the type-specific prediction of glycosylation sites. Five-fold cross-validation and independent tests show that this approach significantly improved the prediction performance compared with four existing prediction tools: NetNGlyc, NetOGlyc, EnsembleGly and GPP. We demonstrated that this tool could identify candidate glycosylation sites in case study proteins and applied it to identify many high-confidence glycosylation target proteins by screening the entire human proteome.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Motivation: Glycosylation is a ubiquitous type of protein post-translational modification (PTM) in eukaryotic cells, which plays vital roles in various biological processes (BPs) such as cellular communication, ligand recognition and subcellular recognition. It is estimated that >50% of the entire human proteome is glycosylated. However, it is still a significant challenge to identify glycosylation sites, which requires expensive/laborious experimental research. Thus, bioinformatics approaches that can predict the glycan occupancy at specific sequons in protein sequences would be useful for understanding and utilizing this important PTM.
Results: In this study, we present a novel bioinformatics tool called GlycoMine, which is a comprehensive tool for the systematic in silico identification of C-linked, N-linked, and O-linked glycosylation sites in the human proteome. GlycoMine was developed using the random forest algorithm and evaluated based on a well-prepared up-to-date benchmark dataset that encompasses all three types of glycosylation sites, which was curated from multiple public resources. Heterogeneous sequences and functional features were derived from various sources, and subjected to further two-step feature selection to characterize a condensed subset of optimal features that contributed most to the type-specific prediction of glycosylation sites. Five-fold cross-validation and independent tests show that this approach significantly improved the prediction performance compared with four existing prediction tools: NetNGlyc, NetOGlyc, EnsembleGly and GPP. We demonstrated that this tool could identify candidate glycosylation sites in case study proteins and applied it to identify many high-confidence glycosylation target proteins by screening the entire human proteome.
Results: In this study, we present a novel bioinformatics tool called GlycoMine, which is a comprehensive tool for the systematic in silico identification of C-linked, N-linked, and O-linked glycosylation sites in the human proteome. GlycoMine was developed using the random forest algorithm and evaluated based on a well-prepared up-to-date benchmark dataset that encompasses all three types of glycosylation sites, which was curated from multiple public resources. Heterogeneous sequences and functional features were derived from various sources, and subjected to further two-step feature selection to characterize a condensed subset of optimal features that contributed most to the type-specific prediction of glycosylation sites. Five-fold cross-validation and independent tests show that this approach significantly improved the prediction performance compared with four existing prediction tools: NetNGlyc, NetOGlyc, EnsembleGly and GPP. We demonstrated that this tool could identify candidate glycosylation sites in case study proteins and applied it to identify many high-confidence glycosylation target proteins by screening the entire human proteome.
Petitjean, F.; Webb, G. I.
Scaling log-linear analysis to datasets with thousands of variables
Proceedings of the 2015 SIAM International Conference on Data Mining, pp. 469-477, 2015.
@inproceedings{PetitjeanWebb15,
title = {Scaling log-linear analysis to datasets with thousands of variables},
author = {F. Petitjean and G. I. Webb},
url = {http://epubs.siam.org/doi/pdf/10.1137/1.9781611974010.53},
year = {2015},
date = {2015-01-01},
booktitle = {Proceedings of the 2015 SIAM International Conference on Data Mining},
pages = {469-477},
abstract = {Association discovery is a fundamental data mining task. The primary statistical approach to association discovery between variables is log-linear analysis. Classical approaches to log-linear analysis do not scale beyond about ten variables. We have recently shown that, if we ensure that the graph supporting the log-linear model is chordal, log-linear analysis can be applied to datasets with hundreds of variables without sacrificing the statistical soundness [21]. However, further scalability remained limited, because state-of-the-art techniques have to examine every edge at every step of the search. This paper makes the following contributions: 1) we prove that only a very small subset of edges has to be considered at each step of the search; 2) we demonstrate how to efficiently find this subset of edges and 3) we show how to efficiently keep track of the best edges to be subsequently added to the initial model. Our experiments, carried out on real datasets with up to 2000 variables, show that our contributions make it possible to gain about 4 orders of magnitude, making log-linear analysis of datasets with thousands of variables possible in seconds instead of days.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Association discovery is a fundamental data mining task. The primary statistical approach to association discovery between variables is log-linear analysis. Classical approaches to log-linear analysis do not scale beyond about ten variables. We have recently shown that, if we ensure that the graph supporting the log-linear model is chordal, log-linear analysis can be applied to datasets with hundreds of variables without sacrificing the statistical soundness [21]. However, further scalability remained limited, because state-of-the-art techniques have to examine every edge at every step of the search. This paper makes the following contributions: 1) we prove that only a very small subset of edges has to be considered at each step of the search; 2) we demonstrate how to efficiently find this subset of edges and 3) we show how to efficiently keep track of the best edges to be subsequently added to the initial model. Our experiments, carried out on real datasets with up to 2000 variables, show that our contributions make it possible to gain about 4 orders of magnitude, making log-linear analysis of datasets with thousands of variables possible in seconds instead of days.
Porebski, B. T.; Nickson, A. A.; Hoke, D. E.; Hunter, M. R.; Zhu, L.; McGowan, S.; Webb, G. I.; Buckle, A. M.
Protein Engineering, Design and Selection, vol. 28, no. 3, pp. 67-78, 2015.
@article{PorebskiEtAl15,
title = {Structural and dynamic properties that govern the stability of an engineered fibronectin type III domain},
author = {B. T. Porebski and A. A. Nickson and D. E. Hoke and M. R. Hunter and L. Zhu and S. McGowan and G. I. Webb and A. M. Buckle},
url = {http://peds.oxfordjournals.org/content/28/3/67.full.pdf+html},
doi = {10.1093/protein/gzv002},
year = {2015},
date = {2015-01-01},
journal = {Protein Engineering, Design and Selection},
volume = {28},
number = {3},
pages = {67-78},
abstract = {Consensus protein design is a rapid and reliable technique for the improvement of protein stability, which relies on the use of homologous protein sequences. To enhance the stability of a fibronectin type III (FN3) domain, consensus design was employed using an alignment of 2123 sequences. The resulting FN3 domain, FN3con, has unprecedented stability, with a melting temperature \>100C, a .GD.N of 15.5 kcal mol.1 and a greatly reduced unfolding rate compared with wild-type. To determine the underlying molecular basis for stability, an X-ray crystal structure of FN3con was determined to 2.0 and compared with other FN3 domains of varying stabilities. The structure of FN3con reveals significantly increased salt bridge interactions that are cooperatively networked, and a highly optimized hydrophobic core. Molecular dynamics simulations of FN3con and comparison structures show the cooperative power of electrostatic and hydrophobic networks in improving FN3con stability. Taken together, our data reveal that FN3con stability does not result from a single mechanism, but rather the combination of several features and the removal of non-conserved, unfavorable interactions. The large number of sequences employed in this study has most likely enhanced the robustness of the consensus design, which is now possible due to the increased sequence availability in the post-genomic era. These studies increase our knowledge of the molecular mechanisms that govern stability and demonstrate the rising potential for enhancing stability via the consensus method.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Consensus protein design is a rapid and reliable technique for the improvement of protein stability, which relies on the use of homologous protein sequences. To enhance the stability of a fibronectin type III (FN3) domain, consensus design was employed using an alignment of 2123 sequences. The resulting FN3 domain, FN3con, has unprecedented stability, with a melting temperature >100C, a .GD.N of 15.5 kcal mol.1 and a greatly reduced unfolding rate compared with wild-type. To determine the underlying molecular basis for stability, an X-ray crystal structure of FN3con was determined to 2.0 and compared with other FN3 domains of varying stabilities. The structure of FN3con reveals significantly increased salt bridge interactions that are cooperatively networked, and a highly optimized hydrophobic core. Molecular dynamics simulations of FN3con and comparison structures show the cooperative power of electrostatic and hydrophobic networks in improving FN3con stability. Taken together, our data reveal that FN3con stability does not result from a single mechanism, but rather the combination of several features and the removal of non-conserved, unfavorable interactions. The large number of sequences employed in this study has most likely enhanced the robustness of the consensus design, which is now possible due to the increased sequence availability in the post-genomic era. These studies increase our knowledge of the molecular mechanisms that govern stability and demonstrate the rising potential for enhancing stability via the consensus method.
Cao, L.; Zhang, C.; Joachims, T.; Webb, G. I.; Margineantu, D. D.; Williams, G. (Ed.)
Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
ACM, 2015.
@proceedings{WebbKDD2015,
title = {Proceedings of the 21st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
editor = {L. Cao and C. Zhang and T. Joachims and G. I. Webb and D. D. Margineantu and G. Williams},
url = {http://dl.acm.org/citation.cfm?id=2783258\&CFID=585807029\&CFTOKEN=47444098},
year = {2015},
date = {2015-01-01},
publisher = {ACM},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Li, Y.; Wang, M.; Wang, H.; Tan, H.; Zhang, Z.; Webb, G. I.; Song, J.
Scientific Reports, vol. 4, 2014.
@article{LiEtAl2014,
title = {Accurate in Silico Identification of Species-Specific Acetylation Sites by Integrating Protein Sequence-Derived and Functional Features},
author = {Y. Li and M. Wang and H. Wang and H. Tan and Z. Zhang and G. I. Webb and J. Song},
doi = {10.1038/srep05765},
year = {2014},
date = {2014-01-01},
journal = {Scientific Reports},
volume = {4},
abstract = {Lysine acetylation is a reversible post-translational modification, playing an important role in cytokine signaling, transcriptional regulation, and apoptosis. To fully understand acetylation mechanisms, identification of substrates and specific acetylation sites is crucial. Experimental identification is often time-consuming and expensive. Alternative bioinformatics methods are cost-effective and can be used in a high-throughput manner to generate relatively precise predictions. Here we develop a method termed as SSPKA for species-specific lysine acetylation prediction, using random forest classifiers that combine sequence-derived and functional features with two-step feature selection. Feature importance analysis indicates functional features, applied for lysine acetylation site prediction for the first time, significantly improve the predictive performance. We apply the SSPKA model to screen the entire human proteome and identify many high-confidence putative substrates that are not previously identified. The results along with the implemented Java tool, serve as useful resources to elucidate the mechanism of lysine acetylation and facilitate hypothesis-driven experimental design and validation.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Lysine acetylation is a reversible post-translational modification, playing an important role in cytokine signaling, transcriptional regulation, and apoptosis. To fully understand acetylation mechanisms, identification of substrates and specific acetylation sites is crucial. Experimental identification is often time-consuming and expensive. Alternative bioinformatics methods are cost-effective and can be used in a high-throughput manner to generate relatively precise predictions. Here we develop a method termed as SSPKA for species-specific lysine acetylation prediction, using random forest classifiers that combine sequence-derived and functional features with two-step feature selection. Feature importance analysis indicates functional features, applied for lysine acetylation site prediction for the first time, significantly improve the predictive performance. We apply the SSPKA model to screen the entire human proteome and identify many high-confidence putative substrates that are not previously identified. The results along with the implemented Java tool, serve as useful resources to elucidate the mechanism of lysine acetylation and facilitate hypothesis-driven experimental design and validation.
Petitjean, F.; Allison, L.; Webb, G. I.
A Statistically Efficient and Scalable Method for Log-Linear Analysis of High-Dimensional Data
Proceedings of the 14th IEEE International Conference on Data Mining, pp. 480-489, 2014.
@inproceedings{PetitjeanEtAl14a,
title = {A Statistically Efficient and Scalable Method for Log-Linear Analysis of High-Dimensional Data},
author = {F. Petitjean and L. Allison and G. I. Webb},
url = {http://dx.doi.org/10.1109/ICDM.2014.23},
year = {2014},
date = {2014-01-01},
booktitle = {Proceedings of the 14th IEEE International Conference on Data Mining},
pages = {480-489},
abstract = {Log-linear analysis is the primary statistical approach to discovering conditional dependencies between the variables of a dataset. A good log-linear analysis method requires both high precision and statistical efficiency. High precision means that the risk of false discoveries should be kept very low. Statistical efficiency means that the method should discover actual associations with as few samples as possible. Classical approaches to log-linear analysis make use of Chi-square tests to control this balance between quality and complexity. We present an information-theoretic approach to log-linear analysis. We show that our approach 1) requires significantly fewer samples to discover the true associations than statistical approaches \textendash statistical efficiency \textendash 2) controls for the risk of false discoveries as well as statistical approaches \textendash high precision - and 3) can perform the discovery on datasets with hundreds of variables on a standard desktop computer \textendash computational efficiency.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Log-linear analysis is the primary statistical approach to discovering conditional dependencies between the variables of a dataset. A good log-linear analysis method requires both high precision and statistical efficiency. High precision means that the risk of false discoveries should be kept very low. Statistical efficiency means that the method should discover actual associations with as few samples as possible. Classical approaches to log-linear analysis make use of Chi-square tests to control this balance between quality and complexity. We present an information-theoretic approach to log-linear analysis. We show that our approach 1) requires significantly fewer samples to discover the true associations than statistical approaches – statistical efficiency – 2) controls for the risk of false discoveries as well as statistical approaches – high precision - and 3) can perform the discovery on datasets with hundreds of variables on a standard desktop computer – computational efficiency.
Petitjean, F.; Forestier, G.; Webb, G. I.; Nicholson, A.; Chen, Y.; Keogh, E.
Dynamic Time Warping Averaging of Time Series Allows Faster and More Accurate Classification
Proceedings of the 14th IEEE International Conference on Data Mining, pp. 470-479, 2014.
@inproceedings{PetitjeanEtAl14b,
title = {Dynamic Time Warping Averaging of Time Series Allows Faster and More Accurate Classification},
author = {F. Petitjean and G. Forestier and G. I. Webb and A. Nicholson and Y. Chen and E. Keogh},
doi = {10.1109/ICDM.2014.27},
year = {2014},
date = {2014-01-01},
booktitle = {Proceedings of the 14th IEEE International Conference on Data Mining},
pages = {470-479},
abstract = {Recent years have seen significant progress in improving both the efficiency and effectiveness of time series classification. However, because the best solution is typically the Nearest Neighbor algorithm with the relatively expensive Dynamic Time Warping as the distance measure, successful deployments on resource constrained devices remain elusive. Moreover, the recent explosion of interest in wearable devices, which typically have limited computational resources, has created a growing need for very efficient classification algorithms. A commonly used technique to glean the benefits of the Nearest Neighbor algorithm, without inheriting its undesirable time complexity, is to use the Nearest Centroid algorithm. However, because of the unique properties of (most) time series data, the centroid typically does not resemble any of the instances, an unintuitive and underappreciated fact. In this work we show that we can exploit a recent result to allow meaningful averaging of 'warped' times series, and that this result allows us to create ultra-efficient Nearest 'Centroid' classifiers that are at least as accurate as their more lethargic Nearest Neighbor cousins.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Recent years have seen significant progress in improving both the efficiency and effectiveness of time series classification. However, because the best solution is typically the Nearest Neighbor algorithm with the relatively expensive Dynamic Time Warping as the distance measure, successful deployments on resource constrained devices remain elusive. Moreover, the recent explosion of interest in wearable devices, which typically have limited computational resources, has created a growing need for very efficient classification algorithms. A commonly used technique to glean the benefits of the Nearest Neighbor algorithm, without inheriting its undesirable time complexity, is to use the Nearest Centroid algorithm. However, because of the unique properties of (most) time series data, the centroid typically does not resemble any of the instances, an unintuitive and underappreciated fact. In this work we show that we can exploit a recent result to allow meaningful averaging of 'warped' times series, and that this result allows us to create ultra-efficient Nearest 'Centroid' classifiers that are at least as accurate as their more lethargic Nearest Neighbor cousins.
Provost, F.; Webb, G. I.; Bekkerman, R.; Etzioni, O.; Fayyad, U.; Perlich, C.
A Data Scientist's Guide to Start-Ups
Big Data, vol. 2, no. 3, pp. 117-128, 2014.
@article{ProvostEtAl14,
title = {A Data Scientist's Guide to Start-Ups},
author = {F. Provost and G. I. Webb and R. Bekkerman and O. Etzioni and U. Fayyad and C. Perlich},
url = {http://dx.doi.org/10.1089/big.2014.0031},
year = {2014},
date = {2014-01-01},
journal = {Big Data},
volume = {2},
number = {3},
pages = {117-128},
abstract = {In August 2013, we held a panel discussion at the KDD 2013 conference in Chicago on the subject of data science, data scientists, and start-ups. KDD is the premier conference on data science research and practice. The panel discussed the pros and cons for top-notch data scientists of the hot data science start-up scene. In this article, we first present background on our panelists. Our four panelists have unquestionable pedigrees in data science and substantial experience with start-ups from multiple perspectives (founders, employees, chief scientists, venture capitalists). For the casual reader, we next present a brief summary of the experts' opinions on eight of the issues the panel discussed. The rest of the article presents a lightly edited transcription of the entire panel discussion.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In August 2013, we held a panel discussion at the KDD 2013 conference in Chicago on the subject of data science, data scientists, and start-ups. KDD is the premier conference on data science research and practice. The panel discussed the pros and cons for top-notch data scientists of the hot data science start-up scene. In this article, we first present background on our panelists. Our four panelists have unquestionable pedigrees in data science and substantial experience with start-ups from multiple perspectives (founders, employees, chief scientists, venture capitalists). For the casual reader, we next present a brief summary of the experts' opinions on eight of the issues the panel discussed. The rest of the article presents a lightly edited transcription of the entire panel discussion.
Hamalainen, Wilhelmiina; Webb, Geoffrey I.
Statistically sound pattern discovery
KDD '14: Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, pp. 1976, 2014.
@inproceedings{HamalainenKDD2014,
title = {Statistically sound pattern discovery},
author = {Wilhelmiina Hamalainen and Geoffrey I. Webb},
doi = {10.1145/2623330.263081},
year = {2014},
date = {2014-01-01},
booktitle = {KDD '14: Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining},
pages = {1976},
abstract = {Pattern discovery is a core data mining activity. Initial approaches were dominated by the frequent pattern discovery paradigm \textendash only patterns that occur frequently in the data were explored. Having been thoroughly researched and its limitations now well understood, this paradigm is giving way to a new one, which can be called statistically sound pattern discovery. In this paradigm, the main impetus is to discover statistically significant patterns, which are unlikely to have occurred by chance and are likely to hold in future data. Thus, the new paradigm provides a strict control over false discoveries and overfitting.
This tutorial covers both classic and cutting-edge research topics on pattern discovery combined to statistical significance testing. We start with an advanced introduction to the relevant forms of statistical significance testing, including different schools and alternative models, their underlying assumptions, practical issues, and limitations. We then discuss their application to data mining specific problems, including evaluation of nested patterns, the multiple testing problem, algorithmic strategies and real-world considerations. We present the current state-of-the art solutions and explore in detail how this approach to pattern discovery can deliver efficient and effective discovery of small sets of interesting patterns.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Pattern discovery is a core data mining activity. Initial approaches were dominated by the frequent pattern discovery paradigm – only patterns that occur frequently in the data were explored. Having been thoroughly researched and its limitations now well understood, this paradigm is giving way to a new one, which can be called statistically sound pattern discovery. In this paradigm, the main impetus is to discover statistically significant patterns, which are unlikely to have occurred by chance and are likely to hold in future data. Thus, the new paradigm provides a strict control over false discoveries and overfitting.
This tutorial covers both classic and cutting-edge research topics on pattern discovery combined to statistical significance testing. We start with an advanced introduction to the relevant forms of statistical significance testing, including different schools and alternative models, their underlying assumptions, practical issues, and limitations. We then discuss their application to data mining specific problems, including evaluation of nested patterns, the multiple testing problem, algorithmic strategies and real-world considerations. We present the current state-of-the art solutions and explore in detail how this approach to pattern discovery can deliver efficient and effective discovery of small sets of interesting patterns.
This tutorial covers both classic and cutting-edge research topics on pattern discovery combined to statistical significance testing. We start with an advanced introduction to the relevant forms of statistical significance testing, including different schools and alternative models, their underlying assumptions, practical issues, and limitations. We then discuss their application to data mining specific problems, including evaluation of nested patterns, the multiple testing problem, algorithmic strategies and real-world considerations. We present the current state-of-the art solutions and explore in detail how this approach to pattern discovery can deliver efficient and effective discovery of small sets of interesting patterns.
Webb, G. I.
Proceedings of the 14th IEEE International Conference on Data Mining, pp. 1031-1036, 2014.
@inproceedings{Webb14,
title = {Contrary to Popular Belief Incremental Discretization can be Sound, Computationally Efficient and Extremely Useful for Streaming Data},
author = {G. I. Webb},
doi = {10.1109/ICDM.2014.123},
year = {2014},
date = {2014-01-01},
booktitle = {Proceedings of the 14th IEEE International Conference on Data Mining},
pages = {1031-1036},
abstract = {Discretization of streaming data has received surprisingly
little attention. This might be because streaming data
require incremental discretization with cutpoints that may vary
over time and this is perceived as undesirable. We argue, to
the contrary, that it can be desirable for a discretization to
evolve in synchronization with an evolving data stream, even
when the learner assumes that attribute values. meanings remain
invariant over time. We examine the issues associated with
discretization in the context of distribution drift and develop
computationally efficient incremental discretization algorithms.
We show that discretization can reduce the error of a classical
incremental learner and that allowing a discretization to drift in
synchronization with distribution drift can further reduce error.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Discretization of streaming data has received surprisingly
little attention. This might be because streaming data
require incremental discretization with cutpoints that may vary
over time and this is perceived as undesirable. We argue, to
the contrary, that it can be desirable for a discretization to
evolve in synchronization with an evolving data stream, even
when the learner assumes that attribute values. meanings remain
invariant over time. We examine the issues associated with
discretization in the context of distribution drift and develop
computationally efficient incremental discretization algorithms.
We show that discretization can reduce the error of a classical
incremental learner and that allowing a discretization to drift in
synchronization with distribution drift can further reduce error.
little attention. This might be because streaming data
require incremental discretization with cutpoints that may vary
over time and this is perceived as undesirable. We argue, to
the contrary, that it can be desirable for a discretization to
evolve in synchronization with an evolving data stream, even
when the learner assumes that attribute values. meanings remain
invariant over time. We examine the issues associated with
discretization in the context of distribution drift and develop
computationally efficient incremental discretization algorithms.
We show that discretization can reduce the error of a classical
incremental learner and that allowing a discretization to drift in
synchronization with distribution drift can further reduce error.
Webb, G. I.; Vreeken, J.
Efficient Discovery of the Most Interesting Associations
ACM Transactions on Knowledge Discovery from Data, vol. 8, no. 3, 2014.
@article{WebbVreeken13,
title = {Efficient Discovery of the Most Interesting Associations},
author = {G. I. Webb and J. Vreeken},
url = {http://dl.acm.org/authorize?N80829},
doi = {10.1145/2601433},
year = {2014},
date = {2014-01-01},
journal = {ACM Transactions on Knowledge Discovery from Data},
volume = {8},
number = {3},
publisher = {ACM},
abstract = {Self-sufficient itemsets have been proposed as an effective approach to summarizing the key associations
in data. However, their computation appears highly demanding, as assessing whether an itemset is selfsufficient
requires consideration of all pairwise partitions of the itemset into pairs of subsets as well as
consideration of all supersets. This paper presents the first published algorithm for efficiently discovering
self-sufficient itemsets. This branch-and-bound algorithm deploys two powerful pruning mechanisms
based on upper-bounds on itemset value and statistical significance level. It demonstrates that finding top-k
productive and non-redundant itemsets, with post processing to identify those that are not independently
productive, can efficiently identify small sets of key associations. We present extensive evaluation of the
strengths and limitations of the technique including comparisons with alternative approaches to finding the
most interesting associations.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Self-sufficient itemsets have been proposed as an effective approach to summarizing the key associations
in data. However, their computation appears highly demanding, as assessing whether an itemset is selfsufficient
requires consideration of all pairwise partitions of the itemset into pairs of subsets as well as
consideration of all supersets. This paper presents the first published algorithm for efficiently discovering
self-sufficient itemsets. This branch-and-bound algorithm deploys two powerful pruning mechanisms
based on upper-bounds on itemset value and statistical significance level. It demonstrates that finding top-k
productive and non-redundant itemsets, with post processing to identify those that are not independently
productive, can efficiently identify small sets of key associations. We present extensive evaluation of the
strengths and limitations of the technique including comparisons with alternative approaches to finding the
most interesting associations.
in data. However, their computation appears highly demanding, as assessing whether an itemset is selfsufficient
requires consideration of all pairwise partitions of the itemset into pairs of subsets as well as
consideration of all supersets. This paper presents the first published algorithm for efficiently discovering
self-sufficient itemsets. This branch-and-bound algorithm deploys two powerful pruning mechanisms
based on upper-bounds on itemset value and statistical significance level. It demonstrates that finding top-k
productive and non-redundant itemsets, with post processing to identify those that are not independently
productive, can efficiently identify small sets of key associations. We present extensive evaluation of the
strengths and limitations of the technique including comparisons with alternative approaches to finding the
most interesting associations.
Zaidi, N.; Carman, M.; Cerquides, J.; Webb, G. I.
Naive-Bayes Inspired Effective Pre-Conditioner for Speeding-up Logistic Regression
Proceedings of the 14th IEEE International Conference on Data Mining, pp. 1097-1102, 2014.
@inproceedings{ZaidiEtAl14,
title = {Naive-Bayes Inspired Effective Pre-Conditioner for Speeding-up Logistic Regression},
author = {N. Zaidi and M. Carman and J. Cerquides and G. I. Webb},
doi = {10.1109/ICDM.2014.53},
year = {2014},
date = {2014-01-01},
booktitle = {Proceedings of the 14th IEEE International Conference on Data Mining},
pages = {1097-1102},
abstract = {We propose an alternative parameterization of
Logistic Regression (LR) for the categorical data, multi-class
setting. LR optimizes the conditional log-likelihood over the
training data and is based on an iterative optimization procedure
to tune this objective function. The optimization procedure
employed may be sensitive to scale and hence an effective
pre-conditioning method is recommended. Many problems in
machine learning involve arbitrary scales or categorical data
(where simple standardization of features is not applicable).
The problem can be alleviated by using optimization routines
that are invariant to scale such as (second-order) Newton
methods. However, computing and inverting the Hessian is a
costly procedure and not feasible for big data. Thus one must
often rely on first-order methods such as gradient descent (GD),
stochastic gradient descent (SGD) or approximate secondorder
such as quasi-Newton (QN) routines, which are not
invariant to scale. This paper proposes a simple yet effective
pre-conditioner for speeding-up LR based on naive Bayes
conditional probability estimates. The idea is to scale each
attribute by the log of the conditional probability of that
attribute given the class. This formulation substantially speeds up
LR's convergence. It also provides a weighted naive Bayes
formulation which yields an effective framework for hybrid
generative-discriminative classification.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We propose an alternative parameterization of
Logistic Regression (LR) for the categorical data, multi-class
setting. LR optimizes the conditional log-likelihood over the
training data and is based on an iterative optimization procedure
to tune this objective function. The optimization procedure
employed may be sensitive to scale and hence an effective
pre-conditioning method is recommended. Many problems in
machine learning involve arbitrary scales or categorical data
(where simple standardization of features is not applicable).
The problem can be alleviated by using optimization routines
that are invariant to scale such as (second-order) Newton
methods. However, computing and inverting the Hessian is a
costly procedure and not feasible for big data. Thus one must
often rely on first-order methods such as gradient descent (GD),
stochastic gradient descent (SGD) or approximate secondorder
such as quasi-Newton (QN) routines, which are not
invariant to scale. This paper proposes a simple yet effective
pre-conditioner for speeding-up LR based on naive Bayes
conditional probability estimates. The idea is to scale each
attribute by the log of the conditional probability of that
attribute given the class. This formulation substantially speeds up
LR's convergence. It also provides a weighted naive Bayes
formulation which yields an effective framework for hybrid
generative-discriminative classification.
Logistic Regression (LR) for the categorical data, multi-class
setting. LR optimizes the conditional log-likelihood over the
training data and is based on an iterative optimization procedure
to tune this objective function. The optimization procedure
employed may be sensitive to scale and hence an effective
pre-conditioning method is recommended. Many problems in
machine learning involve arbitrary scales or categorical data
(where simple standardization of features is not applicable).
The problem can be alleviated by using optimization routines
that are invariant to scale such as (second-order) Newton
methods. However, computing and inverting the Hessian is a
costly procedure and not feasible for big data. Thus one must
often rely on first-order methods such as gradient descent (GD),
stochastic gradient descent (SGD) or approximate secondorder
such as quasi-Newton (QN) routines, which are not
invariant to scale. This paper proposes a simple yet effective
pre-conditioner for speeding-up LR based on naive Bayes
conditional probability estimates. The idea is to scale each
attribute by the log of the conditional probability of that
attribute given the class. This formulation substantially speeds up
LR's convergence. It also provides a weighted naive Bayes
formulation which yields an effective framework for hybrid
generative-discriminative classification.
Chen, S.; Martinez, A.; Webb, G. I.
Highly Scalable Attribute Selection for AODE
Proceedings of the 18th Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. 86-97, 2014.
@inproceedings{ChenEtAl14,
title = {Highly Scalable Attribute Selection for AODE},
author = {S. Chen and A. Martinez and G. I. Webb},
doi = {10.1007/978-3-319-06605-9_8},
year = {2014},
date = {2014-01-01},
booktitle = {Proceedings of the 18th Pacific-Asia Conference on Knowledge Discovery and Data Mining},
pages = {86-97},
abstract = {Averaged One-Dependence Estimators (AODE) is a popular
and effective approach to Bayesian learning. In this paper, a new
attribute selection approach is proposed for AODE. It can search in a
large model space, while it requires only a single extra pass through the
training data, resulting in a computationally efficient two-pass learning
algorithm. The experimental results indicate that the new technique significantly
reduces AODE.s bias at the cost of a modest increase in training
time. Its low bias and computational efficiency make it an attractive
algorithm for learning from big data.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Averaged One-Dependence Estimators (AODE) is a popular
and effective approach to Bayesian learning. In this paper, a new
attribute selection approach is proposed for AODE. It can search in a
large model space, while it requires only a single extra pass through the
training data, resulting in a computationally efficient two-pass learning
algorithm. The experimental results indicate that the new technique significantly
reduces AODE.s bias at the cost of a modest increase in training
time. Its low bias and computational efficiency make it an attractive
algorithm for learning from big data.
and effective approach to Bayesian learning. In this paper, a new
attribute selection approach is proposed for AODE. It can search in a
large model space, while it requires only a single extra pass through the
training data, resulting in a computationally efficient two-pass learning
algorithm. The experimental results indicate that the new technique significantly
reduces AODE.s bias at the cost of a modest increase in training
time. Its low bias and computational efficiency make it an attractive
algorithm for learning from big data.
Petitjean, F.; Webb, G. I.; Nicholson, A. E.
Scaling log-linear analysis to high-dimensional data
Proceedings of the 13th IEEE International Conference on Data Mining, pp. 597-606, 2013.
@inproceedings{PetitjeanEtAl13,
title = {Scaling log-linear analysis to high-dimensional data},
author = {F. Petitjean and G. I. Webb and A. E. Nicholson},
doi = {10.1109/ICDM.2013.17},
year = {2013},
date = {2013-01-01},
booktitle = {Proceedings of the 13th IEEE International Conference on Data Mining},
pages = {597-606},
abstract = {Association discovery is a fundamental data mining task. The primary statistical approach to association discovery between variables is log-linear analysis. Classical approaches to log-linear analysis do not scale beyond about ten variables. We develop an efficient approach to log-linear analysis that scales to hundreds of variables by melding the classical statistical machinery of log-linear analysis with advanced data mining techniques from association discovery and graphical modeling.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Association discovery is a fundamental data mining task. The primary statistical approach to association discovery between variables is log-linear analysis. Classical approaches to log-linear analysis do not scale beyond about ten variables. We develop an efficient approach to log-linear analysis that scales to hundreds of variables by melding the classical statistical machinery of log-linear analysis with advanced data mining techniques from association discovery and graphical modeling.
Provost, F.; Webb, G. I.
Panel: a data scientist's guide to making money from start-ups
Proceedings of the 9th ACM SIGKDD International Conference on knowledge Discovery and Data Mining, pp. 1445-1445, 2013.
@inproceedings{ProvostWebb13,
title = {Panel: a data scientist's guide to making money from start-ups},
author = {F. Provost and G. I. Webb},
url = {http://dl.acm.org/citation.cfm?doid=2487575.2494523},
year = {2013},
date = {2013-01-01},
booktitle = {Proceedings of the 9th ACM SIGKDD International Conference on knowledge Discovery and Data Mining},
pages = {1445-1445},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Suraweera, P.; Webb, G. I.; Evans, I.; Wallace, M.
Learning crew scheduling constraints from historical schedules
Transportation Research Part C: Emerging Technologies, vol. 26, pp. 214-232, 2013.
@article{Suraweera2013,
title = {Learning crew scheduling constraints from historical schedules},
author = {P. Suraweera and G. I. Webb and I. Evans and M. Wallace},
doi = {10.1016/j.trc.2012.08.002},
year = {2013},
date = {2013-01-01},
journal = {Transportation Research Part C: Emerging Technologies},
volume = {26},
pages = {214-232},
abstract = {For most airlines, there are numerous policies, agreements and regulations that govern the workload of airline crew. Although some constraints are formally documented, there are many others based on established practice and tacit understanding. Consequently, the task of developing a formal representation of the constraints that govern the working conditions of an airline's crew requires extensive time and effort involving interviews with the airline's crew schedulers and detailed analysis of historical schedules. We have developed a system that infers crew scheduling constraints from historical crew schedules with the assistance of a domain expert. This system implements the ComCon algorithm developed to learn constraints that prescribe the limits of certain aspects of crew schedules. The algorithm induces complex multivariate constraints based on a set of user provided templates that outline the general structure of important constraints. The results of an evaluation conducted with crew schedules from two commercial airlines show that the system is capable of learning the majority of the minimum rest constraints.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
For most airlines, there are numerous policies, agreements and regulations that govern the workload of airline crew. Although some constraints are formally documented, there are many others based on established practice and tacit understanding. Consequently, the task of developing a formal representation of the constraints that govern the working conditions of an airline's crew requires extensive time and effort involving interviews with the airline's crew schedulers and detailed analysis of historical schedules. We have developed a system that infers crew scheduling constraints from historical crew schedules with the assistance of a domain expert. This system implements the ComCon algorithm developed to learn constraints that prescribe the limits of certain aspects of crew schedules. The algorithm induces complex multivariate constraints based on a set of user provided templates that outline the general structure of important constraints. The results of an evaluation conducted with crew schedules from two commercial airlines show that the system is capable of learning the majority of the minimum rest constraints.
Zaidi, Nayyar A.; Cerquides, Jesus; Carman, Mark J.; Webb, Geoffrey I.
Alleviating Naive Bayes Attribute Independence Assumption by Attribute Weighting
Journal of Machine Learning Research, vol. 14, pp. 1947-1988, 2013.
@article{Zaidi2013,
title = {Alleviating Naive Bayes Attribute Independence Assumption by Attribute Weighting},
author = {Nayyar A. Zaidi and Jesus Cerquides and Mark J. Carman and Geoffrey I. Webb},
url = {http://jmlr.org/papers/volume14/zaidi13a/zaidi13a.pdf},
year = {2013},
date = {2013-01-01},
journal = {Journal of Machine Learning Research},
volume = {14},
pages = {1947-1988},
abstract = {Despite the simplicity of the Naive Bayes classifier, it has continued to perform well against more sophisticated newcomers and has remained, therefore, of great interest to the machine learning community. Of numerous approaches to refining the naive Bayes classifier, attribute weighting has received less attention than it warrants. Most approaches, perhaps influenced by attribute weighting in other machine learning algorithms, use weighting to place more emphasis on highly predictive attributes than those that are less predictive. In this paper, we argue that for naive Bayes attribute weighting should instead be used to alleviate the conditional independence assumption. Based on this premise, we propose a weighted naive Bayes algorithm, called WANBIA, that selects weights to minimize either the negative conditional log likelihood or the mean squared error objective functions. We perform extensive evaluations and find that WANBIA is a competitive alternative to state of the art classifiers like Random Forest, Logistic Regression and A1DE.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Despite the simplicity of the Naive Bayes classifier, it has continued to perform well against more sophisticated newcomers and has remained, therefore, of great interest to the machine learning community. Of numerous approaches to refining the naive Bayes classifier, attribute weighting has received less attention than it warrants. Most approaches, perhaps influenced by attribute weighting in other machine learning algorithms, use weighting to place more emphasis on highly predictive attributes than those that are less predictive. In this paper, we argue that for naive Bayes attribute weighting should instead be used to alleviate the conditional independence assumption. Based on this premise, we propose a weighted naive Bayes algorithm, called WANBIA, that selects weights to minimize either the negative conditional log likelihood or the mean squared error objective functions. We perform extensive evaluations and find that WANBIA is a competitive alternative to state of the art classifiers like Random Forest, Logistic Regression and A1DE.
Zaidi, N.; Webb, G. I.
Fast and Effective Single Pass Bayesian Learning
Proceedings of the 17th Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. 149-160, 2013.
@inproceedings{ZaidiWebb13,
title = {Fast and Effective Single Pass Bayesian Learning},
author = {N. Zaidi and G. I. Webb},
doi = {10.1007/978-3-642-37453-1_13},
year = {2013},
date = {2013-01-01},
booktitle = {Proceedings of the 17th Pacific-Asia Conference on Knowledge Discovery and Data Mining},
pages = {149-160},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Mahmood, K.; Webb, G. I.; Song, J.; Whisstock, J. C.; Konagurthu, A. S.
Nucleic Acids Research, vol. 40, no. 6, 2012.
@article{MahmoodEtAl2012,
title = {Efficient large-scale protein sequence comparison and gene matching to identify orthologs and co-orthologs},
author = {K. Mahmood and G. I. Webb and J. Song and J. C. Whisstock and A. S. Konagurthu},
url = {http://nar.oxfordjournals.org/content/early/2011/12/29/nar.gkr1261.abstract},
doi = {10.1093/nar/gkr1261},
year = {2012},
date = {2012-01-01},
journal = {Nucleic Acids Research},
volume = {40},
number = {6},
publisher = {Oxford Journals},
abstract = {Broadly, computational approaches for ortholog assignment is a three steps process: (i) identify all putative homologs between the genomes, (ii) identify gene anchors and (iii) link anchors to identify best gene matches given their order and context. In this article, we engineer two methods to improve two important aspects of this pipeline [specifically steps (ii) and (iii)]. First, computing sequence similarity data [step (i)] is a computationally intensive task for large sequence sets, creating a bottleneck in the ortholog assignment pipeline. We have designed a fast and highly scalable sort-join method (afree) based on k-mer counts to rapidly compare all pairs of sequences in a large protein sequence set to identify putative homologs. Second, availability of complex genomes containing large gene families with prevalence of complex evolutionary events, such as duplications, has made the task of assigning orthologs and co-orthologs difficult. Here, we have developed an iterative graph matching strategy where at each iteration the best gene assignments are identified resulting in a set of orthologs and co-orthologs. We find that the afree algorithm is faster than existing methods and maintains high accuracy in identifying similar genes. The iterative graph matching strategy also showed high accuracy in identifying complex gene relationships. Standalone afree available from http://vbc.med.monash.edu.au/kmahmood/afree. EGM2, complete ortholog assignment pipeline (including afree and the iterative graph matching method) available from http://vbc.med.monash.edu.au/kmahmood/EGM2.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Broadly, computational approaches for ortholog assignment is a three steps process: (i) identify all putative homologs between the genomes, (ii) identify gene anchors and (iii) link anchors to identify best gene matches given their order and context. In this article, we engineer two methods to improve two important aspects of this pipeline [specifically steps (ii) and (iii)]. First, computing sequence similarity data [step (i)] is a computationally intensive task for large sequence sets, creating a bottleneck in the ortholog assignment pipeline. We have designed a fast and highly scalable sort-join method (afree) based on k-mer counts to rapidly compare all pairs of sequences in a large protein sequence set to identify putative homologs. Second, availability of complex genomes containing large gene families with prevalence of complex evolutionary events, such as duplications, has made the task of assigning orthologs and co-orthologs difficult. Here, we have developed an iterative graph matching strategy where at each iteration the best gene assignments are identified resulting in a set of orthologs and co-orthologs. We find that the afree algorithm is faster than existing methods and maintains high accuracy in identifying similar genes. The iterative graph matching strategy also showed high accuracy in identifying complex gene relationships. Standalone afree available from http://vbc.med.monash.edu.au/kmahmood/afree. EGM2, complete ortholog assignment pipeline (including afree and the iterative graph matching method) available from http://vbc.med.monash.edu.au/kmahmood/EGM2.
Martinez, A.; Webb, G. I.; Flores, M.; Gamez, J.
Non-Disjoint Discretization for Aggregating One-Dependence Estimator Classifiers
Proceedings of the 7th International Conference on Hybrid Artificial Intelligent Systems, pp. 151-162, Springer, Berlin / Heidelberg, 2012, ISBN: 978-3-642-28930-9.
@inproceedings{MartinezEtAl12,
title = {Non-Disjoint Discretization for Aggregating One-Dependence Estimator Classifiers},
author = {A. Martinez and G. I. Webb and M. Flores and J. Gamez},
isbn = {978-3-642-28930-9},
year = {2012},
date = {2012-01-01},
booktitle = {Proceedings of the 7th International Conference on Hybrid Artificial Intelligent Systems},
pages = {151-162},
publisher = {Springer},
address = {Berlin / Heidelberg},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Salem, H.; Suraweera, P.; Webb, G. I.; Boughton, J. R.
Techniques for Efficient Learning without Search
Proceedings of the 16th Pacific-Asia Conference, PAKDD 2012, pp. 50-61, Springer, Kuala Lumpur, Malaysia, 2012.
@inproceedings{SalemEtAl12,
title = {Techniques for Efficient Learning without Search},
author = {H. Salem and P. Suraweera and G. I. Webb and J. R. Boughton},
url = {http://link.springer.com/chapter/10.1007%2F978-3-642-30217-6_5},
year = {2012},
date = {2012-01-01},
booktitle = {Proceedings of the 16th Pacific-Asia Conference, PAKDD 2012},
pages = {50-61},
publisher = {Springer},
address = {Kuala Lumpur, Malaysia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Zaki, Mohammed Javeed; Siebes, Arno; Yu, Jeffrey Xu; Goethals, Bart; Webb, Geoffrey I.; Wu, Xindong (Ed.)
ICDM 2012, The 12th IEEE International Conference on Data Mining
IEEE Computer Society, 2012.
@proceedings{ICDM2012,
title = {ICDM 2012, The 12th IEEE International Conference on Data Mining},
editor = {Mohammed Javeed Zaki and Arno Siebes and Jeffrey Xu Yu and Bart Goethals and Geoffrey I. Webb and Xindong Wu},
url = {https://www.computer.org/csdl/proceedings/icdm/2010/4256/00/index.html},
year = {2012},
date = {2012-01-01},
publisher = {IEEE Computer Society},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Song, J.; Tan, H.; Perry, A. J.; Akutsu, T.; Webb, G. I.; Whisstock, J. C.; Pike, R. N.
PROSPER: An Integrated Feature-Based Tool for Predicting Protease Substrate Cleavage Sites
PLoS ONE, vol. 7, no. 11, 2012.
@article{SongEtAl12b,
title = {PROSPER: An Integrated Feature-Based Tool for Predicting Protease Substrate Cleavage Sites},
author = {J. Song and H. Tan and A. J. Perry and T. Akutsu and G. I. Webb and J. C. Whisstock and R. N. Pike},
doi = {10.1371/journal.pone.0050300},
year = {2012},
date = {2012-01-01},
journal = {PLoS ONE},
volume = {7},
number = {11},
publisher = {Public Library of Science},
abstract = {The ability to catalytically cleave protein substrates after synthesis is fundamental for all forms of life. Accordingly, site-specific proteolysis is one of the most important post-translational modifications. The key to understanding the physiological role of a protease is to identify its natural substrate(s). Knowledge of the substrate specificity of a protease can dramatically improve our ability to predict its target protein substrates, but this information must be utilized in an effective manner in order to efficiently identify protein substrates by \<italic\>in silico\</italic\> approaches. To address this problem, we present PROSPER, an integrated feature-based server for \<italic\>in silico\</italic\> identification of protease substrates and their cleavage sites for twenty-four different proteases. PROSPER utilizes established specificity information for these proteases (derived from the MEROPS database) with a machine learning approach to predict protease cleavage sites by using different, but complementary sequence and structure characteristics. Features used by PROSPER include local amino acid sequence profile, predicted secondary structure, solvent accessibility and predicted native disorder. Thus, for proteases with known amino acid specificity, PROSPER provides a convenient, pre-prepared tool for use in identifying protein substrates for the enzymes. Systematic prediction analysis for the twenty-four proteases thus far included in the database revealed that the features we have included in the tool strongly improve performance in terms of cleavage site prediction, as evidenced by their contribution to performance improvement in terms of identifying known cleavage sites in substrates for these enzymes. In comparison with two state-of-the-art prediction tools, PoPS and SitePrediction, PROSPER achieves greater accuracy and coverage. To our knowledge, PROSPER is the first comprehensive server capable of predicting cleavage sites of multiple proteases within a single substrate sequence using machine learning techniques. It is freely available at \<ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="http://lightning.med.monash.edu.au/PROSPER/" xlink:type="simple"\>http://lightning.med.monash.edu.au/PROSPER/\</ext-link\>.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The ability to catalytically cleave protein substrates after synthesis is fundamental for all forms of life. Accordingly, site-specific proteolysis is one of the most important post-translational modifications. The key to understanding the physiological role of a protease is to identify its natural substrate(s). Knowledge of the substrate specificity of a protease can dramatically improve our ability to predict its target protein substrates, but this information must be utilized in an effective manner in order to efficiently identify protein substrates by <italic>in silico</italic> approaches. To address this problem, we present PROSPER, an integrated feature-based server for <italic>in silico</italic> identification of protease substrates and their cleavage sites for twenty-four different proteases. PROSPER utilizes established specificity information for these proteases (derived from the MEROPS database) with a machine learning approach to predict protease cleavage sites by using different, but complementary sequence and structure characteristics. Features used by PROSPER include local amino acid sequence profile, predicted secondary structure, solvent accessibility and predicted native disorder. Thus, for proteases with known amino acid specificity, PROSPER provides a convenient, pre-prepared tool for use in identifying protein substrates for the enzymes. Systematic prediction analysis for the twenty-four proteases thus far included in the database revealed that the features we have included in the tool strongly improve performance in terms of cleavage site prediction, as evidenced by their contribution to performance improvement in terms of identifying known cleavage sites in substrates for these enzymes. In comparison with two state-of-the-art prediction tools, PoPS and SitePrediction, PROSPER achieves greater accuracy and coverage. To our knowledge, PROSPER is the first comprehensive server capable of predicting cleavage sites of multiple proteases within a single substrate sequence using machine learning techniques. It is freely available at <ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="http://lightning.med.monash.edu.au/PROSPER/" xlink:type="simple">http://lightning.med.monash.edu.au/PROSPER/</ext-link>.
Song, Jiangning; Tan, Hao; Wang, Mingjun; Webb, Geoffrey I.; Akutsu, Tatsuya
PLoS ONE, vol. 7, no. 2, 2012.
@article{SongEtAl12,
title = {TANGLE: Two-Level Support Vector Regression Approach for Protein Backbone Torsion Angle Prediction from Primary Sequences},
author = {Jiangning Song and Hao Tan and Mingjun Wang and Geoffrey I. Webb and Tatsuya Akutsu},
url = {http://dx.doi.org/10.1371%2Fjournal.pone.0030361},
doi = {10.1371/journal.pone.0030361},
year = {2012},
date = {2012-01-01},
journal = {PLoS ONE},
volume = {7},
number = {2},
publisher = {Public Library of Science},
abstract = {Protein backbone torsion angles (Phi) and (Psi) involve two rotation angles rotating around the C_{}-N bond (Phi)
and the C_{}-C bond (Psi). Due to the planarity of the linked rigid peptide bonds, these two angles can essentially determine
the backbone geometry of proteins. Accordingly, the accurate prediction of protein backbone torsion angle from sequence information
can assist the prediction of protein structures. In this study, we develop a new approach called TANGLE (Torsion ANGLE predictor) to
predict the protein backbone torsion angles from amino acid sequences. TANGLE uses a two-level support vector regression approach to
perform real-value torsion angle prediction using a variety of features derived from amino acid sequences, including the evolutionary
profiles in the form of position-specific scoring matrices, predicted secondary structure, solvent accessibility and natively disordered
region as well as other global sequence features. When evaluated based on a large benchmark dataset of 1,526 non-homologous proteins,
the mean absolute errors (MAEs) of the Phi and Psi angle prediction are 27.8 and 44.6, respectively, which are 1% and 3% respectively
lower than that using one of the state-of-the-art prediction tools ANGLOR. Moreover, the prediction of TANGLE is significantly better than a
random predictor that was built on the amino acid-specific basis, with the \<italic\>p\</italic\>-value\<1.46e-147 and 7.97e-150, respectively by the
Wilcoxon signed rank test. As a complementary approach to the current torsion angle prediction algorithms, TANGLE should prove useful in predicting
protein structural properties and assisting protein fold recognition by applying the predicted torsion angles as useful restraints. TANGLE is freely accessible at \<ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="http://sunflower.kuicr.kyoto-u.ac.jp/~sjn/TANGLE/" xlink:type="simple"\>http://sunflower.kuicr.kyoto-u.ac.jp/~sjn/TANGLE/\</ext-link\>.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Protein backbone torsion angles (Phi) and (Psi) involve two rotation angles rotating around the C-N bond (Phi)
and the C-C bond (Psi). Due to the planarity of the linked rigid peptide bonds, these two angles can essentially determine
the backbone geometry of proteins. Accordingly, the accurate prediction of protein backbone torsion angle from sequence information
can assist the prediction of protein structures. In this study, we develop a new approach called TANGLE (Torsion ANGLE predictor) to
predict the protein backbone torsion angles from amino acid sequences. TANGLE uses a two-level support vector regression approach to
perform real-value torsion angle prediction using a variety of features derived from amino acid sequences, including the evolutionary
profiles in the form of position-specific scoring matrices, predicted secondary structure, solvent accessibility and natively disordered
region as well as other global sequence features. When evaluated based on a large benchmark dataset of 1,526 non-homologous proteins,
the mean absolute errors (MAEs) of the Phi and Psi angle prediction are 27.8 and 44.6, respectively, which are 1% and 3% respectively
lower than that using one of the state-of-the-art prediction tools ANGLOR. Moreover, the prediction of TANGLE is significantly better than a
random predictor that was built on the amino acid-specific basis, with the <italic>p</italic>-value<1.46e-147 and 7.97e-150, respectively by the
Wilcoxon signed rank test. As a complementary approach to the current torsion angle prediction algorithms, TANGLE should prove useful in predicting
protein structural properties and assisting protein fold recognition by applying the predicted torsion angles as useful restraints. TANGLE is freely accessible at <ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="http://sunflower.kuicr.kyoto-u.ac.jp/~sjn/TANGLE/" xlink:type="simple">http://sunflower.kuicr.kyoto-u.ac.jp/~sjn/TANGLE/</ext-link>.
and the C-C bond (Psi). Due to the planarity of the linked rigid peptide bonds, these two angles can essentially determine
the backbone geometry of proteins. Accordingly, the accurate prediction of protein backbone torsion angle from sequence information
can assist the prediction of protein structures. In this study, we develop a new approach called TANGLE (Torsion ANGLE predictor) to
predict the protein backbone torsion angles from amino acid sequences. TANGLE uses a two-level support vector regression approach to
perform real-value torsion angle prediction using a variety of features derived from amino acid sequences, including the evolutionary
profiles in the form of position-specific scoring matrices, predicted secondary structure, solvent accessibility and natively disordered
region as well as other global sequence features. When evaluated based on a large benchmark dataset of 1,526 non-homologous proteins,
the mean absolute errors (MAEs) of the Phi and Psi angle prediction are 27.8 and 44.6, respectively, which are 1% and 3% respectively
lower than that using one of the state-of-the-art prediction tools ANGLOR. Moreover, the prediction of TANGLE is significantly better than a
random predictor that was built on the amino acid-specific basis, with the <italic>p</italic>-value<1.46e-147 and 7.97e-150, respectively by the
Wilcoxon signed rank test. As a complementary approach to the current torsion angle prediction algorithms, TANGLE should prove useful in predicting
protein structural properties and assisting protein fold recognition by applying the predicted torsion angles as useful restraints. TANGLE is freely accessible at <ext-link xmlns:xlink="http://www.w3.org/1999/xlink" ext-link-type="uri" xlink:href="http://sunflower.kuicr.kyoto-u.ac.jp/~sjn/TANGLE/" xlink:type="simple">http://sunflower.kuicr.kyoto-u.ac.jp/~sjn/TANGLE/</ext-link>.
Webb, G. I.; Boughton, J.; Zheng, F.; Ting, K. M.; Salem, H.
Machine Learning, vol. 86, no. 2, pp. 233-272, 2012, ISSN: 0885-6125.
@article{WebbEtAl12,
title = {Learning by extrapolation from marginal to full-multivariate probability distributions: Decreasingly naive Bayesian classification},
author = {G. I. Webb and J. Boughton and F. Zheng and K. M. Ting and H. Salem},
url = {http://dx.doi.org/10.1007/s10994-011-5263-6},
issn = {0885-6125},
year = {2012},
date = {2012-01-01},
journal = {Machine Learning},
volume = {86},
number = {2},
pages = {233-272},
publisher = {Springer},
address = {Netherlands},
abstract = {Averaged n-Dependence Estimators (AnDE) is an approach to probabilistic classification learning that learns by extrapolation from marginal
to full-multivariate probability distributions. It utilizes a single parameter that transforms the approach between a low-variance high-bias learner
(Naive Bayes) and a high-variance low-bias learner with Bayes optimal
asymptotic error. It extends the underlying strategy of Averaged One-Dependence Estimators (AODE), which relaxes the Naive Bayes independence assumption while retaining many of Naive Bayes' desirable computational and theoretical properties. AnDE further relaxes the independence assumption by generalizing AODE to higher-levels of dependence.
Extensive experimental evaluation shows that the bias-variance trade-off
for Averaged 2-Dependence Estimators results in strong predictive accuracy over a wide range of data sets. It has training time linear with
respect to the number of examples, supports incremental learning, handles directly missing values, and is robust in the face of noise. Beyond
the practical utility of its lower-dimensional variants, AnDE is of interest
in that it demonstrates that it is possible to create low-bias high-variance
generative learners and suggests strategies for developing even more powerful classifiers.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Averaged n-Dependence Estimators (AnDE) is an approach to probabilistic classification learning that learns by extrapolation from marginal
to full-multivariate probability distributions. It utilizes a single parameter that transforms the approach between a low-variance high-bias learner
(Naive Bayes) and a high-variance low-bias learner with Bayes optimal
asymptotic error. It extends the underlying strategy of Averaged One-Dependence Estimators (AODE), which relaxes the Naive Bayes independence assumption while retaining many of Naive Bayes' desirable computational and theoretical properties. AnDE further relaxes the independence assumption by generalizing AODE to higher-levels of dependence.
Extensive experimental evaluation shows that the bias-variance trade-off
for Averaged 2-Dependence Estimators results in strong predictive accuracy over a wide range of data sets. It has training time linear with
respect to the number of examples, supports incremental learning, handles directly missing values, and is robust in the face of noise. Beyond
the practical utility of its lower-dimensional variants, AnDE is of interest
in that it demonstrates that it is possible to create low-bias high-variance
generative learners and suggests strategies for developing even more powerful classifiers.
to full-multivariate probability distributions. It utilizes a single parameter that transforms the approach between a low-variance high-bias learner
(Naive Bayes) and a high-variance low-bias learner with Bayes optimal
asymptotic error. It extends the underlying strategy of Averaged One-Dependence Estimators (AODE), which relaxes the Naive Bayes independence assumption while retaining many of Naive Bayes' desirable computational and theoretical properties. AnDE further relaxes the independence assumption by generalizing AODE to higher-levels of dependence.
Extensive experimental evaluation shows that the bias-variance trade-off
for Averaged 2-Dependence Estimators results in strong predictive accuracy over a wide range of data sets. It has training time linear with
respect to the number of examples, supports incremental learning, handles directly missing values, and is robust in the face of noise. Beyond
the practical utility of its lower-dimensional variants, AnDE is of interest
in that it demonstrates that it is possible to create low-bias high-variance
generative learners and suggests strategies for developing even more powerful classifiers.
Zheng, F.; Webb, G. I.; Suraweera, P.; Zhu, L.
Subsumption Resolution: An Efficient and Effective Technique for Semi-Naive Bayesian Learning
Machine Learning, vol. 87, no. 1, pp. 93-125, 2012, ISSN: 0885-6125.
@article{ZhengEtAl12,
title = {Subsumption Resolution: An Efficient and Effective Technique for Semi-Naive Bayesian Learning},
author = {F. Zheng and G. I. Webb and P. Suraweera and L. Zhu},
doi = {10.1007/s10994-011-5275-2},
issn = {0885-6125},
year = {2012},
date = {2012-01-01},
journal = {Machine Learning},
volume = {87},
number = {1},
pages = {93-125},
publisher = {Springer},
address = {Netherlands},
abstract = {Semi-naive Bayesian techniques seek to improve the accuracy of naive
Bayes (NB) by relaxing the attribute independence assumption. We present a new
type of semi-naive Bayesian operation, Subsumption Resolution (SR), which efficiently identifies occurrences of the specialization-generalization relationship and
eliminates generalizations at classification time.We extend SR to Near-Subsumption
Resolution (NSR) to delete near.generalizations in addition to generalizations. We
develop two versions of SR: one that performs SR during training, called eager SR
(ESR), and another that performs SR during testing, called lazy SR (LSR).We inves-
tigate the effect of ESR, LSR, NSR and conventional attribute elimination (BSE) on
NB and Averaged One-Dependence Estimators (AODE), a powerful alternative to
NB. BSE imposes very high training time overheads on NB and AODE accompanied
by varying decreases in classification time overheads. ESR, LSR and NSR impose
high training time and test time overheads on NB. However, LSR imposes no extra
training time overheads and only modest test time overheads on AODE, while ESR
and NSR impose modest training and test time overheads on AODE. Our extensive
experimental comparison on sixty UCI data sets shows that applying BSE, LSR or
NSR to NB significantly improves both zero-one loss and RMSE, while applying
BSE, ESR or NSR to AODE significantly improves zero-one loss and RMSE and
applying LSR to AODE significantly improves zero-one loss. The Friedman test and
Nemenyi test show that AODE with ESR or NSR have a significant zero-one loss and
RMSE advantage over Logistic Regression and a zero-one loss advantage overWeka.s
LibSVM implementation with a grid parameter search on categorical data. AODE
with LSR has a zero-one loss advantage over Logistic Regression and comparable
zero-one loss with LibSVM. Finally, we examine the circumstances under which the
elimination of near-generalizations proves beneficial.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Semi-naive Bayesian techniques seek to improve the accuracy of naive
Bayes (NB) by relaxing the attribute independence assumption. We present a new
type of semi-naive Bayesian operation, Subsumption Resolution (SR), which efficiently identifies occurrences of the specialization-generalization relationship and
eliminates generalizations at classification time.We extend SR to Near-Subsumption
Resolution (NSR) to delete near.generalizations in addition to generalizations. We
develop two versions of SR: one that performs SR during training, called eager SR
(ESR), and another that performs SR during testing, called lazy SR (LSR).We inves-
tigate the effect of ESR, LSR, NSR and conventional attribute elimination (BSE) on
NB and Averaged One-Dependence Estimators (AODE), a powerful alternative to
NB. BSE imposes very high training time overheads on NB and AODE accompanied
by varying decreases in classification time overheads. ESR, LSR and NSR impose
high training time and test time overheads on NB. However, LSR imposes no extra
training time overheads and only modest test time overheads on AODE, while ESR
and NSR impose modest training and test time overheads on AODE. Our extensive
experimental comparison on sixty UCI data sets shows that applying BSE, LSR or
NSR to NB significantly improves both zero-one loss and RMSE, while applying
BSE, ESR or NSR to AODE significantly improves zero-one loss and RMSE and
applying LSR to AODE significantly improves zero-one loss. The Friedman test and
Nemenyi test show that AODE with ESR or NSR have a significant zero-one loss and
RMSE advantage over Logistic Regression and a zero-one loss advantage overWeka.s
LibSVM implementation with a grid parameter search on categorical data. AODE
with LSR has a zero-one loss advantage over Logistic Regression and comparable
zero-one loss with LibSVM. Finally, we examine the circumstances under which the
elimination of near-generalizations proves beneficial.
Bayes (NB) by relaxing the attribute independence assumption. We present a new
type of semi-naive Bayesian operation, Subsumption Resolution (SR), which efficiently identifies occurrences of the specialization-generalization relationship and
eliminates generalizations at classification time.We extend SR to Near-Subsumption
Resolution (NSR) to delete near.generalizations in addition to generalizations. We
develop two versions of SR: one that performs SR during training, called eager SR
(ESR), and another that performs SR during testing, called lazy SR (LSR).We inves-
tigate the effect of ESR, LSR, NSR and conventional attribute elimination (BSE) on
NB and Averaged One-Dependence Estimators (AODE), a powerful alternative to
NB. BSE imposes very high training time overheads on NB and AODE accompanied
by varying decreases in classification time overheads. ESR, LSR and NSR impose
high training time and test time overheads on NB. However, LSR imposes no extra
training time overheads and only modest test time overheads on AODE, while ESR
and NSR impose modest training and test time overheads on AODE. Our extensive
experimental comparison on sixty UCI data sets shows that applying BSE, LSR or
NSR to NB significantly improves both zero-one loss and RMSE, while applying
BSE, ESR or NSR to AODE significantly improves zero-one loss and RMSE and
applying LSR to AODE significantly improves zero-one loss. The Friedman test and
Nemenyi test show that AODE with ESR or NSR have a significant zero-one loss and
RMSE advantage over Logistic Regression and a zero-one loss advantage overWeka.s
LibSVM implementation with a grid parameter search on categorical data. AODE
with LSR has a zero-one loss advantage over Logistic Regression and comparable
zero-one loss with LibSVM. Finally, we examine the circumstances under which the
elimination of near-generalizations proves beneficial.
Ng, N. M.; Pierce, J. D.; Webb, G. I.; Ratnikov, B. I.; Wijeyewickrema, L. C.; Duncan, R. C.; Robertson, A. L.; Bottomley, S. P.; Boyd, S. E.; Pike, R. N.
Discovery of Amino Acid Motifs for Thrombin Cleavage and Validation Using a Model Substrate
Biochemistry, vol. 50, no. 48, pp. 10499-10507, 2011.
@article{NgEtAl11,
title = {Discovery of Amino Acid Motifs for Thrombin Cleavage and Validation Using a Model Substrate},
author = {N. M. Ng and J. D. Pierce and G. I. Webb and B. I. Ratnikov and L. C. Wijeyewickrema and R. C. Duncan and A. L. Robertson and S. P. Bottomley and S. E. Boyd and R. N. Pike},
url = {http://pubs.acs.org/doi/abs/10.1021/bi201333g},
doi = {10.1021/bi201333g},
year = {2011},
date = {2011-01-01},
journal = {Biochemistry},
volume = {50},
number = {48},
pages = {10499-10507},
abstract = {Understanding the active site preferences of an enzyme is critical to the design of effective inhibitors and to gaining insights into its mechanisms of action on substrates. While the subsite specificity of thrombin is understood, it is not clear whether the enzyme prefers individual amino acids at each subsite in isolation or prefers to cleave combinations of amino acids as a motif. To investigate whether preferred peptide motifs for cleavage could be identified for thrombin, we exposed a phage-displayed peptide library to thrombin. The resulting preferentially cleaved substrates were analyzed using the technique of association rule discovery. The results revealed that thrombin selected for amino acid motifs in cleavage sites. The contribution of these hypothetical motifs to substrate cleavage efficiency was further investigated using the B1 IgG-binding domain of streptococcal protein G as a model substrate. Introduction of a P2.P1. LRS thrombin cleavage sequence within a major loop of the protein led to cleavage of the protein by thrombin, with the cleavage efficiency increasing with the length of the loop. Introduction of further P3.P1 and P1.P1..P3. amino acid motifs into the loop region yielded greater cleavage efficiencies, suggesting that the susceptibility of a protein substrate to cleavage by thrombin is influenced by these motifs, perhaps because of cooperative effects between subsites closest to the scissile peptide bond.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Understanding the active site preferences of an enzyme is critical to the design of effective inhibitors and to gaining insights into its mechanisms of action on substrates. While the subsite specificity of thrombin is understood, it is not clear whether the enzyme prefers individual amino acids at each subsite in isolation or prefers to cleave combinations of amino acids as a motif. To investigate whether preferred peptide motifs for cleavage could be identified for thrombin, we exposed a phage-displayed peptide library to thrombin. The resulting preferentially cleaved substrates were analyzed using the technique of association rule discovery. The results revealed that thrombin selected for amino acid motifs in cleavage sites. The contribution of these hypothetical motifs to substrate cleavage efficiency was further investigated using the B1 IgG-binding domain of streptococcal protein G as a model substrate. Introduction of a P2.P1. LRS thrombin cleavage sequence within a major loop of the protein led to cleavage of the protein by thrombin, with the cleavage efficiency increasing with the length of the loop. Introduction of further P3.P1 and P1.P1..P3. amino acid motifs into the loop region yielded greater cleavage efficiencies, suggesting that the susceptibility of a protein substrate to cleavage by thrombin is influenced by these motifs, perhaps because of cooperative effects between subsites closest to the scissile peptide bond.
Song, J.; Tan, H.; Boyd, S. E.; Shen, H.; Mahmood, K.; Webb, G. I.; Akutsu, T.; Whisstock, J. C.; Pike, R. N.
Bioinformatic Approaches for Predicting Substrates of Proteases
Journal of Bioinformatics and Computational Biology, vol. 9, no. 1, pp. 149-178, 2011.
@article{SongEtAl11,
title = {Bioinformatic Approaches for Predicting Substrates of Proteases},
author = {J. Song and H. Tan and S. E. Boyd and H. Shen and K. Mahmood and G. I. Webb and T. Akutsu and J. C. Whisstock and R. N. Pike},
doi = {10.1142/S0219720011005288},
year = {2011},
date = {2011-01-01},
journal = {Journal of Bioinformatics and Computational Biology},
volume = {9},
number = {1},
pages = {149-178},
publisher = {World Scientific},
abstract = {Proteases have central roles in "life and death" processes due to their important ability to catalytically hydrolyse protein substrates, usually altering the function and/or activity of the target in the process. Knowledge of the substrate specificity of a protease should, in theory, dramatically improve the ability to predict target protein substrates. However, experimental identification and characterization of protease substrates is often difficult and time-consuming. Thus solving the "substrate identification" problem is fundamental to both understanding protease biology and the development of therapeutics that target specific protease-regulated pathways. In this context, bioinformatic prediction of protease substrates may provide useful and experimentally testable information about novel potential cleavage sites in candidate substrates. In this article, we provide an overview of recent advances in developing bioinformatic approaches for predicting protease substrate cleavage sites and identifying novel putative substrates. We discuss the advantages and drawbacks of the current methods and detail how more accurate models can be built by deriving multiple sequence and structural features of substrates. We also provide some suggestions about how future studies might further improve the accuracy of protease substrate specificity prediction.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Proteases have central roles in "life and death" processes due to their important ability to catalytically hydrolyse protein substrates, usually altering the function and/or activity of the target in the process. Knowledge of the substrate specificity of a protease should, in theory, dramatically improve the ability to predict target protein substrates. However, experimental identification and characterization of protease substrates is often difficult and time-consuming. Thus solving the "substrate identification" problem is fundamental to both understanding protease biology and the development of therapeutics that target specific protease-regulated pathways. In this context, bioinformatic prediction of protease substrates may provide useful and experimentally testable information about novel potential cleavage sites in candidate substrates. In this article, we provide an overview of recent advances in developing bioinformatic approaches for predicting protease substrate cleavage sites and identifying novel putative substrates. We discuss the advantages and drawbacks of the current methods and detail how more accurate models can be built by deriving multiple sequence and structural features of substrates. We also provide some suggestions about how future studies might further improve the accuracy of protease substrate specificity prediction.
Ting, K. M.; Wells, J.; Tan, S.; Teng, S.; Webb, G. I.
Feature-subspace aggregating: Ensembles for stable and unstable learners
Machine Learning, vol. 82, no. 3, pp. 375-397, 2011, ISSN: 0885-6125.
@article{TingEtAl11,
title = {Feature-subspace aggregating: Ensembles for stable and unstable learners},
author = {K. M. Ting and J. Wells and S. Tan and S. Teng and G. I. Webb},
doi = {10.1007/s10994-010-5224-5},
issn = {0885-6125},
year = {2011},
date = {2011-01-01},
journal = {Machine Learning},
volume = {82},
number = {3},
pages = {375-397},
publisher = {Springer},
address = {Netherlands},
abstract = {This paper introduces a new ensemble approach, Feature-Subspace Aggregating (Feating), which builds local models instead of global models. Feating is a generic ensemble approach that can enhance the predictive performance of both stable and unstable learners. In contrast, most existing ensemble approaches can improve the predictive performance of unstable learners only. Our analysis shows that the new approach reduces the execution time to generate a model in an ensemble through an increased level of localisation in Feating. Our empirical evaluation shows that Feating performs significantly better than Boosting, Random Subspace and Bagging in terms of predictive accuracy, when a stable learner SVM is used as the base learner. The speed up achieved by Feating makes feasible SVM ensembles that would otherwise be infeasible for large data sets. When SVM is the preferred base learner, we show that Feating SVM performs better than Boosting decision trees and Random Forests. We further demonstrate that Feating also substantially reduces the error of another stable learner, k-nearest neighbour, and an unstable learner, decision tree.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper introduces a new ensemble approach, Feature-Subspace Aggregating (Feating), which builds local models instead of global models. Feating is a generic ensemble approach that can enhance the predictive performance of both stable and unstable learners. In contrast, most existing ensemble approaches can improve the predictive performance of unstable learners only. Our analysis shows that the new approach reduces the execution time to generate a model in an ensemble through an increased level of localisation in Feating. Our empirical evaluation shows that Feating performs significantly better than Boosting, Random Subspace and Bagging in terms of predictive accuracy, when a stable learner SVM is used as the base learner. The speed up achieved by Feating makes feasible SVM ensembles that would otherwise be infeasible for large data sets. When SVM is the preferred base learner, we show that Feating SVM performs better than Boosting decision trees and Random Forests. We further demonstrate that Feating also substantially reduces the error of another stable learner, k-nearest neighbour, and an unstable learner, decision tree.
Webb, G. I.
Filtered-top-k Association Discovery
WIREs Data Mining and Knowledge Discovery, vol. 1, no. 3, pp. 183-192, 2011.
@article{Webb11,
title = {Filtered-top-k Association Discovery},
author = {G. I. Webb},
doi = {10.1002/widm.28},
year = {2011},
date = {2011-01-01},
journal = {WIREs Data Mining and Knowledge Discovery},
volume = {1},
number = {3},
pages = {183-192},
publisher = {Wiley},
abstract = {Association mining has been one of the most intensively researched areas of data mining. However, direct uptake of the resulting technologies has been relatively low. This paper examines some of the reasons why the dominant paradigms in association mining have not lived up to their promise, and argues that a powerful alternative is provided by top-k techniques coupled with appropriate statistical and other filtering.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Association mining has been one of the most intensively researched areas of data mining. However, direct uptake of the resulting technologies has been relatively low. This paper examines some of the reasons why the dominant paradigms in association mining have not lived up to their promise, and argues that a powerful alternative is provided by top-k techniques coupled with appropriate statistical and other filtering.
Mahmood, K.; Konagurthu, A. S.; Song, J.; Buckle, A. M.; Webb, G. I.; Whisstock, J. C.
Bioinformatics, vol. 26, no. 17, pp. 2076-2084, 2010.
@article{MahmoodEtAl10,
title = {EGM: Encapsulated Gene-by-Gene Matching to Identify Gene Orthologs and Homologous Segments in Genomes},
author = {K. Mahmood and A. S. Konagurthu and J. Song and A. M. Buckle and G. I. Webb and J. C. Whisstock},
doi = {10.1093/bioinformatics/btq339},
year = {2010},
date = {2010-01-01},
journal = {Bioinformatics},
volume = {26},
number = {17},
pages = {2076-2084},
publisher = {Oxford Univ Press},
abstract = {Motivation: Identification of functionally equivalent genes in different species is essential to understand the evolution of biological pathways and processes. At the same time, identification of strings of conserved orthologous genes helps identify complex genomic rearrangements across different organisms. Such an insight is particularly useful, for example, in the transfer of experimental results between different experimental systems such as Drosophila and mammals.
Results: Here we describe the Encapsulated Gene-by-gene Matching (EGM) approach, a method that employs a graph matching strategy to identify gene orthologs and conserved gene segments. Given a pair of genomes, EGM constructs a global gene match for all genes taking into account gene context and family information. The Hungarian method for identifying the maximum weight matching in bipartite graphs is employed, where the resulting matching reveals one-to-one correspondences between nodes (genes) in a manner that maximizes the gene similarity and context.
Conclusion: We tested our approach by performing several comparisons including a detailed Human v Mouse genome mapping. We find that the algorithm is robust and sensitive in detecting orthologs and conserved gene segments. EGM can sensitively detect rearrangements within large and small chromosomal segments. The EGM tool is fully automated and easy to use compared to other more complex methods that also require extensive manual intervention and input.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Motivation: Identification of functionally equivalent genes in different species is essential to understand the evolution of biological pathways and processes. At the same time, identification of strings of conserved orthologous genes helps identify complex genomic rearrangements across different organisms. Such an insight is particularly useful, for example, in the transfer of experimental results between different experimental systems such as Drosophila and mammals.
Results: Here we describe the Encapsulated Gene-by-gene Matching (EGM) approach, a method that employs a graph matching strategy to identify gene orthologs and conserved gene segments. Given a pair of genomes, EGM constructs a global gene match for all genes taking into account gene context and family information. The Hungarian method for identifying the maximum weight matching in bipartite graphs is employed, where the resulting matching reveals one-to-one correspondences between nodes (genes) in a manner that maximizes the gene similarity and context.
Conclusion: We tested our approach by performing several comparisons including a detailed Human v Mouse genome mapping. We find that the algorithm is robust and sensitive in detecting orthologs and conserved gene segments. EGM can sensitively detect rearrangements within large and small chromosomal segments. The EGM tool is fully automated and easy to use compared to other more complex methods that also require extensive manual intervention and input.
Results: Here we describe the Encapsulated Gene-by-gene Matching (EGM) approach, a method that employs a graph matching strategy to identify gene orthologs and conserved gene segments. Given a pair of genomes, EGM constructs a global gene match for all genes taking into account gene context and family information. The Hungarian method for identifying the maximum weight matching in bipartite graphs is employed, where the resulting matching reveals one-to-one correspondences between nodes (genes) in a manner that maximizes the gene similarity and context.
Conclusion: We tested our approach by performing several comparisons including a detailed Human v Mouse genome mapping. We find that the algorithm is robust and sensitive in detecting orthologs and conserved gene segments. EGM can sensitively detect rearrangements within large and small chromosomal segments. The EGM tool is fully automated and easy to use compared to other more complex methods that also require extensive manual intervention and input.
Song, J.; Tan, H.; Shen, H.; Mahmood, K.; Boyd, S. E.; Webb, G. I.; Akutsu, T.; Whisstock, J. C.
Cascleave: Towards More Accurate Prediction of Caspase Substrate Cleavage Sites
Bioinformatics, vol. 26, no. 6, pp. 752-760, 2010.
@article{SongEtAl10,
title = {Cascleave: Towards More Accurate Prediction of Caspase Substrate Cleavage Sites},
author = {J. Song and H. Tan and H. Shen and K. Mahmood and S. E. Boyd and G. I. Webb and T. Akutsu and J. C. Whisstock},
doi = {10.1093/bioinformatics/btq043},
year = {2010},
date = {2010-01-01},
journal = {Bioinformatics},
volume = {26},
number = {6},
pages = {752-760},
publisher = {Oxford Univ Press},
abstract = {Motivation: The caspase family of cysteine proteases play essential roles in key biological processes such as programmed cell death, differentiation, proliferation, necrosis and inflammation. The complete repertoire of caspase substrates remains to be fully characterized. Accordingly, systematic computational screening studies of caspase substrate cleavage sites may provide insight into the substrate specificity of caspases and further facilitating the discovery of putative novel substrates. Results: In this article we develop an approach (termed Cascleave) to predict both classical (i.e. following a P1 Asp) and non-typical caspase cleavage sites. When using local sequence-derived profiles, Cascleave successfully predicted 82.2% of the known substrate cleavage sites, with a Matthews correla tion coefficient (MCC) of 0.667. We found that prediction performance could be further improved by incorporating information such as predicted solvent accessibility and whether a cleavage sequence lies in a region that is most likely natively unstructured. Novel bi-profile Bayesian signatures were found to significantly improve the prediction performance and yielded the best performance with an overall accuracy of 87.6% and a MCC of 0.747, which is higher accuracy than published methods that essentially rely on amino acid sequence alone. It is anticipated that Cascleave will be a powerful tool for predicting novel substrate cleavage sites of caspases and shedding new insights on the unknown caspase-substrate interactivity relationship.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Motivation: The caspase family of cysteine proteases play essential roles in key biological processes such as programmed cell death, differentiation, proliferation, necrosis and inflammation. The complete repertoire of caspase substrates remains to be fully characterized. Accordingly, systematic computational screening studies of caspase substrate cleavage sites may provide insight into the substrate specificity of caspases and further facilitating the discovery of putative novel substrates. Results: In this article we develop an approach (termed Cascleave) to predict both classical (i.e. following a P1 Asp) and non-typical caspase cleavage sites. When using local sequence-derived profiles, Cascleave successfully predicted 82.2% of the known substrate cleavage sites, with a Matthews correla tion coefficient (MCC) of 0.667. We found that prediction performance could be further improved by incorporating information such as predicted solvent accessibility and whether a cleavage sequence lies in a region that is most likely natively unstructured. Novel bi-profile Bayesian signatures were found to significantly improve the prediction performance and yielded the best performance with an overall accuracy of 87.6% and a MCC of 0.747, which is higher accuracy than published methods that essentially rely on amino acid sequence alone. It is anticipated that Cascleave will be a powerful tool for predicting novel substrate cleavage sites of caspases and shedding new insights on the unknown caspase-substrate interactivity relationship.
Webb, G. I.
ACM Transactions on Knowledge Discovery from Data, vol. 4, iss. 1, 2010.
@article{Webb10,
title = {Self-Sufficient Itemsets: An Approach to Screening Potentially Interesting Associations Between Items},
author = {G. I. Webb},
url = {http://dl.acm.org/authorize?270473},
doi = {10.1145/1644873.1644876},
year = {2010},
date = {2010-01-01},
journal = {ACM Transactions on Knowledge Discovery from Data},
volume = {4},
issue = {1},
publisher = {ACM},
abstract = {Self-sufficient itemsets are those whose frequency cannot explained solely by the frequency of either their subsets or of their
supersets. We argue that itemsets that are not
self-sufficient will often be of little interest to the data
analyst, as their frequency should be expected once that of the
itemsets on which their frequency depends is known. We present
statistical tests for statistically sound discovery of
self-sufficient itemsets, and computational techniques that allow
those tests to be applied as a post-processing step for any itemset
discovery algorithm. We also present a measure for assessing the degree of potential interest in an itemset that complements these statistical measures.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Self-sufficient itemsets are those whose frequency cannot explained solely by the frequency of either their subsets or of their
supersets. We argue that itemsets that are not
self-sufficient will often be of little interest to the data
analyst, as their frequency should be expected once that of the
itemsets on which their frequency depends is known. We present
statistical tests for statistically sound discovery of
self-sufficient itemsets, and computational techniques that allow
those tests to be applied as a post-processing step for any itemset
discovery algorithm. We also present a measure for assessing the degree of potential interest in an itemset that complements these statistical measures.
supersets. We argue that itemsets that are not
self-sufficient will often be of little interest to the data
analyst, as their frequency should be expected once that of the
itemsets on which their frequency depends is known. We present
statistical tests for statistically sound discovery of
self-sufficient itemsets, and computational techniques that allow
those tests to be applied as a post-processing step for any itemset
discovery algorithm. We also present a measure for assessing the degree of potential interest in an itemset that complements these statistical measures.
Sammut, C.; Webb, G. I. (Ed.)
Encyclopedia of Machine Learning
Springer, Berlin, 2010.
@book{SammutWebb10,
title = {Encyclopedia of Machine Learning},
editor = {C. Sammut and G. I. Webb},
url = {http://www.springer.com/us/book/9780387307688},
year = {2010},
date = {2010-01-01},
publisher = {Springer},
address = {Berlin},
keywords = {},
pubstate = {published},
tppubtype = {book}
}
Webb, G. I.; Liu, B.; Zhang, C.; Gunopulos, D.; Wu, X. (Ed.)
ICDM 2010, The 10th IEEE International Conference on Data Mining
IEEE Computer Society, 2010.
@proceedings{WebbICDM2010,
title = {ICDM 2010, The 10th IEEE International Conference on Data Mining},
editor = {G. I. Webb and B. Liu and C. Zhang and D. Gunopulos and X. Wu},
url = {https://www.computer.org/csdl/proceedings/icdm/2010/4256/00/index.html},
year = {2010},
date = {2010-01-01},
publisher = {IEEE Computer Society},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Liu, B.; Yang, Y.; Webb, G. I.; Boughton, J.
Proceedings of the 13th Pacific-Asia Conference, PAKDD 2009, pp. 302-313, Springer, Bangkok, Thailand, 2009.
@inproceedings{LiuYangWebbBoughton09,
title = {A Comparative Study of Bandwidth Choice in Kernel Density Estimation for Naive Bayesian Classification},
author = {B. Liu and Y. Yang and G. I. Webb and J. Boughton},
doi = {10.1007/2F978-3-642-01307-2_29},
year = {2009},
date = {2009-01-01},
booktitle = {Proceedings of the 13th Pacific-Asia Conference, PAKDD 2009},
pages = {302-313},
publisher = {Springer},
address = {Bangkok, Thailand},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Novak, P.; Lavrac, N.; Webb, G. I.
Journal of Machine Learning Research, vol. 10, pp. 377-403, 2009.
@article{NovakLavracWebb09,
title = {Supervised Descriptive Rule Discovery: A Unifying Survey of Contrast Set, Emerging Pattern and Subgroup Mining},
author = {P. Novak and N. Lavrac and G. I. Webb},
url = {http://www.jmlr.org/papers/volume10/kralj-novak09a/kralj-novak09a.pdf},
year = {2009},
date = {2009-01-01},
journal = {Journal of Machine Learning Research},
volume = {10},
pages = {377-403},
abstract = {This paper gives a survey of contrast set mining (CSM), emerging pattern mining (EPM), and subgroup discovery (SD) in a unifying framework named supervised descriptive rule discovery. While all these research areas aim at discovering patterns in the form of rules induced from labeled data, they use different terminology and task definitions, claim to have different goals, claim to use different rule learning heuristics, and use different means for selecting subsets of induced patterns. This paper contributes a novel understanding of these subareas of data mining by presenting a unified terminology, by explaining the apparent differences between the learning tasks as variants of a unique supervised descriptive rule discovery task and by exploring the apparent differences between the approaches. It also shows that various rule learning heuristics used in CSM, EPM and SD algorithms all aim at optimizing a trade off between rule coverage and precision. The commonalities (and differences) between the approaches are showcased on a selection of best known variants of CSM, EPM and SD algorithms. The paper also provides a critical survey of existing supervised descriptive rule discovery visualization methods.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper gives a survey of contrast set mining (CSM), emerging pattern mining (EPM), and subgroup discovery (SD) in a unifying framework named supervised descriptive rule discovery. While all these research areas aim at discovering patterns in the form of rules induced from labeled data, they use different terminology and task definitions, claim to have different goals, claim to use different rule learning heuristics, and use different means for selecting subsets of induced patterns. This paper contributes a novel understanding of these subareas of data mining by presenting a unified terminology, by explaining the apparent differences between the learning tasks as variants of a unique supervised descriptive rule discovery task and by exploring the apparent differences between the approaches. It also shows that various rule learning heuristics used in CSM, EPM and SD algorithms all aim at optimizing a trade off between rule coverage and precision. The commonalities (and differences) between the approaches are showcased on a selection of best known variants of CSM, EPM and SD algorithms. The paper also provides a critical survey of existing supervised descriptive rule discovery visualization methods.
Song, J.; Tan, H.; Mahmood, K.; Law, R. H. P.; Buckle, A. M.; Webb, G. I.; Akutsu, T.; Whisstock, J. C.
Prodepth: Predict Residue Depth by Support Vector Regression Approach from Protein Sequences Only
PLoS ONE, vol. 4, no. 9, 2009.
@article{SongEtAl09,
title = {Prodepth: Predict Residue Depth by Support Vector Regression Approach from Protein Sequences Only},
author = {J. Song and H. Tan and K. Mahmood and R. H. P. Law and A. M. Buckle and G. I. Webb and T. Akutsu and J. C. Whisstock},
doi = {10.1371/journal.pone.0007072},
year = {2009},
date = {2009-01-01},
journal = {PLoS ONE},
volume = {4},
number = {9},
publisher = {PLOS},
abstract = {Residue depth (RD) is a solvent exposure measure that complements the information provided by conventional accessible surface area (ASA) and describes to what extent a residue is buried in the protein structure space. Previous studies have established that RD is correlated with several protein properties, such as protein stability, residue conservation and amino acid types. Accurate prediction of RD has many potentially important applications in the field of structural bioinformatics, for example, facilitating the identification of functionally important residues, or residues in the folding nucleus, or enzyme active sites from sequence information. In this work, we introduce an efficient approach that uses support vector regression to quantify the relationship between RD and protein sequence. We systematically investigated eight different sequence encoding schemes including both local and global sequence characteristics and examined their respective prediction performances. For the objective evaluation of our approach, we used 5-fold cross-validation to assess the prediction accuracies and showed that the overall best performance could be achieved with a correlation coefficient (CC) of 0.71 between the observed and predicted RD values and a root mean square error (RMSE) of 1.74, after incorporating the relevant multiple sequence features. The results suggest that residue depth could be reliably predicted solely from protein primary sequences: local sequence environments are the major determinants, while global sequence features could influence the prediction performance marginally. We highlight two examples as a comparison in order to illustrate the applicability of this approach. We also discuss the potential implications of this new structural parameter in the field of protein structure prediction and homology modeling. This method might prove to be a powerful tool for sequence analysis.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Residue depth (RD) is a solvent exposure measure that complements the information provided by conventional accessible surface area (ASA) and describes to what extent a residue is buried in the protein structure space. Previous studies have established that RD is correlated with several protein properties, such as protein stability, residue conservation and amino acid types. Accurate prediction of RD has many potentially important applications in the field of structural bioinformatics, for example, facilitating the identification of functionally important residues, or residues in the folding nucleus, or enzyme active sites from sequence information. In this work, we introduce an efficient approach that uses support vector regression to quantify the relationship between RD and protein sequence. We systematically investigated eight different sequence encoding schemes including both local and global sequence characteristics and examined their respective prediction performances. For the objective evaluation of our approach, we used 5-fold cross-validation to assess the prediction accuracies and showed that the overall best performance could be achieved with a correlation coefficient (CC) of 0.71 between the observed and predicted RD values and a root mean square error (RMSE) of 1.74, after incorporating the relevant multiple sequence features. The results suggest that residue depth could be reliably predicted solely from protein primary sequences: local sequence environments are the major determinants, while global sequence features could influence the prediction performance marginally. We highlight two examples as a comparison in order to illustrate the applicability of this approach. We also discuss the potential implications of this new structural parameter in the field of protein structure prediction and homology modeling. This method might prove to be a powerful tool for sequence analysis.
Ting, K. M.; Wells, J. R.; Tan, S. C.; Teng, S. W.; Webb, G. I.
FaSS: Ensembles for Stable Learners
Proceedings of the 8th International Workshop on Multiple Classifier Systems, MCS 2009, pp. 364-374, Springer, Reykjavik, Iceland, 2009.
@inproceedings{TingEtAl09,
title = {FaSS: Ensembles for Stable Learners},
author = {K. M. Ting and J. R. Wells and S. C. Tan and S. W. Teng and G. I. Webb},
doi = {10.1007/978-3-642-02326-2_37},
year = {2009},
date = {2009-01-01},
booktitle = {Proceedings of the 8th International Workshop on Multiple Classifier Systems, MCS 2009},
pages = {364-374},
publisher = {Springer},
address = {Reykjavik, Iceland},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Yang, Y.; Webb, G. I.
Discretization for Naive-Bayes Learning: Managing Discretization Bias and Variance
Machine Learning, vol. 74, no. 1, pp. 39-74, 2009.
@article{YangWebb09,
title = {Discretization for Naive-Bayes Learning: Managing Discretization Bias and Variance},
author = {Y. Yang and G. I. Webb},
doi = {10.1007/s10994-008-5083-5},
year = {2009},
date = {2009-01-01},
journal = {Machine Learning},
volume = {74},
number = {1},
pages = {39-74},
publisher = {Springer},
address = {Netherlands},
abstract = {Quantitative attributes are usually discretized in Naive-Bayes learning. We establish simple conditions under which discretization is equivalent to use of the true probability density function during naive-Bayes learning. The use of different discretization techniques can be expected to affect the classification bias and variance of generated naive-Bayes classifiers, effects we name discretization bias and variance. We argue that by properly managing discretization bias and variance, we can effectively reduce naive-Bayes classification error. In particular, we supply insights into managing discretization bias and variance by adjusting the number of intervals and the number of training instances contained in each interval. We accordingly propose proportional discretization and fixed frequency discretization, two efficient unsupervised discretization methods that are able to effectively manage discretization bias and variance. We evaluate our new techniques against four key discretization methods for naive-Bayes classifiers. The experimental results support our theoretical analyses by showing that with statistically significant frequency, naive-Bayes classifiers trained on data discretized by our new methods are able to achieve lower classification error than those trained on data discretized by current established discretization methods.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Quantitative attributes are usually discretized in Naive-Bayes learning. We establish simple conditions under which discretization is equivalent to use of the true probability density function during naive-Bayes learning. The use of different discretization techniques can be expected to affect the classification bias and variance of generated naive-Bayes classifiers, effects we name discretization bias and variance. We argue that by properly managing discretization bias and variance, we can effectively reduce naive-Bayes classification error. In particular, we supply insights into managing discretization bias and variance by adjusting the number of intervals and the number of training instances contained in each interval. We accordingly propose proportional discretization and fixed frequency discretization, two efficient unsupervised discretization methods that are able to effectively manage discretization bias and variance. We evaluate our new techniques against four key discretization methods for naive-Bayes classifiers. The experimental results support our theoretical analyses by showing that with statistically significant frequency, naive-Bayes classifiers trained on data discretized by our new methods are able to achieve lower classification error than those trained on data discretized by current established discretization methods.
Hui, B.; Yang, Y.; Webb, G. I.
Anytime Classification for a Pool of Instances
Machine Learning, vol. 77, no. 1, pp. 61-102, 2009.
@article{HuiYangWebb09,
title = {Anytime Classification for a Pool of Instances},
author = {B. Hui and Y. Yang and G. I. Webb},
doi = {10.1007/s10994-009-5118-6},
year = {2009},
date = {2009-01-01},
journal = {Machine Learning},
volume = {77},
number = {1},
pages = {61-102},
publisher = {Springer},
address = {Netherlands},
abstract = {In many real-world applications of classification learning, such as
credit card transaction vetting or classification embedded in sensor
nodes, multiple instances simultaneously require classification under
computational resource constraints such as limited time or limited
battery capacity. In such a situation, available computational
resources should be allocated across the instances in order to
optimize the overall classification efficacy and efficiency. We
propose a novel anytime classification framework, Scheduling Anytime
Averaged Probabilistic Estimators (SAAPE), which is capable of
classifying a pool of instances, delivering accurate results whenever
interrupted and optimizing the collective classification
performance. Following the practice of our previous anytime
classification system AAPE, SAAPE runs a sequence of very efficient
Bayesian probabilistic classifiers to classify each single
instance. Furthermore, SAAPE implements seven alternative scheduling
schemes to decide which instance gets available computational
resources next such that a new classifier can be applied to refine its
classification. We formally present each scheduling scheme's
definition, rationale and time complexity. We conduct large-scale
experiments using 60 benchmark data sets and diversified statistical
tests to evaluate SAAPE's performance on zero-one loss classification
as well as on probability estimation. We analyze each scheduling
scheme's advantage and disadvantage according to both theoretical
understandings and empirical observations. Consequently we identify
effective scheduling schemes that enable SAAPE to accomplish accurate
anytime classification for a pool of instances.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In many real-world applications of classification learning, such as
credit card transaction vetting or classification embedded in sensor
nodes, multiple instances simultaneously require classification under
computational resource constraints such as limited time or limited
battery capacity. In such a situation, available computational
resources should be allocated across the instances in order to
optimize the overall classification efficacy and efficiency. We
propose a novel anytime classification framework, Scheduling Anytime
Averaged Probabilistic Estimators (SAAPE), which is capable of
classifying a pool of instances, delivering accurate results whenever
interrupted and optimizing the collective classification
performance. Following the practice of our previous anytime
classification system AAPE, SAAPE runs a sequence of very efficient
Bayesian probabilistic classifiers to classify each single
instance. Furthermore, SAAPE implements seven alternative scheduling
schemes to decide which instance gets available computational
resources next such that a new classifier can be applied to refine its
classification. We formally present each scheduling scheme's
definition, rationale and time complexity. We conduct large-scale
experiments using 60 benchmark data sets and diversified statistical
tests to evaluate SAAPE's performance on zero-one loss classification
as well as on probability estimation. We analyze each scheduling
scheme's advantage and disadvantage according to both theoretical
understandings and empirical observations. Consequently we identify
effective scheduling schemes that enable SAAPE to accomplish accurate
anytime classification for a pool of instances.
credit card transaction vetting or classification embedded in sensor
nodes, multiple instances simultaneously require classification under
computational resource constraints such as limited time or limited
battery capacity. In such a situation, available computational
resources should be allocated across the instances in order to
optimize the overall classification efficacy and efficiency. We
propose a novel anytime classification framework, Scheduling Anytime
Averaged Probabilistic Estimators (SAAPE), which is capable of
classifying a pool of instances, delivering accurate results whenever
interrupted and optimizing the collective classification
performance. Following the practice of our previous anytime
classification system AAPE, SAAPE runs a sequence of very efficient
Bayesian probabilistic classifiers to classify each single
instance. Furthermore, SAAPE implements seven alternative scheduling
schemes to decide which instance gets available computational
resources next such that a new classifier can be applied to refine its
classification. We formally present each scheduling scheme's
definition, rationale and time complexity. We conduct large-scale
experiments using 60 benchmark data sets and diversified statistical
tests to evaluate SAAPE's performance on zero-one loss classification
as well as on probability estimation. We analyze each scheduling
scheme's advantage and disadvantage according to both theoretical
understandings and empirical observations. Consequently we identify
effective scheduling schemes that enable SAAPE to accomplish accurate
anytime classification for a pool of instances.
Webb, G. I.
Layered Critical Values: A Powerful Direct-Adjustment Approach to Discovering Significant Patterns
Machine Learning, vol. 71, no. 2-3, pp. 307-323, 2008.
@article{Webb08,
title = {Layered Critical Values: A Powerful Direct-Adjustment Approach to Discovering Significant Patterns},
author = {G. I. Webb},
doi = {10.1007/s10994-008-5046-x},
year = {2008},
date = {2008-01-01},
journal = {Machine Learning},
volume = {71},
number = {2-3},
pages = {307-323},
publisher = {Springer},
address = {Netherlands},
abstract = {Standard pattern discovery techniques, such as association rules, suffer an extreme risk of finding very large numbers of spurious patterns for many knowledge discovery tasks. The direct-adjustment approach to controlling this risk applies a statistical test during the discovery process, using a critical value adjusted to take account of the size of the search space. However, a problem with the direct-adjustment strategy is that it may discard numerous true patterns. This paper investigates the assignment of different critical values to different areas of the search space as an approach to alleviating this problem, using a variant of a technique originally developed for other purposes. This approach is shown to be effective at increasing the number of discoveries while still maintaining strict control over the risk of false discoveries.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Standard pattern discovery techniques, such as association rules, suffer an extreme risk of finding very large numbers of spurious patterns for many knowledge discovery tasks. The direct-adjustment approach to controlling this risk applies a statistical test during the discovery process, using a critical value adjusted to take account of the size of the search space. However, a problem with the direct-adjustment strategy is that it may discard numerous true patterns. This paper investigates the assignment of different critical values to different areas of the search space as an approach to alleviating this problem, using a variant of a technique originally developed for other purposes. This approach is shown to be effective at increasing the number of discoveries while still maintaining strict control over the risk of false discoveries.
Webb, G. I.
Multi-Strategy Ensemble Learning, Ensembles of Bayesian Classifiers, and the Problem of False Discoveries
Proceedings of the Seventh Australasian Data Mining Conference (AusDM 2008), pp. 15, Australian Computer Society, Adelaide, Australia, 2008.
@inproceedings{Webb08b,
title = {Multi-Strategy Ensemble Learning, Ensembles of Bayesian Classifiers, and the Problem of False Discoveries},
author = {G. I. Webb},
year = {2008},
date = {2008-01-01},
booktitle = {Proceedings of the Seventh Australasian Data Mining Conference (AusDM 2008)},
pages = {15},
publisher = {Australian Computer Society},
address = {Adelaide, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Webb, G. I.
Finding the Real Patterns (Extended Abstract)
Zhou, Zhi-Hua; Li, Hang; Yang, Qiang (Ed.): Lecture Notes in Computer Science Vol. 4426 : Advances in Knowledge Discovery and Data Mining Proceedings of the 11th Pacific-Asia Conference, PAKDD 2007, pp. 6, Springer, Nanjing, China, 2007.
@inproceedings{Webb07a,
title = {Finding the Real Patterns (Extended Abstract)},
author = {G. I. Webb},
editor = {Zhi-Hua Zhou and Hang Li and Qiang Yang},
year = {2007},
date = {2007-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 4426 : Advances in Knowledge Discovery and Data Mining Proceedings of the 11th Pacific-Asia Conference, PAKDD 2007},
pages = {6},
publisher = {Springer},
address = {Nanjing, China},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Webb, G. I.
Discovering Significant Patterns
Machine Learning, vol. 68, no. 1, pp. 1-33, 2007.
@article{Webb07,
title = {Discovering Significant Patterns},
author = {G. I. Webb},
doi = {10.1007/s10994-007-5006-x},
year = {2007},
date = {2007-01-01},
journal = {Machine Learning},
volume = {68},
number = {1},
pages = {1-33},
publisher = {Springer},
address = {Netherlands},
abstract = {Exploratory pattern discovery techniques, such as association rule discovery, explore large search spaces of potential patterns to find those that satisfy some user-specified constraints. Due to the large number of patterns considered, they suffer from an extreme risk of type-1 error, that is, of finding patterns that appear due to chance alone to satisfy the constraints on the sample data. This paper proposes techniques to overcome this problem by applying well-established statistical practices. These allow the user to enforce a strict upper limit on the risk of experimentwise error. Empirical studies demonstrate that standard exploratory pattern discovery techniques can discover numerous spurious patterns when applied to random data and when applied to real-world data result in large numbers of patterns that are rejected when subjected to statistical evaluation on holdout data. They also reveal that modification of the pattern discovery process to anticipate subsequent statistical evaluation can increase the number of patterns that are accepted by statistical evaluation on holdout data.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Exploratory pattern discovery techniques, such as association rule discovery, explore large search spaces of potential patterns to find those that satisfy some user-specified constraints. Due to the large number of patterns considered, they suffer from an extreme risk of type-1 error, that is, of finding patterns that appear due to chance alone to satisfy the constraints on the sample data. This paper proposes techniques to overcome this problem by applying well-established statistical practices. These allow the user to enforce a strict upper limit on the risk of experimentwise error. Empirical studies demonstrate that standard exploratory pattern discovery techniques can discover numerous spurious patterns when applied to random data and when applied to real-world data result in large numbers of patterns that are rejected when subjected to statistical evaluation on holdout data. They also reveal that modification of the pattern discovery process to anticipate subsequent statistical evaluation can increase the number of patterns that are accepted by statistical evaluation on holdout data.
Webb, G. I.
Tenth Anniversary Edition Editorial
Data Mining and Knowledge Discovery, vol. 15, no. 1, pp. 1-2, 2007.
@article{Webb07b,
title = {Tenth Anniversary Edition Editorial},
author = {G. I. Webb},
doi = {10.1007/s10618-007-0075-9},
year = {2007},
date = {2007-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {15},
number = {1},
pages = {1-2},
publisher = {Springer},
address = {Netherlands},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Yang, Y.; Webb, G. I.; Cerquides, J.; Korb, K.; Boughton, J.; Ting, K-M.
IEEE Transactions on Knowledge and Data Engineering, vol. 19, no. 12, pp. 1652-1665, 2007.
@article{YangWebbCerquideszKorbBoughtonTing07,
title = {To Select or To Weigh: A Comparative Study of Linear Combination Schemes for SuperParent-One-Dependence Estimators},
author = {Y. Yang and G. I. Webb and J. Cerquides and K. Korb and J. Boughton and K-M. Ting},
doi = {10.1109/TKDE.2007.190650},
year = {2007},
date = {2007-01-01},
journal = {IEEE Transactions on Knowledge and Data Engineering},
volume = {19},
number = {12},
pages = {1652-1665},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA},
abstract = {We conduct a large-scale comparative study on linearly combining superparent-one-dependence estimators (SPODEs), a popular family of semi-naive Bayesian classifiers. Altogether 16 model selection and weighing schemes, 58 benchmark data sets, as well as various statistical tests are employed. This paper's main contributions are three-fold. First, it formally presents each scheme's definition, rationale and time complexity; and hence can serve as a comprehensive reference for researchers interested in ensemble learning. Second, it offers bias-variance analysis for each scheme's classification error performance. Third, it identifies effective schemes that meet various needs in practice. This leads to accurate and fast classification algorithms with immediate and significant impact on real-world applications. Another important feature of our study is using a variety of statistical tests to evaluate multiple learning methods across multiple data sets.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We conduct a large-scale comparative study on linearly combining superparent-one-dependence estimators (SPODEs), a popular family of semi-naive Bayesian classifiers. Altogether 16 model selection and weighing schemes, 58 benchmark data sets, as well as various statistical tests are employed. This paper's main contributions are three-fold. First, it formally presents each scheme's definition, rationale and time complexity; and hence can serve as a comprehensive reference for researchers interested in ensemble learning. Second, it offers bias-variance analysis for each scheme's classification error performance. Third, it identifies effective schemes that meet various needs in practice. This leads to accurate and fast classification algorithms with immediate and significant impact on real-world applications. Another important feature of our study is using a variety of statistical tests to evaluate multiple learning methods across multiple data sets.
Yang, Y.; Webb, G. I.; Korb, K.; Ting, K-M.
Machine Learning, vol. 69, no. 1, pp. 35-53, 2007.
@article{YangWebbKorbTing07,
title = {Classifying under Computational Resource Constraints: Anytime Classification Using Probabilistic Estimators},
author = {Y. Yang and G. I. Webb and K. Korb and K-M. Ting},
doi = {10.1007/s10994-007-5020-z},
year = {2007},
date = {2007-01-01},
journal = {Machine Learning},
volume = {69},
number = {1},
pages = {35-53},
publisher = {Springer},
address = {Netherlands},
abstract = {In many online applications of machine learning, the computational resources available for classification will vary from time to time. Most techniques are designed to operate within the constraints of the minimum expected resources and fail to utilize further resources when they are available. We propose a novel anytime classification algorithm, anytime averaged probabilistic estimators (AAPE), which is capable of delivering strong prediction accuracy with little CPU time and utilizing additional CPU time to increase classification accuracy. The idea is to run an ordered sequence of very efficient Bayesian probabilistic estimators (single improvement steps) until classification time runs out. Theoretical studies and empirical validations reveal that by properly identifying, ordering, invoking and ensembling single improvement steps, AAPE is able to accomplish accurate classification whenever it is interrupted. It is also able to output class probability estimates beyond simple 0/1-loss classifications, as well as adeptly handle incremental learning.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In many online applications of machine learning, the computational resources available for classification will vary from time to time. Most techniques are designed to operate within the constraints of the minimum expected resources and fail to utilize further resources when they are available. We propose a novel anytime classification algorithm, anytime averaged probabilistic estimators (AAPE), which is capable of delivering strong prediction accuracy with little CPU time and utilizing additional CPU time to increase classification accuracy. The idea is to run an ordered sequence of very efficient Bayesian probabilistic estimators (single improvement steps) until classification time runs out. Theoretical studies and empirical validations reveal that by properly identifying, ordering, invoking and ensembling single improvement steps, AAPE is able to accomplish accurate classification whenever it is interrupted. It is also able to output class probability estimates beyond simple 0/1-loss classifications, as well as adeptly handle incremental learning.
Zheng, F.; Webb, G. I.
Finding the Right Family: Parent and Child Selection for Averaged One-Dependence Estimators
Lecture Notes in Artificial Intelligence 4710: Proceedings of the 18th European Conference on Machine Learning (ECML'07), pp. 490-501, Springer-Verlag, Warsaw, Poland, 2007.
@inproceedings{ZhengWebb07,
title = {Finding the Right Family: Parent and Child Selection for Averaged One-Dependence Estimators},
author = {F. Zheng and G. I. Webb},
year = {2007},
date = {2007-01-01},
booktitle = {Lecture Notes in Artificial Intelligence 4710: Proceedings of the 18th European Conference on Machine Learning (ECML'07)},
pages = {490-501},
publisher = {Springer-Verlag},
address = {Warsaw, Poland},
abstract = {Averaged One-Dependence Estimators (AODE) classifies by uniformly aggregating all qualified one-dependence estimators (ODEs). Its capacity to significantly improve naive Bayes' accuracy without undue time complexity has attracted substantial interest. Forward Sequential Selection and Backwards Sequential Elimination are effective wrapper techniques to identify and repair harmful interdependencies which have been profitably applied to naive Bayes. However, their straightforward application to AODE has previously proved ineffective. We investigate novel variants of these strategies. Our extensive experiments show that elimination of child attributes from within the constituent ODEs results in a significant improvement in probability estimate and reductions in bias and error relative to unmodified AODE. In contrast, elimination of complete constituent ODEs and the four types of attribute addition are found to be less effective and do not demonstrate any strong advantage over AODE. These surprising results lead to effective techniques for improving AODE's prediction accuracy.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Averaged One-Dependence Estimators (AODE) classifies by uniformly aggregating all qualified one-dependence estimators (ODEs). Its capacity to significantly improve naive Bayes' accuracy without undue time complexity has attracted substantial interest. Forward Sequential Selection and Backwards Sequential Elimination are effective wrapper techniques to identify and repair harmful interdependencies which have been profitably applied to naive Bayes. However, their straightforward application to AODE has previously proved ineffective. We investigate novel variants of these strategies. Our extensive experiments show that elimination of child attributes from within the constituent ODEs results in a significant improvement in probability estimate and reductions in bias and error relative to unmodified AODE. In contrast, elimination of complete constituent ODEs and the four types of attribute addition are found to be less effective and do not demonstrate any strong advantage over AODE. These surprising results lead to effective techniques for improving AODE's prediction accuracy.
Faux, N. G.; Huttley, G. A.; Mahmood, K.; Webb, G. I.; de la Banda, M. Garcia; Whisstock, J. C.
RCPdb: An evolutionary classification and codon usage database for repeat-containing proteins
Genome Research, vol. 17, no. 1, pp. 1118-1127, 2007.
@article{FauxHuttleyMahmoodWebbGarciaWhisstock07,
title = {RCPdb: An evolutionary classification and codon usage database for repeat-containing proteins},
author = {N. G. Faux and G. A. Huttley and K. Mahmood and G. I. Webb and M. Garcia de la Banda and J. C. Whisstock},
doi = {10.1101/gr.6255407},
year = {2007},
date = {2007-01-01},
journal = {Genome Research},
volume = {17},
number = {1},
pages = {1118-1127},
publisher = {Cold Spring Harbor Laboratory Press, ISSN 1088-9051/07},
address = {Woodbury, New York},
abstract = {Over 3% of human proteins contain single amino acid repeats (repeat-containing proteins, RCPs). Many repeats (homopeptides) localize to important proteins involved in transcription, and the expansion of certain repeats, in particular poly-Q and poly-A tracts, can also lead to the development of neurological diseases. Previous studies have suggested that the homopeptide makeup is a result of the presence of G+C-rich tracts in the encoding genes and that expansion occurs via replication slippage. Here, we have performed a large-scale genomic analysis of the variation of the genes encoding RCPs in 13 species and present these data in an online database (http://repeats.med.monash.edu.au/genetic_analysis/). This resource allows rapid comparison and analysis of RCPs, homopeptides, and their underlying genetic tracts across the eukaryotic species considered. We report three major findings. First, there is a bias for a small subset of codons being reiterated within homopeptides, and there is no G+C or A+T bias relative to the organism's transcriptome. Second, single base pair transversions from the homocodon are unusually common and may represent a mechanism of reducing the rate of homopeptide mutations. Third, homopeptides that are conserved across different species lie within regions that are under stronger purifying selection in contrast to nonconserved homopeptides.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Over 3% of human proteins contain single amino acid repeats (repeat-containing proteins, RCPs). Many repeats (homopeptides) localize to important proteins involved in transcription, and the expansion of certain repeats, in particular poly-Q and poly-A tracts, can also lead to the development of neurological diseases. Previous studies have suggested that the homopeptide makeup is a result of the presence of G+C-rich tracts in the encoding genes and that expansion occurs via replication slippage. Here, we have performed a large-scale genomic analysis of the variation of the genes encoding RCPs in 13 species and present these data in an online database (http://repeats.med.monash.edu.au/genetic_analysis/). This resource allows rapid comparison and analysis of RCPs, homopeptides, and their underlying genetic tracts across the eukaryotic species considered. We report three major findings. First, there is a bias for a small subset of codons being reiterated within homopeptides, and there is no G+C or A+T bias relative to the organism's transcriptome. Second, single base pair transversions from the homocodon are unusually common and may represent a mechanism of reducing the rate of homopeptide mutations. Third, homopeptides that are conserved across different species lie within regions that are under stronger purifying selection in contrast to nonconserved homopeptides.
Lu, J.; Yang, Y.; Webb, G. I.
Incremental Discretization for Naive-Bayes Classifier
Li, Xue; Zaiane, Osmar R.; Li, Zhanhuai (Ed.): Lecture Notes in Computer Science 4093: Proceedings of the Second International Conference on Advanced Data Mining and Applications (ADMA 2006), pp. 223-238, Springer, Xian, China, 2006.
@inproceedings{LuYangWebb06,
title = {Incremental Discretization for Naive-Bayes Classifier},
author = {J. Lu and Y. Yang and G. I. Webb},
editor = {Xue Li and Osmar R. Zaiane and Zhanhuai Li},
year = {2006},
date = {2006-01-01},
booktitle = {Lecture Notes in Computer Science 4093: Proceedings of the Second International Conference on Advanced Data Mining and Applications (ADMA 2006)},
pages = {223-238},
publisher = {Springer},
address = {Xian, China},
abstract = {Naive-Bayes classifiers (NB) support incremental learning. However, the lack of effective incremental discretization methods has been hindering NB's incremental learning in face of quantitative data. This problem is further compounded by the fact that quantitative data are everywhere, from temperature readings to share prices. In this paper, we present a novel incremental discretization method for NB, incremental flexible frequency discretization (IFFD). IFFD discretizes values of a quantitative attribute into a sequence of intervals of flexible sizes. It allows online insertion and splitting operation on intervals. Theoretical analysis and experimental test are conducted to compare IFFD with alternative methods. Empirical evidence suggests that IFFD is efficient and effective. NB coupled with IFFD achieves a rapport between high learning efficiency and high classification accuracy in the context of incremental learning.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Naive-Bayes classifiers (NB) support incremental learning. However, the lack of effective incremental discretization methods has been hindering NB's incremental learning in face of quantitative data. This problem is further compounded by the fact that quantitative data are everywhere, from temperature readings to share prices. In this paper, we present a novel incremental discretization method for NB, incremental flexible frequency discretization (IFFD). IFFD discretizes values of a quantitative attribute into a sequence of intervals of flexible sizes. It allows online insertion and splitting operation on intervals. Theoretical analysis and experimental test are conducted to compare IFFD with alternative methods. Empirical evidence suggests that IFFD is efficient and effective. NB coupled with IFFD achieves a rapport between high learning efficiency and high classification accuracy in the context of incremental learning.
Webb, G. I.
Anytime Learning and Classification for Online Applications
Li, Y.; Looi, M.; Zhong, N. (Ed.): Advances in Intelligent IT: Proceedings of the Fourth International Conference on Active Media Technology (AMT'06). [Extended Abstract], pp. 7-12, IOS Press, Brisbane, Australia, 2006.
@inproceedings{Webb06b,
title = {Anytime Learning and Classification for Online Applications},
author = {G. I. Webb},
editor = {Y. Li and M. Looi and N. Zhong},
year = {2006},
date = {2006-01-01},
booktitle = {Advances in Intelligent IT: Proceedings of the Fourth International Conference on Active Media Technology (AMT'06). [Extended Abstract]},
pages = {7-12},
publisher = {IOS Press},
address = {Brisbane, Australia},
abstract = {Many online applications of machine learning require fast classification and hence utilise efficient classifiers such as naive Bayes. However, outside periods of peak computation load, additional computational resources will often be available. Anytime classification can use whatever computational resources may be available at classification time to improve the accuracy of the classifications made.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Many online applications of machine learning require fast classification and hence utilise efficient classifiers such as naive Bayes. However, outside periods of peak computation load, additional computational resources will often be available. Anytime classification can use whatever computational resources may be available at classification time to improve the accuracy of the classifications made.
Webb, G. I.
Ungar, L.; Craven, M.; Gunopulos, D.; Eliassi-Rad, T. (Ed.): Proceedings of the Twelfth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2006), pp. 434-443, The Association for Computing Machinery, Philadelphia, PA, 2006.
@inproceedings{Webb06a,
title = {Discovering Significant Rules},
author = {G. I. Webb},
editor = {L. Ungar and M. Craven and D. Gunopulos and T. Eliassi-Rad},
url = {http://dl.acm.org/authorize?N00546},
year = {2006},
date = {2006-01-01},
booktitle = {Proceedings of the Twelfth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2006)},
pages = {434-443},
publisher = {The Association for Computing Machinery},
address = {Philadelphia, PA},
abstract = {In many applications, association rules will only be interesting if they represent non-trivial correlations between all constituent items. Numerous techniques have been developed that seek to avoid false discoveries. However, while all provide useful solutions to aspects of this problem, none provides a generic solution that is both flexible enough to accommodate varying definitions of true and false discoveries and powerful enough to provide strict control over the risk of false discoveries. This paper presents generic techniques that allow definitions of true and false discoveries to be specified in terms of arbitrary statistical hypothesis tests and which provide strict control over the experimentwise risk of false discoveries.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
In many applications, association rules will only be interesting if they represent non-trivial correlations between all constituent items. Numerous techniques have been developed that seek to avoid false discoveries. However, while all provide useful solutions to aspects of this problem, none provides a generic solution that is both flexible enough to accommodate varying definitions of true and false discoveries and powerful enough to provide strict control over the risk of false discoveries. This paper presents generic techniques that allow definitions of true and false discoveries to be specified in terms of arbitrary statistical hypothesis tests and which provide strict control over the experimentwise risk of false discoveries.
Webb, G. I.; Brain, D.
Generality is Predictive of Prediction Accuracy
LNAI State-of-the-Art Survey series, 'Data Mining: Theory, Methodology, Techniques, and Applications', pp. 1-13, Springer, Berlin/Heidelberg, 2006, (An earlier version of this paper was published in the Proceedings of PKAW 2002, pp 117-130).
@inproceedings{WebbBrain05,
title = {Generality is Predictive of Prediction Accuracy},
author = {G. I. Webb and D. Brain},
year = {2006},
date = {2006-01-01},
booktitle = {LNAI State-of-the-Art Survey series, 'Data Mining: Theory, Methodology, Techniques, and Applications'},
pages = {1-13},
publisher = {Springer},
address = {Berlin/Heidelberg},
abstract = {During knowledge acquisition it frequently occurs that multiple alternative potential rules all appear equally credible. This paper addresses the dearth of formal analysis about how to select between such alternatives. It presents two hypotheses about the expected impact of selecting between classification rules of differing levels of generality in the absence of other evidence about their likely relative performance on unseen data. We argue that the accuracy on unseen data of the more general rule will tend to be closer to that of a default rule for the class than will that of the more specific rule. We also argue that in comparison to the more general rule, the accuracy of the more specific rule on unseen cases will tend to be closer to the accuracy obtained on training data. Experimental evidence is provided in support of these hypotheses. These hypotheses can be useful for selecting between rules in order to achieve specific knowledge acquisition objectives.},
note = {An earlier version of this paper was published in the Proceedings of PKAW 2002, pp 117-130},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
During knowledge acquisition it frequently occurs that multiple alternative potential rules all appear equally credible. This paper addresses the dearth of formal analysis about how to select between such alternatives. It presents two hypotheses about the expected impact of selecting between classification rules of differing levels of generality in the absence of other evidence about their likely relative performance on unseen data. We argue that the accuracy on unseen data of the more general rule will tend to be closer to that of a default rule for the class than will that of the more specific rule. We also argue that in comparison to the more general rule, the accuracy of the more specific rule on unseen cases will tend to be closer to the accuracy obtained on training data. Experimental evidence is provided in support of these hypotheses. These hypotheses can be useful for selecting between rules in order to achieve specific knowledge acquisition objectives.
Yang, Y.; Webb, G. I.; Cerquides, J.; Korb, K.; Boughton, J.; Ting, K-M.
To Select or To Weigh: A Comparative Study of Model Selection and Model Weighing for SPODE Ensembles
Furkranz, J.; Scheffer, T.; Spiliopoulou, M. (Ed.): Lecture Notes in Computer Science 4212: Proceedings of the 17th European Conference on Machine Learning (ECML'06), pp. 533-544, Springer-Verlag, Berlin, Germany, 2006.
@inproceedings{YangWebbCerquideKorbBoughtonTing06,
title = {To Select or To Weigh: A Comparative Study of Model Selection and Model Weighing for SPODE Ensembles},
author = {Y. Yang and G. I. Webb and J. Cerquides and K. Korb and J. Boughton and K-M. Ting},
editor = {J. Furkranz and T. Scheffer and M. Spiliopoulou},
year = {2006},
date = {2006-01-01},
booktitle = {Lecture Notes in Computer Science 4212: Proceedings of the 17th European Conference on Machine Learning (ECML'06)},
pages = {533-544},
publisher = {Springer-Verlag},
address = {Berlin, Germany},
abstract = {An ensemble of Super-Parent-One-Dependence Estimators (SPODEs) offers a powerful yet simple alternative to naive Bayes classifiers, achieving significantly higher classification accuracy at a moderate cost in classification efficiency. Currently there exist two families of methodologies that ensemble candidate SPODEs for classification. One is to select only helpful SPODEs and uniformly average their probability estimates, a type of model selection. Another is to assign a weight to each SPODE and linearly combine their probability estimates, a methodology named model weighing. This paper presents a theoretical and empirical study comparing model selection and model weighing for ensembling SPODEs. The focus is on maximizing the ensemble's classification accuracy while minimizing its computational time. A number of representative selection and weighing schemes are studied, providing a comprehensive research on this topic and identifying effective schemes that provide alternative trades-offs between speed and expected error},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
An ensemble of Super-Parent-One-Dependence Estimators (SPODEs) offers a powerful yet simple alternative to naive Bayes classifiers, achieving significantly higher classification accuracy at a moderate cost in classification efficiency. Currently there exist two families of methodologies that ensemble candidate SPODEs for classification. One is to select only helpful SPODEs and uniformly average their probability estimates, a type of model selection. Another is to assign a weight to each SPODE and linearly combine their probability estimates, a methodology named model weighing. This paper presents a theoretical and empirical study comparing model selection and model weighing for ensembling SPODEs. The focus is on maximizing the ensemble's classification accuracy while minimizing its computational time. A number of representative selection and weighing schemes are studied, providing a comprehensive research on this topic and identifying effective schemes that provide alternative trades-offs between speed and expected error
Yang, Y.; Webb, G. I.
Discretization for Data Mining
Wang, John (Ed.): The Encyclopedia of Data Warehousing and Mining, pp. 392-396, Idea Group Inc., Hershey, PA, 2006.
@incollection{YangWebb05,
title = {Discretization for Data Mining},
author = {Y. Yang and G. I. Webb},
editor = {John Wang},
doi = {10.4018/978-1-59140-557-3.ch075},
year = {2006},
date = {2006-01-01},
booktitle = {The Encyclopedia of Data Warehousing and Mining},
pages = {392-396},
publisher = {Idea Group Inc.},
address = {Hershey, PA},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Zheng, F.; Webb, G. I.
Efficient Lazy Elimination for Averaged One-Dependence Estimators
Cohen, W.; Moore, A. (Ed.): ACM International Conference Proceeding Series, Vol. 148: The Proceedings of the Twenty-third International Conference on Machine Learning (ICML'06), pp. 1113-1120, ACM Press, Pittsburgh, Pennsylvania, 2006.
@inproceedings{ZhengWebb06,
title = {Efficient Lazy Elimination for Averaged One-Dependence Estimators},
author = {F. Zheng and G. I. Webb},
editor = {W. Cohen and A. Moore},
url = {http://dl.acm.org/authorize?N00547},
year = {2006},
date = {2006-01-01},
booktitle = {ACM International Conference Proceeding Series, Vol. 148: The Proceedings of the Twenty-third International Conference on Machine Learning (ICML'06)},
pages = {1113-1120},
publisher = {ACM Press},
address = {Pittsburgh, Pennsylvania},
abstract = {Semi-naive Bayesian classifiers seek to retain the numerous strengths of naive Bayes while reducing error by weakening the attribute independence assumption. Backwards Sequential Elimination (BSE) is a wrapper technique for attribute elimination that has proved effective at this task. We explore a new efficient technique, Lazy Elimination (LE), which eliminates highly related attribute-values at classification time without the computational overheads inherent in wrapper techniques. We analyze the effect of LE and BSE on Averaged One-Dependence Estimators (AODE), a state-of-the-art semi-naive Bayesian algorithm. Our extensive experiments show that LE significantly reduces bias and error without undue additional computation, while BSE significantly reduces bias but not error, with high training time complexity. In the context of AODE, LE has a significant advantage over BSE in both computational efficiency and error.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Semi-naive Bayesian classifiers seek to retain the numerous strengths of naive Bayes while reducing error by weakening the attribute independence assumption. Backwards Sequential Elimination (BSE) is a wrapper technique for attribute elimination that has proved effective at this task. We explore a new efficient technique, Lazy Elimination (LE), which eliminates highly related attribute-values at classification time without the computational overheads inherent in wrapper techniques. We analyze the effect of LE and BSE on Averaged One-Dependence Estimators (AODE), a state-of-the-art semi-naive Bayesian algorithm. Our extensive experiments show that LE significantly reduces bias and error without undue additional computation, while BSE significantly reduces bias but not error, with high training time complexity. In the context of AODE, LE has a significant advantage over BSE in both computational efficiency and error.
Yang, Q.; Webb, G. I. (Ed.)
Lecture Notes in Artificial Intelligence 4099: Proceedings of the 9th Pacific Rim International Conference on Artificial Intelligence (PRICAI 2006)
Springer, Guilin, China, 2006.
@proceedings{YangWebb06,
title = {Lecture Notes in Artificial Intelligence 4099: Proceedings of the 9th Pacific Rim International Conference on Artificial Intelligence (PRICAI 2006)},
editor = {Q. Yang and G. I. Webb},
year = {2006},
date = {2006-01-01},
publisher = {Springer},
address = {Guilin, China},
series = {Lecture Notes in Artificial Intelligence},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Butler, S.; Webb, G. I.
Wang, John (Ed.): The Encyclopedia of Data Warehousing and Mining, pp. 795-799, Idea Group Inc., Hershey, PA, 2006.
@incollection{ButlerWebb05,
title = {Mining Group Differences},
author = {S. Butler and G. I. Webb},
editor = {John Wang},
doi = {10.4018/978-1-60566-010-3.ch199},
year = {2006},
date = {2006-01-01},
booktitle = {The Encyclopedia of Data Warehousing and Mining},
pages = {795-799},
publisher = {Idea Group Inc.},
address = {Hershey, PA},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Huang, S.; Webb, G. I.
Efficiently Identifying Exploratory Rules' Significance
LNAI State-of-the-Art Survey series, 'Data Mining: Theory, Methodology, Techniques, and Applications', pp. 64-77, Springer, Berlin/Heidelberg, 2006, (An earlier version of this paper was published in S.J. Simoff and G.J. Williams (Eds.), Proceedings of the Third Australasian Data Mining Conference (AusDM04) Cairns, Australia. Sydney: University of Technology, pages 169-182.).
@inproceedings{HuangWebb05b,
title = {Efficiently Identifying Exploratory Rules' Significance},
author = {S. Huang and G. I. Webb},
doi = {10.1007/11677437_6},
year = {2006},
date = {2006-01-01},
booktitle = {LNAI State-of-the-Art Survey series, 'Data Mining: Theory, Methodology, Techniques, and Applications'},
pages = {64-77},
publisher = {Springer},
address = {Berlin/Heidelberg},
abstract = {How to efficiently discard potentially uninteresting rules in exploratory rule discovery is one of the important research foci in data mining. Many researchers have presented algorithms to automatically remove potentially uninteresting rules utilizing background knowledge and user-specified constraints. Identifying the significance of exploratory rules using a significance test is desirable for removing rules that may appear interesting by chance, hence providing the users with a more compact set of resulting rules. However, applying statistical tests to identify significant rules requires considerable computation and data access in order to obtain the necessary statistics. The situation gets worse as the size of the database increases. In this paper, we propose two approaches for improving the efficiency of significant exploratory rule discovery. We also evaluate the experimental effect in impact rule discovery which is suitable for discovering exploratory rules in very large, dense databases.},
note = {An earlier version of this paper was published in S.J. Simoff and G.J. Williams (Eds.), Proceedings of the Third Australasian Data Mining Conference (AusDM04) Cairns, Australia. Sydney: University of Technology, pages 169-182.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
How to efficiently discard potentially uninteresting rules in exploratory rule discovery is one of the important research foci in data mining. Many researchers have presented algorithms to automatically remove potentially uninteresting rules utilizing background knowledge and user-specified constraints. Identifying the significance of exploratory rules using a significance test is desirable for removing rules that may appear interesting by chance, hence providing the users with a more compact set of resulting rules. However, applying statistical tests to identify significant rules requires considerable computation and data access in order to obtain the necessary statistics. The situation gets worse as the size of the database increases. In this paper, we propose two approaches for improving the efficiency of significant exploratory rule discovery. We also evaluate the experimental effect in impact rule discovery which is suitable for discovering exploratory rules in very large, dense databases.
Siu, K. K. W.; Butler, S. M.; Beveridge, T.; Gillam, J. E.; Hall, C. J.; Kaye, A. H.; Lewis, R. A.; Mannan, K.; McLoughlin, G.; Pearson, S.; Round, A. R.; E., Schultke; Webb, G. I.; Wilkinson, S. J.
Identifying markers of pathology in SAXS data of malignant tissues of the brain
Nuclear Instruments and Methods in Physics Research A, vol. 548, pp. 140-146, 2005.
@article{SiuEtAl05,
title = {Identifying markers of pathology in SAXS data of malignant tissues of the brain},
author = {K. K. W. Siu and S. M. Butler and T. Beveridge and J. E. Gillam and C. J. Hall and A. H. Kaye and R. A. Lewis and K. Mannan and G. McLoughlin and S. Pearson and A. R. Round and Schultke E. and G. I. Webb and S. J. Wilkinson},
doi = {10.1016/j.nima.2005.03.081},
year = {2005},
date = {2005-01-01},
journal = {Nuclear Instruments and Methods in Physics Research A},
volume = {548},
pages = {140-146},
publisher = {Elsevier},
abstract = {Conventional neuropathological analysis for brain malignancies is heavily reliant on the observation of morphological abnormalities, observed in thin, stained sections of tissue. Small Angle X-ray Scattering (SAXS) data provide an alternative means of distinguishing pathology by examining the ultra-structural (nanometer length scales) characteristics of tissue. To evaluate the diagnostic potential of SAXS for brain tumors, data was collected from normal, malignant and benign tissues of the human brain at station 2.1 of the Daresbury Laboratory Synchrotron Radiation Source and subjected to data mining and multivariate statistical analysis. The results suggest SAXS data may be an effective classi.er of malignancy.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Conventional neuropathological analysis for brain malignancies is heavily reliant on the observation of morphological abnormalities, observed in thin, stained sections of tissue. Small Angle X-ray Scattering (SAXS) data provide an alternative means of distinguishing pathology by examining the ultra-structural (nanometer length scales) characteristics of tissue. To evaluate the diagnostic potential of SAXS for brain tumors, data was collected from normal, malignant and benign tissues of the human brain at station 2.1 of the Daresbury Laboratory Synchrotron Radiation Source and subjected to data mining and multivariate statistical analysis. The results suggest SAXS data may be an effective classi.er of malignancy.
Webb, G. I.
K-Optimal Pattern Discovery: An Efficient and Effective Approach to Exploratory Data Mining
Zhang, S.; Jarvis, R. (Ed.): Lecture Notes in Computer Science 3809: Advances in Artificial Intelligence, Proceedings of the 18th Australian Joint Conference on Artificial Intelligence (AI 2005)[Extended Abstract], pp. 1-2, Springer, Sydney, Australia, 2005.
@inproceedings{Webb05a,
title = {K-Optimal Pattern Discovery: An Efficient and Effective Approach to Exploratory Data Mining},
author = {G. I. Webb},
editor = {S. Zhang and R. Jarvis},
doi = {10.1007/11589990_1},
year = {2005},
date = {2005-01-01},
booktitle = {Lecture Notes in Computer Science 3809: Advances in Artificial Intelligence, Proceedings of the 18th Australian Joint Conference on Artificial Intelligence (AI 2005)[Extended Abstract]},
pages = {1-2},
publisher = {Springer},
address = {Sydney, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Webb, G. I.; Boughton, J.; Wang, Z.
Not So Naive Bayes: Aggregating One-Dependence Estimators
Machine Learning, vol. 58, no. 1, pp. 5-24, 2005.
@article{WebbBoughtonWang05,
title = {Not So Naive Bayes: Aggregating One-Dependence Estimators},
author = {G. I. Webb and J. Boughton and Z. Wang},
doi = {10.1007/s10994-005-4258-6},
year = {2005},
date = {2005-01-01},
journal = {Machine Learning},
volume = {58},
number = {1},
pages = {5-24},
publisher = {Springer},
address = {Netherlands},
abstract = {Of numerous proposals to improve the accuracy of naive Bayes by weakening its attribute independence assumption, both LBR and TAN have demonstrated remarkable error performance. However, both techniques obtain this outcome at a considerable computational cost. We present a new approach to weakening the attribute independence assumption by averaging all of a constrained class of classifiers. In extensive experiments this technique delivers comparable prediction accuracy to LBR and TAN with substantially improved computational efficiency.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Of numerous proposals to improve the accuracy of naive Bayes by weakening its attribute independence assumption, both LBR and TAN have demonstrated remarkable error performance. However, both techniques obtain this outcome at a considerable computational cost. We present a new approach to weakening the attribute independence assumption by averaging all of a constrained class of classifiers. In extensive experiments this technique delivers comparable prediction accuracy to LBR and TAN with substantially improved computational efficiency.
Webb, G. I.; Ting, K. M.
On the Application of ROC Analysis to Predict Classification Performance Under Varying Class Distributions
Machine Learning, vol. 58, no. 1, pp. 25-32, 2005.
@article{WebbTing05,
title = {On the Application of ROC Analysis to Predict Classification Performance Under Varying Class Distributions},
author = {G. I. Webb and K. M. Ting},
year = {2005},
date = {2005-01-01},
journal = {Machine Learning},
volume = {58},
number = {1},
pages = {25-32},
publisher = {Springer},
address = {Netherlands},
abstract = {We counsel caution in the application of ROC analysis for prediction of classifier accuracy under varying class distributions. The heart of our contention is that in real-world applications variations of class distribution are likely to result from forces that affect the distribution of the attribute-values, rather than forces that directly affect the class distribution. In statistical terms, it is usually the class, rather than the attributes, that is the dependent variable. If the class distribution alters as an indirect consequence of changes in the distribution of the attribute values, rather than vice versa, performance estimates derived through ROC analysis may be grossly inaccurate.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We counsel caution in the application of ROC analysis for prediction of classifier accuracy under varying class distributions. The heart of our contention is that in real-world applications variations of class distribution are likely to result from forces that affect the distribution of the attribute-values, rather than forces that directly affect the class distribution. In statistical terms, it is usually the class, rather than the attributes, that is the dependent variable. If the class distribution alters as an indirect consequence of changes in the distribution of the attribute values, rather than vice versa, performance estimates derived through ROC analysis may be grossly inaccurate.
Webb, G. I.; Zhang, S.
Data Mining and Knowledge Discovery, vol. 10, no. 1, pp. 39-79, 2005.
@article{WebbZhang05,
title = {k-Optimal-Rule-Discovery},
author = {G. I. Webb and S. Zhang},
doi = {10.1007/s10618-005-0255-4},
year = {2005},
date = {2005-01-01},
journal = {Data Mining and Knowledge Discovery},
volume = {10},
number = {1},
pages = {39-79},
publisher = {Springer},
address = {Netherlands},
abstract = {K-most-interesting rule discovery finds the k rules that optimize a user-specified measure of interestingness with respect to a set of sample data and user-specified constraints. This approach avoids many limitations of the frequent itemset approach of association rule discovery. This paper presents a scalable algorithm applicable to a wide range of k-most-interesting rule discovery tasks and demonstrates its efficiency.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
K-most-interesting rule discovery finds the k rules that optimize a user-specified measure of interestingness with respect to a set of sample data and user-specified constraints. This approach avoids many limitations of the frequent itemset approach of association rule discovery. This paper presents a scalable algorithm applicable to a wide range of k-most-interesting rule discovery tasks and demonstrates its efficiency.
Yang, Y.; Korb, K.; Ting, K-M.; Webb, G. I.
Ensemble Selection for SuperParent-One-Dependence Estimators
Zhang, S.; Jarvis, R. (Ed.): Lecture Notes in Computer Science 3809: Advances in Artificial Intelligence, Proceedings of the 18th Australian Joint Conference on Artificial Intelligence (AI 2005), pp. 102-111, Springer, Sydney, Australia, 2005.
@inproceedings{YangKorbTingWebb05,
title = {Ensemble Selection for SuperParent-One-Dependence Estimators},
author = {Y. Yang and K. Korb and K-M. Ting and G. I. Webb},
editor = {S. Zhang and R. Jarvis},
doi = {10.1007/11589990_13},
year = {2005},
date = {2005-01-01},
urldate = {2005-01-01},
booktitle = {Lecture Notes in Computer Science 3809: Advances in Artificial Intelligence, Proceedings of the 18th Australian Joint Conference on Artificial Intelligence (AI 2005)},
pages = {102-111},
publisher = {Springer},
address = {Sydney, Australia},
abstract = {SuperParent-One-Dependence Estimators (SPODEs) loosen Naive-Bayes\' attribute independence assumption by allowing each attribute to depend on a common single attribute (superparent) in addition to the class. An ensemble of SPODEs is able to achieve high classification accuracy with modest computational cost. This paper investigates how to select SPODEs for ensembling. Various popular model selection strategies are presented. Their learning efficacy and efficiency are theoretically analyzed and empirically verified. Accordingly, guidelines are investigated for choosing between selection criteria in differing contexts.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
SuperParent-One-Dependence Estimators (SPODEs) loosen Naive-Bayes' attribute independence assumption by allowing each attribute to depend on a common single attribute (superparent) in addition to the class. An ensemble of SPODEs is able to achieve high classification accuracy with modest computational cost. This paper investigates how to select SPODEs for ensembling. Various popular model selection strategies are presented. Their learning efficacy and efficiency are theoretically analyzed and empirically verified. Accordingly, guidelines are investigated for choosing between selection criteria in differing contexts.
Yang, Y.; Webb, G. I.; Wu, X.
Maimon, O.; Rokach, L. (Ed.): The Data Mining and Knowledge Discovery Handbook, pp. 113-130, Springer, Berlin, 2005.
@incollection{YangWebbWu05,
title = {Discretization Methods},
author = {Y. Yang and G. I. Webb and X. Wu},
editor = {O. Maimon and L. Rokach},
doi = {10.1007/978-0-387-09823-4_6},
year = {2005},
date = {2005-01-01},
booktitle = {The Data Mining and Knowledge Discovery Handbook},
pages = {113-130},
publisher = {Springer},
address = {Berlin},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Zheng, F.; I., Webb. G.
A Comparative Study of Semi-naive Bayes Methods in Classification Learning
Simoff, S. J.; Williams, G. J.; Galloway, J.; Kolyshkina, I. (Ed.): Proceedings of the Fourth Australasian Data Mining Conference (AusDM05), pp. 141-156, University of Technology, Sydney, Australia, 2005.
@inproceedings{ZhengWebb05,
title = {A Comparative Study of Semi-naive Bayes Methods in Classification Learning},
author = {F. Zheng and Webb. G. I.},
editor = {S. J. Simoff and G. J. Williams and J. Galloway and I. Kolyshkina},
year = {2005},
date = {2005-01-01},
booktitle = {Proceedings of the Fourth Australasian Data Mining Conference (AusDM05)},
pages = {141-156},
publisher = {University of Technology},
address = {Sydney, Australia},
abstract = {Numerous techniques have sought to improve the accuracy of Naive Bayes (NB) by alleviating the attribute interdependence problem. This paper summarizes these semi-naive Bayesian methods into two groups: those that apply conventional NB with a new attribute set, and those that alter NB by allowing inter-dependencies between attributes. We review eight typical semi-naive Bayesian learning algorithms and perform error analysis using the bias-variance decomposition on thirty-six natural domains from the UCI Machine Learning Repository. In analysing the results of these experiments we provide general recommendations for selection between methods.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Numerous techniques have sought to improve the accuracy of Naive Bayes (NB) by alleviating the attribute interdependence problem. This paper summarizes these semi-naive Bayesian methods into two groups: those that apply conventional NB with a new attribute set, and those that alter NB by allowing inter-dependencies between attributes. We review eight typical semi-naive Bayesian learning algorithms and perform error analysis using the bias-variance decomposition on thirty-six natural domains from the UCI Machine Learning Repository. In analysing the results of these experiments we provide general recommendations for selection between methods.
Huang, S.; Webb, G. I.
Pruning Derivative Partial Rules During Impact Rule Discovery
Ho, T. B.; Cheung, D.; Liu, H. (Ed.): Lecture Notes in Computer Science Vol. 3518: Proceedings of the 9th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining (PAKDD 2005), pp. 71-80, Springer, Hanoi, Vietnam, 2005.
@inproceedings{HuangWebb05a,
title = {Pruning Derivative Partial Rules During Impact Rule Discovery},
author = {S. Huang and G. I. Webb},
editor = {T. B. Ho and D. Cheung and H. Liu},
year = {2005},
date = {2005-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 3518: Proceedings of the 9th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining (PAKDD 2005)},
pages = {71-80},
publisher = {Springer},
address = {Hanoi, Vietnam},
abstract = {Because exploratory rule discovery works with data that is only a sample of the phenomena to be investigated, some resulting rules may appear interesting only by chance. Techniques are developed for automatically discarding statistically insignificant exploratory rules that cannot survive a hypothesis with regard to its ancestors. We call such insignificant rules derivative extended rules. In this paper, we argue that there is another type of derivative exploratory rules, which is derivative with regard to their children. We also argue that considerable amount of such derivative partial rules can not be successfully removed using existing rule pruning techniques. We propose a new technique to address this problem. Experiments are done in impact rule discovery to evaluate the effect of this derivative partial rule filter. Results show that the inherent problem of too many resulting rules in exploratory rule discovery is alleviated.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Because exploratory rule discovery works with data that is only a sample of the phenomena to be investigated, some resulting rules may appear interesting only by chance. Techniques are developed for automatically discarding statistically insignificant exploratory rules that cannot survive a hypothesis with regard to its ancestors. We call such insignificant rules derivative extended rules. In this paper, we argue that there is another type of derivative exploratory rules, which is derivative with regard to their children. We also argue that considerable amount of such derivative partial rules can not be successfully removed using existing rule pruning techniques. We propose a new technique to address this problem. Experiments are done in impact rule discovery to evaluate the effect of this derivative partial rule filter. Results show that the inherent problem of too many resulting rules in exploratory rule discovery is alleviated.
Huang, S.; Webb, G. I.
Discarding Insignificant Rules During Impact Rule Discovery in Large, Dense Databases
Kargupta, H.; Kamath, C.; Srivastava, J.; Goodman, A. (Ed.): Proceedings of the Fifth SIAM International Conference on Data Mining (SDM'05) [short paper], pp. 541-545, Society for Industrial and Applied Mathematics, Newport Beach, CA, 2005.
@inproceedings{HuangWebb05,
title = {Discarding Insignificant Rules During Impact Rule Discovery in Large, Dense Databases},
author = {S. Huang and G. I. Webb},
editor = {H. Kargupta and C. Kamath and J. Srivastava and A. Goodman},
year = {2005},
date = {2005-01-01},
booktitle = {Proceedings of the Fifth SIAM International Conference on Data Mining (SDM'05) [short paper]},
pages = {541-545},
publisher = {Society for Industrial and Applied Mathematics},
address = {Newport Beach, CA},
abstract = {Considerable progress has been made on how to reduce the number of spurious exploratory rules with quantitative attributes. However, little has been done for rules with undiscretized quantitative attributes. It is argued that propositional rules can not effectively describe the interactions between quantitative and qualitative attributes. Aumann and Lindell proposed quantitative association rules to provide a better description of such relationship, together with a rule pruning techniques . Since their technique is based on the frequent itemset framework, it is not suitable for rule discovery in large, dense databases. In this paper, an efficient technique for automatically discarding insignificant rules during rule discovery is proposed, based on the OPUS search algorithm. Experiments demonstrate that the algorithm we propose can efficiently remove potentially uninteresting rules even in very large, dense databases.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Considerable progress has been made on how to reduce the number of spurious exploratory rules with quantitative attributes. However, little has been done for rules with undiscretized quantitative attributes. It is argued that propositional rules can not effectively describe the interactions between quantitative and qualitative attributes. Aumann and Lindell proposed quantitative association rules to provide a better description of such relationship, together with a rule pruning techniques . Since their technique is based on the frequent itemset framework, it is not suitable for rule discovery in large, dense databases. In this paper, an efficient technique for automatically discarding insignificant rules during rule discovery is proposed, based on the OPUS search algorithm. Experiments demonstrate that the algorithm we propose can efficiently remove potentially uninteresting rules even in very large, dense databases.
Newlands, D. A.; Webb, G. I.
Convex Hulls as an Hypothesis Language Bias
Ebecken, N. F. F. E.; Brebbia, C. A.; Zanasi, A. (Ed.): Proceedings of the Fourth International Conference on Data Mining (DATA MINING IV), pp. 285-294, WIT Press, Rio de Janeiro, Brazil, 2004.
@inproceedings{NewlandsWebb04,
title = {Convex Hulls as an Hypothesis Language Bias},
author = {D. A. Newlands and G. I. Webb},
editor = {N. F. F. E. Ebecken and C. A. Brebbia and A. Zanasi},
year = {2004},
date = {2004-01-01},
booktitle = {Proceedings of the Fourth International Conference on Data Mining (DATA MINING IV)},
pages = {285-294},
publisher = {WIT Press},
address = {Rio de Janeiro, Brazil},
abstract = {Classification learning is dominated by systems which induce large numbers of small axis-orthogonal decision surfaces which biases such systems towards particular hypothesis types. However, there is reason to believe that many domains have underlying concepts which do not involve axis orthogonal surfaces. Further, the multiplicity of small decision regions mitigates against any holistic appreciation of the theories produced by these systems, notwithstanding the fact that many of the small regions are individually comprehensible. We propose the use of less strongly biased hypothesis languages which might be expected to model concepts using a number of structures close to the number of actual structures in the domain. An instantiation of such a language, a convex hull based classifier, CH1, has been implemented to investigate modeling concepts as a small number of large geometric structures in n-dimensional space. A comparison of the number of regions induced is made against other well-known systems on a representative selection of largely or wholly continuous valued machine learning tasks. The convex hull system is shown to produce a number of induced regions about an order of magnitude less than well-known systems and very close to the number of actual concepts. This representation, as convex hulls, allows the possibility of extraction of higher level mathematical descriptions of the induced concepts, using the techniques of computational geometry.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Classification learning is dominated by systems which induce large numbers of small axis-orthogonal decision surfaces which biases such systems towards particular hypothesis types. However, there is reason to believe that many domains have underlying concepts which do not involve axis orthogonal surfaces. Further, the multiplicity of small decision regions mitigates against any holistic appreciation of the theories produced by these systems, notwithstanding the fact that many of the small regions are individually comprehensible. We propose the use of less strongly biased hypothesis languages which might be expected to model concepts using a number of structures close to the number of actual structures in the domain. An instantiation of such a language, a convex hull based classifier, CH1, has been implemented to investigate modeling concepts as a small number of large geometric structures in n-dimensional space. A comparison of the number of regions induced is made against other well-known systems on a representative selection of largely or wholly continuous valued machine learning tasks. The convex hull system is shown to produce a number of induced regions about an order of magnitude less than well-known systems and very close to the number of actual concepts. This representation, as convex hulls, allows the possibility of extraction of higher level mathematical descriptions of the induced concepts, using the techniques of computational geometry.
Newlands, D. A.; Webb, G. I.
Alternative Strategies for Decision List Construction
Ebecken, N. F. F. E.; Brebbia, C. A.; Zanasi, A. (Ed.): Proceedings of the Fourth International Conference on Data Mining (DATA MINING IV), pp. 265-273, WIT Press, Rio de Janeiro, Brazil, 2004.
@inproceedings{NewlandsWebb04a,
title = {Alternative Strategies for Decision List Construction},
author = {D. A. Newlands and G. I. Webb},
editor = {N. F. F. E. Ebecken and C. A. Brebbia and A. Zanasi},
year = {2004},
date = {2004-01-01},
booktitle = {Proceedings of the Fourth International Conference on Data Mining (DATA MINING IV)},
pages = {265-273},
publisher = {WIT Press},
address = {Rio de Janeiro, Brazil},
abstract = {This work surveys well-known approaches to building decision lists. Some novel variations to strategies based on default rules for the most common class and insertion of new rules before the default rule are presented. These are expected to offer speed up in the construction of the decision list as well as compression of the length of the list. These strategies and a testing regime have been implemented and some empirical studies done to compare the strategies. Experimental results are presented and interpreted. We show that all strategies deliver decision lists of comparable accuracy. However, two techniques are shown to deliver this accuracy with lists composed of significantly fewer rules than alternative strategies. Of these, one also demonstrates significant computational advantages. The prepending strategy is also demonstrated to produce decision lists which are as much as an order of magnitude shorter than those produced by CN2.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This work surveys well-known approaches to building decision lists. Some novel variations to strategies based on default rules for the most common class and insertion of new rules before the default rule are presented. These are expected to offer speed up in the construction of the decision list as well as compression of the length of the list. These strategies and a testing regime have been implemented and some empirical studies done to compare the strategies. Experimental results are presented and interpreted. We show that all strategies deliver decision lists of comparable accuracy. However, two techniques are shown to deliver this accuracy with lists composed of significantly fewer rules than alternative strategies. Of these, one also demonstrates significant computational advantages. The prepending strategy is also demonstrated to produce decision lists which are as much as an order of magnitude shorter than those produced by CN2.
Thiruvady, D. R.; Webb, G. I.
Mining Negative Rules using GRD
Dai, H.; Srikant, R.; Zhang, C. (Ed.): Lecture Notes in Computer Science Vol. 3056: Proceedings of the Eighth Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 04) [Short Paper], pp. 161-165, Springer, Sydney, Australia, 2004.
@inproceedings{ThiruvadyWebb04,
title = {Mining Negative Rules using GRD},
author = {D. R. Thiruvady and G. I. Webb},
editor = {H. Dai and R. Srikant and C. Zhang},
year = {2004},
date = {2004-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 3056: Proceedings of the Eighth Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 04) [Short Paper]},
pages = {161-165},
publisher = {Springer},
address = {Sydney, Australia},
abstract = {GRD is an algorithm for k-most interesting rule discovery. In contrast to association rule discovery, GRD does not require the use of a minimum support constraint. Rather, the user must specify a measure of interestingness and the number of rules sought (k). This paper reports efficient techniques to extend GRD to support mining of negative rules. We demonstrate that the new approach provides tractable discovery of both negative and positive rules.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
GRD is an algorithm for k-most interesting rule discovery. In contrast to association rule discovery, GRD does not require the use of a minimum support constraint. Rather, the user must specify a measure of interestingness and the number of rules sought (k). This paper reports efficient techniques to extend GRD to support mining of negative rules. We demonstrate that the new approach provides tractable discovery of both negative and positive rules.
Wang, Z.; Webb, G. I.; Zheng, F.
Selective Augmented Bayesian Network Classifiers Based on Rough Set Theory
Dai, H.; Srikant, R.; Zhang, C. (Ed.): Lecture Notes in Computer Science Vol. 3056: Proceedings of the Eighth Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 04), pp. 319-328, Springer, Sydney, Australia, 2004.
@inproceedings{WangWebbZheng04,
title = {Selective Augmented Bayesian Network Classifiers Based on Rough Set Theory},
author = {Z. Wang and G. I. Webb and F. Zheng},
editor = {H. Dai and R. Srikant and C. Zhang},
year = {2004},
date = {2004-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 3056: Proceedings of the Eighth Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 04)},
pages = {319-328},
publisher = {Springer},
address = {Sydney, Australia},
abstract = {The naive Bayes classifier is widely used in interactive applications due to its computational efficiency, direct theoretical base, and competitive accuracy. However, its attribute independence assumption can result in sub-optimal accuracy. A number of techniques have explored simple relaxations of the attribute independence assumption in order to increase accuracy. TAN, Tree-Augmented Naive Bayes, is a state-of-the-art extension of naive Bayes, that can express limited forms of inter-dependence among attributes. Rough sets theory provides tools for expressing inexact or partial dependencies within dataset. In this paper, we present a variant of TAN and compare their tree classifier structures, which can be thought of as a selective restricted trees Bayesian classifier. It delivers lower error than both pre-existing state-of-the-art TAN-based classifiers, with substantially less computation than is required by the SuperParent approach.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The naive Bayes classifier is widely used in interactive applications due to its computational efficiency, direct theoretical base, and competitive accuracy. However, its attribute independence assumption can result in sub-optimal accuracy. A number of techniques have explored simple relaxations of the attribute independence assumption in order to increase accuracy. TAN, Tree-Augmented Naive Bayes, is a state-of-the-art extension of naive Bayes, that can express limited forms of inter-dependence among attributes. Rough sets theory provides tools for expressing inexact or partial dependencies within dataset. In this paper, we present a variant of TAN and compare their tree classifier structures, which can be thought of as a selective restricted trees Bayesian classifier. It delivers lower error than both pre-existing state-of-the-art TAN-based classifiers, with substantially less computation than is required by the SuperParent approach.
Webb, G. I.; Zheng, Z.
Multistrategy Ensemble Learning: Reducing Error by Combining Ensemble Learning Techniques
IEEE Transactions on Knowledge and Data Engineering, vol. 16, no. 8, pp. 980-991, 2004.
@article{WebbZheng04,
title = {Multistrategy Ensemble Learning: Reducing Error by Combining Ensemble Learning Techniques},
author = {G. I. Webb and Z. Zheng},
doi = {10.1109/TKDE.2004.29},
year = {2004},
date = {2004-01-01},
journal = {IEEE Transactions on Knowledge and Data Engineering},
volume = {16},
number = {8},
pages = {980-991},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA},
abstract = {Ensemble learning strategies, especially Boosting and Bagging decision trees, have demonstrated impressive capacities to improve the prediction accuracy of base learning algorithms. Further gains have been demonstrated by strategies that combine simple ensemble formation approaches. In this paper, we investigate the hypothesis that the improvement inaccuracy of multi-strategy approaches to ensemble learning is due to an increase in the diversity of ensemble members that are formed. In addition, guided by this hypothesis, we develop three new multi-strategy ensemble-learning techniques. Experimental results in a wide variety of natural domains suggest that these multi-strategy ensemble-learning techniques are, on average, more accurate than their component ensemble learning techniques},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Ensemble learning strategies, especially Boosting and Bagging decision trees, have demonstrated impressive capacities to improve the prediction accuracy of base learning algorithms. Further gains have been demonstrated by strategies that combine simple ensemble formation approaches. In this paper, we investigate the hypothesis that the improvement inaccuracy of multi-strategy approaches to ensemble learning is due to an increase in the diversity of ensemble members that are formed. In addition, guided by this hypothesis, we develop three new multi-strategy ensemble-learning techniques. Experimental results in a wide variety of natural domains suggest that these multi-strategy ensemble-learning techniques are, on average, more accurate than their component ensemble learning techniques
Webb, Geoffrey I; Conilione, Paul
Estimating bias and variance from data
2004, (Unpublished manuscript).
@unpublished{WebbConilione04,
title = {Estimating bias and variance from data},
author = {Geoffrey I Webb and Paul Conilione},
year = {2004},
date = {2004-01-01},
abstract = {The bias-variance decomposition of error provides useful insights into the error performance of a classifier as it is applied to di#erent types of learning task. Most notably, it has been used to explain the extraordinary e#ectiveness of ensemble learning techniques. It is important that the research community have e#ective tools for assessing such explanations. To this end, techniques have been developed for estimating bias and variance from data. The most widely deployed of these uses repeated sub-sampling with a holdout set. We argue, with empirical support, that this approach has serious limitations. First, it provides very little flexibility in the types of distributions of training sets that may be studied. It requires that the training sets be relatively small and that the degree of variation between training sets be very circumscribed. Second, the approach leads to bias and variance estimates that have high statistical variance and hence low reliability. We develop an alternative method that is based on cross-validation. We show that this method allows far greater flexibility in the types of distribution that are examined and that the estimates derived are much more stable. Finally, we show that changing the distributions of training sets from which bias and variance estimates are drawn can alter substantially the bias and variance estimates that are derived.},
note = {Unpublished manuscript},
keywords = {},
pubstate = {published},
tppubtype = {unpublished}
}
The bias-variance decomposition of error provides useful insights into the error performance of a classifier as it is applied to di#erent types of learning task. Most notably, it has been used to explain the extraordinary e#ectiveness of ensemble learning techniques. It is important that the research community have e#ective tools for assessing such explanations. To this end, techniques have been developed for estimating bias and variance from data. The most widely deployed of these uses repeated sub-sampling with a holdout set. We argue, with empirical support, that this approach has serious limitations. First, it provides very little flexibility in the types of distributions of training sets that may be studied. It requires that the training sets be relatively small and that the degree of variation between training sets be very circumscribed. Second, the approach leads to bias and variance estimates that have high statistical variance and hence low reliability. We develop an alternative method that is based on cross-validation. We show that this method allows far greater flexibility in the types of distribution that are examined and that the estimates derived are much more stable. Finally, we show that changing the distributions of training sets from which bias and variance estimates are drawn can alter substantially the bias and variance estimates that are derived.
Webb, G. I.; Yu, X. (Ed.)
Lecture Notes in Computer Science 3339: Proceedings of the 17th Australian Joint Conference on Artificial Intelligence (AI 2004)
Springer, Cairns, Australia, 2004.
@proceedings{WebbYu04,
title = {Lecture Notes in Computer Science 3339: Proceedings of the 17th Australian Joint Conference on Artificial Intelligence (AI 2004)},
editor = {G. I. Webb and X. Yu},
year = {2004},
date = {2004-01-01},
publisher = {Springer},
address = {Cairns, Australia},
series = {Lecture Notes in Computer Science},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
Rolfe, B.; Frayman, Y.; Webb, G. I.; Hodgson, P.
Analysis of Stamping Production Data with View Towards Quality Management
Proceedings of the 9th International Conference on Manufacturing Excellence (ICME 03), Melbourne, Australia, 2003.
@inproceedings{RolfeFraymanWebbHodgson03,
title = {Analysis of Stamping Production Data with View Towards Quality Management},
author = {B. Rolfe and Y. Frayman and G. I. Webb and P. Hodgson},
year = {2003},
date = {2003-01-01},
booktitle = {Proceedings of the 9th International Conference on Manufacturing Excellence (ICME 03)},
address = {Melbourne, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Rolfe, B.; Hodgson, P.; Webb, G. I.
Improving the Prediction of the Roll Separating Force in a Hot Steel Finishing Mill
Meech, J. A. (Ed.): Intelligence in a Small World - Nanomaterials for the 21st Century. Selected Papers from IPMM-2003, CRC-Press, Sendai, Japan, 2003.
@inproceedings{RolfeHodgsonWebb03,
title = {Improving the Prediction of the Roll Separating Force in a Hot Steel Finishing Mill},
author = {B. Rolfe and P. Hodgson and G. I. Webb},
editor = {J. A. Meech},
year = {2003},
date = {2003-01-01},
booktitle = {Intelligence in a Small World - Nanomaterials for the 21st Century. Selected Papers from IPMM-2003},
publisher = {CRC-Press},
address = {Sendai, Japan},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Shi, H.; Wang, Z.; Webb, G. I.; Huang, H.
A New Restricted Bayesian Network Classifier
Whang, K-Y.; Jeon, J.; Shim, K.; Srivastava, J. (Ed.): Lecture Notes in Artificial Intelligence Vol. 2637: Proceedings of the Seventh Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD'03), pp. 265-270, Springer-Verlag, Seoul, Korea, 2003.
@inproceedings{ShiWangWebbHuang03,
title = {A New Restricted Bayesian Network Classifier},
author = {H. Shi and Z. Wang and G. I. Webb and H. Huang},
editor = {K-Y. Whang and J. Jeon and K. Shim and J. Srivastava},
year = {2003},
date = {2003-01-01},
booktitle = {Lecture Notes in Artificial Intelligence Vol. 2637: Proceedings of the Seventh Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD'03)},
pages = {265-270},
publisher = {Springer-Verlag},
address = {Seoul, Korea},
abstract = {On the basis of examining the existing restricted Bayesian network classifiers, a new Bayes-theorem-based and more strictly restricted Bayesian-network-based classification model DLBAN is proposed, which can be viewed as a double-level Bayesian network augmented naive Bayes classification. The experimental results show that the DLBAN classifier is better than the TAN classifier in the most cases.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
On the basis of examining the existing restricted Bayesian network classifiers, a new Bayes-theorem-based and more strictly restricted Bayesian-network-based classification model DLBAN is proposed, which can be viewed as a double-level Bayesian network augmented naive Bayes classification. The experimental results show that the DLBAN classifier is better than the TAN classifier in the most cases.
Wang, Z.; Webb, G. I.; Zheng, F.
Adjusting Dependence Relations for Semi-Lazy TAN Classifiers
Gedeon, T. D.; Fung, L. C. C. (Ed.): Lecture Notes in Artificial Intelligence Vol. 2903: Proceedings of the 16th Australian Conference on Artificial Intelligence (AI 03), pp. 453-465, Springer, Perth, Australia, 2003.
@inproceedings{WangWebbZheng03,
title = {Adjusting Dependence Relations for Semi-Lazy TAN Classifiers},
author = {Z. Wang and G. I. Webb and F. Zheng},
editor = {T. D. Gedeon and L. C. C. Fung},
year = {2003},
date = {2003-01-01},
booktitle = {Lecture Notes in Artificial Intelligence Vol. 2903: Proceedings of the 16th Australian Conference on Artificial Intelligence (AI 03)},
pages = {453-465},
publisher = {Springer},
address = {Perth, Australia},
abstract = {The naive Bayesian classifier is a simple and effective classification method, which assumes a Bayesian network in which each attribute has the class label as its only one parent. But this assumption is not obviously hold in many real world domains. Tree-Augmented Na?ve Bayes (TAN) is a state-of-the-art extension of the naive Bayes, which can express partial dependence relations among attributes. In this paper, we analyze the implementations of two different TAN classifiers and their tree structures. Experiments show how different dependence relations impact on accuracy of TAN classifiers. We present a kind of semi-lazy TAN classifier, which builds a TAN identical to the original TAN at training time, but adjusts the dependence relations for a new test instance at classification time. Our extensive experimental results show that this kind of semi-lazy classifier delivers lower error than the original TAN and is more efficient than SuperParent TAN.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The naive Bayesian classifier is a simple and effective classification method, which assumes a Bayesian network in which each attribute has the class label as its only one parent. But this assumption is not obviously hold in many real world domains. Tree-Augmented Na?ve Bayes (TAN) is a state-of-the-art extension of the naive Bayes, which can express partial dependence relations among attributes. In this paper, we analyze the implementations of two different TAN classifiers and their tree structures. Experiments show how different dependence relations impact on accuracy of TAN classifiers. We present a kind of semi-lazy TAN classifier, which builds a TAN identical to the original TAN at training time, but adjusts the dependence relations for a new test instance at classification time. Our extensive experimental results show that this kind of semi-lazy classifier delivers lower error than the original TAN and is more efficient than SuperParent TAN.
Webb, G. I.
Preliminary Investigations into Statistically Valid Exploratory Rule Discovery
Simoff, S. J.; Williams, G. J.; Hegland, M. (Ed.): Proceedings of the Second Australasian Data Mining Conference (AusDM03), pp. 1-9, University of Technology, Canberra, Australia, 2003.
@inproceedings{Webb03a,
title = {Preliminary Investigations into Statistically Valid Exploratory Rule Discovery},
author = {G. I. Webb},
editor = {S. J. Simoff and G. J. Williams and M. Hegland},
year = {2003},
date = {2003-01-01},
booktitle = {Proceedings of the Second Australasian Data Mining Conference (AusDM03)},
pages = {1-9},
publisher = {University of Technology},
address = {Canberra, Australia},
abstract = {Exploratory rule discovery, as exemplified by association rule discovery, has proven very popular. In this paper I investigate issues surrounding the statistical validity of rules found using this approach and methods that might be employed to deliver statistically sound exploratory rule discovery.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Exploratory rule discovery, as exemplified by association rule discovery, has proven very popular. In this paper I investigate issues surrounding the statistical validity of rules found using this approach and methods that might be employed to deliver statistically sound exploratory rule discovery.
Webb, G. I.
Association Rules
Ye, Nong (Ed.): The Handbook of Data Mining, Chapter 2, pp. 25-39, Lawrence Erlbaum Associates, 2003.
@incollection{Webb03,
title = {Association Rules},
author = {G. I. Webb},
editor = {Nong Ye},
year = {2003},
date = {2003-01-01},
booktitle = {The Handbook of Data Mining, Chapter 2},
pages = {25-39},
publisher = {Lawrence Erlbaum Associates},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Webb, G. I.; Butler, S.; Newlands, D.
On Detecting Differences Between Groups
Domingos, P.; Faloutsos, C.; Senator, T.; Kargupta, H.; Getoor, L. (Ed.): Proceedings of The Ninth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2003), pp. 256-265, The Association for Computing Machinery, Washington, DC, 2003.
@inproceedings{WebbButlerNewlands03,
title = {On Detecting Differences Between Groups},
author = {G. I. Webb and S. Butler and D. Newlands},
editor = {P. Domingos and C. Faloutsos and T. Senator and H. Kargupta and L. Getoor},
year = {2003},
date = {2003-01-01},
booktitle = {Proceedings of The Ninth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2003)},
pages = {256-265},
publisher = {The Association for Computing Machinery},
address = {Washington, DC},
abstract = {Understanding the differences between contrasting groups is a fundamental task in data analysis. This realization has led to the development of a new special purpose data mining technique, em contrast-set mining. We undertook a study with a retail collaborator to compare contrast-set mining with existing rule-discovery techniques. To our surprise we observed that straightforward application of an existing commercial rule-discovery system, Magnum Opus, could successfully perform the contrast-set-mining task. This led to the realization that contrast-set mining is a special case of the more general rule-discovery task. We present the results of our study together with a proof of this conclusion},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Understanding the differences between contrasting groups is a fundamental task in data analysis. This realization has led to the development of a new special purpose data mining technique, em contrast-set mining. We undertook a study with a retail collaborator to compare contrast-set mining with existing rule-discovery techniques. To our surprise we observed that straightforward application of an existing commercial rule-discovery system, Magnum Opus, could successfully perform the contrast-set-mining task. This led to the realization that contrast-set mining is a special case of the more general rule-discovery task. We present the results of our study together with a proof of this conclusion
Yang, Y.; Webb, G. I.
Weighted Proportional k-Interval Discretization for Naive-Bayes Classifiers
Whang, K-Y.; Jeon, J.; Shim, K.; Srivastava, J. (Ed.): Lecture Notes in Artificial Intelligence Vol. 2637: Proceedings of the Seventh Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD'03), pp. 501-512, Springer-Verlag, Seoul, Korea, 2003.
@inproceedings{YangWebb03,
title = {Weighted Proportional k-Interval Discretization for Naive-Bayes Classifiers},
author = {Y. Yang and G. I. Webb},
editor = {K-Y. Whang and J. Jeon and K. Shim and J. Srivastava},
year = {2003},
date = {2003-01-01},
booktitle = {Lecture Notes in Artificial Intelligence Vol. 2637: Proceedings of the Seventh Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD'03)},
pages = {501-512},
publisher = {Springer-Verlag},
address = {Seoul, Korea},
abstract = {The use of different discretization techniques can be expected to affect the bias and variance of a learning algorithm. We call such an effect discretization bias and variance. Proportional k-interval discretization (PKID) tunes discretization bias and variance by adjusting discretized interval size and number proportional to the number of training instances. Theoretical analysis suggests that this is desirable for naive-Bayes classifiers. However PKID has sub-optimal performance when learning from small training data. We argue that this is because PKID equally weighs bias reduction and variance reduction. But for small data, variance reduction can contribute more to lower learning error and thus should be given greater weight than bias reduction. Accordingly we propose weighted proportional k-interval discretization (WPKID), which establishes a more suitable bias and variance trade-off for small data while allowing additional training data to be used to reduce both bias and variance. Our experiments demonstrate that for naive-Bayes classifiers, WPKID improves upon PKID for smaller datasets with significant frequency; and WPKID delivers lower classification error significantly more often than not in comparison to the other three leading alternative discretization techniques studied.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The use of different discretization techniques can be expected to affect the bias and variance of a learning algorithm. We call such an effect discretization bias and variance. Proportional k-interval discretization (PKID) tunes discretization bias and variance by adjusting discretized interval size and number proportional to the number of training instances. Theoretical analysis suggests that this is desirable for naive-Bayes classifiers. However PKID has sub-optimal performance when learning from small training data. We argue that this is because PKID equally weighs bias reduction and variance reduction. But for small data, variance reduction can contribute more to lower learning error and thus should be given greater weight than bias reduction. Accordingly we propose weighted proportional k-interval discretization (WPKID), which establishes a more suitable bias and variance trade-off for small data while allowing additional training data to be used to reduce both bias and variance. Our experiments demonstrate that for naive-Bayes classifiers, WPKID improves upon PKID for smaller datasets with significant frequency; and WPKID delivers lower classification error significantly more often than not in comparison to the other three leading alternative discretization techniques studied.
Yang, Y.; Webb, G. I.
On Why Discretization Works for Naive-Bayes Classifiers
Gedeon, T. D.; Fung, L. C. C. (Ed.): Lecture Notes in Artificial Intelligence Vol. 2903: Proceedings of the 16th Australian Conference on Artificial Intelligence (AI 03), pp. 440-452, Springer, Perth, Australia, 2003.
@inproceedings{YangWebb03c,
title = {On Why Discretization Works for Naive-Bayes Classifiers},
author = {Y. Yang and G. I. Webb},
editor = {T. D. Gedeon and L. C. C. Fung},
year = {2003},
date = {2003-01-01},
booktitle = {Lecture Notes in Artificial Intelligence Vol. 2903: Proceedings of the 16th Australian Conference on Artificial Intelligence (AI 03)},
pages = {440-452},
publisher = {Springer},
address = {Perth, Australia},
abstract = {We investigate why discretization is effective in naive-Bayes learning. We prove a theorem that identifies particular conditions under which discretization will result in naive Bayes classifiers delivering the same probability estimates as would be obtained if the correct probability density functions were employed. We discuss the factors that might affect naive-Bayes classification error under discretization. We suggest that the use of different discretization techniques can affect the classification bias and variance of the generated classifiers, an effect named discretization bias and variance. We argue that by properly managing discretization bias and variance, we can effectively reduce naive-Bayes classification error},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We investigate why discretization is effective in naive-Bayes learning. We prove a theorem that identifies particular conditions under which discretization will result in naive Bayes classifiers delivering the same probability estimates as would be obtained if the correct probability density functions were employed. We discuss the factors that might affect naive-Bayes classification error under discretization. We suggest that the use of different discretization techniques can affect the classification bias and variance of the generated classifiers, an effect named discretization bias and variance. We argue that by properly managing discretization bias and variance, we can effectively reduce naive-Bayes classification error
Zhang, C.; Zhang, S.; Webb, G. I.
Identifying Approximate Itemsets of Interest In Large Databases
Applied Intelligence, vol. 18, pp. 91-104, 2003.
@article{ZhangZhangWebb03,
title = {Identifying Approximate Itemsets of Interest In Large Databases},
author = {C. Zhang and S. Zhang and G. I. Webb},
url = {http://link.springer.com/content/pdf/10.1023%2FA%3A1020995206763.pdf},
year = {2003},
date = {2003-01-01},
journal = {Applied Intelligence},
volume = {18},
pages = {91-104},
publisher = {Springer},
address = {Netherlands},
abstract = {This paper presents a method for discovering approximate frequent itemsets of interest in large scale databases. This method uses the central limit theorem to increase efficiency, enabling us to reduce the sample size by about half compared to previous approximations. Further efficiency is gained by pruning from the search space uninteresting frequent itemsets. In addition to improving efficiency, this measure also reduces the number of itemsets that the user need consider. The model and algorithm have been implemented and evaluated using both synthetic and real-world databases. Our experimental results demonstrate the efficiency of the approach},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper presents a method for discovering approximate frequent itemsets of interest in large scale databases. This method uses the central limit theorem to increase efficiency, enabling us to reduce the sample size by about half compared to previous approximations. Further efficiency is gained by pruning from the search space uninteresting frequent itemsets. In addition to improving efficiency, this measure also reduces the number of itemsets that the user need consider. The model and algorithm have been implemented and evaluated using both synthetic and real-world databases. Our experimental results demonstrate the efficiency of the approach
Butler, S. M.; Webb, G. I.; Lewis, R. A.
A Case Study in Feature Invention for Breast Cancer Diagnosis Using X-Ray Scatter Images
Gedeon, T. D.; Fung, L. C. C. (Ed.): Lecture Notes in Artificial Intelligence Vol. 2903: Proceedings of the 16th Australian Conference on Artificial Intelligence (AI 03), pp. 677-685, Springer, Berlin/Heidelberg, 2003.
@inproceedings{ButlerWebbLewis03,
title = {A Case Study in Feature Invention for Breast Cancer Diagnosis Using X-Ray Scatter Images},
author = {S. M. Butler and G. I. Webb and R. A. Lewis},
editor = {T. D. Gedeon and L. C. C. Fung},
doi = {10.1007/978-3-540-24581-0_58},
year = {2003},
date = {2003-01-01},
booktitle = {Lecture Notes in Artificial Intelligence Vol. 2903: Proceedings of the 16th Australian Conference on Artificial Intelligence (AI 03)},
pages = {677-685},
publisher = {Springer},
address = {Berlin/Heidelberg},
abstract = {X-ray mammography is the current method for screening for breast cancer, and like any technique, has its limitations. Several groups have reported differences in the X-ray scattering patterns of normal and tumour tissue from the breast. This gives rise to the hope that X-ray scatter analysis techniques may lead to a more accurate and cost effective method of diagnosing beast cancer which lends itself to automation. This is a particularly challenging exercise due to the inherent complexity of the information content in X-ray scatter patterns from complex heterogenous tissue samples. We use a simple naive Bayes classier, coupled with Equal Frequency Discretization (EFD) as our classification system. High-level features are extracted from the low-level pixel data. This paper reports some preliminary results in the ongoing development of this classification method that can distinguish between the diffraction patterns of normal and cancerous tissue, with particular emphasis on the invention of features for classification.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
X-ray mammography is the current method for screening for breast cancer, and like any technique, has its limitations. Several groups have reported differences in the X-ray scattering patterns of normal and tumour tissue from the breast. This gives rise to the hope that X-ray scatter analysis techniques may lead to a more accurate and cost effective method of diagnosing beast cancer which lends itself to automation. This is a particularly challenging exercise due to the inherent complexity of the information content in X-ray scatter patterns from complex heterogenous tissue samples. We use a simple naive Bayes classier, coupled with Equal Frequency Discretization (EFD) as our classification system. High-level features are extracted from the low-level pixel data. This paper reports some preliminary results in the ongoing development of this classification method that can distinguish between the diffraction patterns of normal and cancerous tissue, with particular emphasis on the invention of features for classification.
Pearce, J.; Webb, G. I.; Shaw, R.; Garner, B.
A Systemic Approach to the Database Marketing Process
Proceedings of the Australian and New Zealand Marketing Academy Conference (ANZMAC 02), pp. 2941-2948, Deakin University (CD Rom), Geelong, Australia, 2002.
@inproceedings{PearceWebbShawGarner02,
title = {A Systemic Approach to the Database Marketing Process},
author = {J. Pearce and G. I. Webb and R. Shaw and B. Garner},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the Australian and New Zealand Marketing Academy Conference (ANZMAC 02)},
pages = {2941-2948},
publisher = {Deakin University (CD Rom)},
address = {Geelong, Australia},
abstract = {The role of database marketing (DBM) has become increasingly important for organisations that have large databases of information on customers with whom they deal directly. At the same time, DBM models used in practice have increased in sophistication. This paper examines a systemic view of DBM and the role of analytical techniques within DBM. It extends existing process models to develop a systemic model that encompasses the increased complexity of DBM in practice. The systemic model provides a framework to integrate data mining, experimental design and prioritisation decisions. This paper goes on to identify opportunities for research in DBM, including DBM process models used in practice, the use of evolutionary operations techniques in DBM, prioritisation decisions, and the factors that surround the uptake of DBM.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The role of database marketing (DBM) has become increasingly important for organisations that have large databases of information on customers with whom they deal directly. At the same time, DBM models used in practice have increased in sophistication. This paper examines a systemic view of DBM and the role of analytical techniques within DBM. It extends existing process models to develop a systemic model that encompasses the increased complexity of DBM in practice. The systemic model provides a framework to integrate data mining, experimental design and prioritisation decisions. This paper goes on to identify opportunities for research in DBM, including DBM process models used in practice, the use of evolutionary operations techniques in DBM, prioritisation decisions, and the factors that surround the uptake of DBM.
Rolfe, B.; Frayman, Y.; Hodgson, P.; Webb, G. I.
Fault Detection in a Cold Forging Process Through Feature Extraction with a Neural Network
Proceedings of the IASTED International Conference on Artificial Intelligence and Applications (AIA 2002), pp. 155-159, ACTA Press, Benalmdena, Spain, 2002.
@inproceedings{RolfeFraymanHodgsonWebb02,
title = {Fault Detection in a Cold Forging Process Through Feature Extraction with a Neural Network},
author = {B. Rolfe and Y. Frayman and P. Hodgson and G. I. Webb},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the IASTED International Conference on Artificial Intelligence and Applications (AIA 2002)},
pages = {155-159},
publisher = {ACTA Press},
address = {Benalmdena, Spain},
abstract = {This paper investigates the application of neural networks to the recognition of lubrication defects typical to an industrial cold forging process employed by fastener manufacturers. The accurate recognition of lubrication errors, such as coating not being applied properly or damaged during material handling, is very important to the quality of the final product in fastener manufacture. Lubrication errors lead to increased forging loads and premature tool failure, as well as to increased defect sorting and the re-processing of the coated rod. The lubrication coating provides a barrier between the work material and the die during the drawing operation; moreover it needs be sufficiently robust to remain on the wire during the transfer to the cold forging operation. In the cold forging operation the wire undergoes multi-stage deformation without the application of any additional lubrication. Four types of lubrication errors, typical to production of fasteners, were introduced to a set of sample rods, which were subsequently drawn under laboratory conditions. The drawing force was measured, from which a limited set of features was extracted. The neural network based model learned from these features is able to recognize all types of lubrication errors to a high accuracy. The overall accuracy of the neural network model is around 98% with almost uniform distribution of errors between all four errors and the normal condition.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper investigates the application of neural networks to the recognition of lubrication defects typical to an industrial cold forging process employed by fastener manufacturers. The accurate recognition of lubrication errors, such as coating not being applied properly or damaged during material handling, is very important to the quality of the final product in fastener manufacture. Lubrication errors lead to increased forging loads and premature tool failure, as well as to increased defect sorting and the re-processing of the coated rod. The lubrication coating provides a barrier between the work material and the die during the drawing operation; moreover it needs be sufficiently robust to remain on the wire during the transfer to the cold forging operation. In the cold forging operation the wire undergoes multi-stage deformation without the application of any additional lubrication. Four types of lubrication errors, typical to production of fasteners, were introduced to a set of sample rods, which were subsequently drawn under laboratory conditions. The drawing force was measured, from which a limited set of features was extracted. The neural network based model learned from these features is able to recognize all types of lubrication errors to a high accuracy. The overall accuracy of the neural network model is around 98% with almost uniform distribution of errors between all four errors and the normal condition.
Pearce, J. E.; Shaw, R. N.; Webb, G. I.; Garner, B.
Experimentation and self learning in continuous database marketing
Proceedings of the IEEE International Conference on Data Mining (ICDM-2002), pp. 775-778, IEEE Comput. Soc, 2002.
@inproceedings{PearceWebbShawGarner02b,
title = {Experimentation and self learning in continuous database marketing},
author = {J. E. Pearce and R. N. Shaw and G. I. Webb and B. Garner},
doi = {10.1109/icdm.2002.1184051},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the IEEE International Conference on Data Mining (ICDM-2002)},
pages = {775-778},
publisher = {IEEE Comput. Soc},
series = {ICDM-02},
abstract = {We present a method for continuous database marketing that identifies target customers for a number of marketing offers using predictive models. The algorithm then selects the appropriate offer for the customer. Experimental design principles are encapsulated to capture more information that will be used to monitor and refine the predictive models. The updated predictive models are then used for the next round of marketing offers.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We present a method for continuous database marketing that identifies target customers for a number of marketing offers using predictive models. The algorithm then selects the appropriate offer for the customer. Experimental design principles are encapsulated to capture more information that will be used to monitor and refine the predictive models. The updated predictive models are then used for the next round of marketing offers.
Wang, Z.; Webb, G. I.
A Heuristic Lazy Bayesian Rules Algorithm
Simoff, S. J; Williams, G. J; Hegland, M. (Ed.): Proceedings of the First Australasian Data Mining Workshop (AusDM02), pp. 57-63, University of Technology, Canberra, Australia, 2002.
@inproceedings{WangWebb02b,
title = {A Heuristic Lazy Bayesian Rules Algorithm},
author = {Z. Wang and G. I. Webb},
editor = {S. J Simoff and G. J Williams and M. Hegland},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the First Australasian Data Mining Workshop (AusDM02)},
pages = {57-63},
publisher = {University of Technology},
address = {Canberra, Australia},
abstract = {Lazy Bayesian rule has demonstrated outstanding classification accuracy. However, it has high computational overheads when large numbers of instances are classified from a single training set. We compare lazy Bayesian rule and the tree-augmented Bayesian classifier, and present a new heuristic lazy Bayesian rule classifier that combines elements of the two. It requires less computation than lazy Bayesian rule, but demonstrates similar prediction accuracy.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Lazy Bayesian rule has demonstrated outstanding classification accuracy. However, it has high computational overheads when large numbers of instances are classified from a single training set. We compare lazy Bayesian rule and the tree-augmented Bayesian classifier, and present a new heuristic lazy Bayesian rule classifier that combines elements of the two. It requires less computation than lazy Bayesian rule, but demonstrates similar prediction accuracy.
Wang, Z.; Webb, G. I.
Comparison of Lazy Bayesian Rule Learning and Tree-Augmented Bayesian Learning
Proceedings of the IEEE International Conference on Data Mining (ICDM-2002), pp. 775-778, IEEE Computer Society, Maebashi City, Japan, 2002.
@inproceedings{WangWebb02,
title = {Comparison of Lazy Bayesian Rule Learning and Tree-Augmented Bayesian Learning},
author = {Z. Wang and G. I. Webb},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the IEEE International Conference on Data Mining (ICDM-2002)},
pages = {775-778},
publisher = {IEEE Computer Society},
address = {Maebashi City, Japan},
abstract = {The naive Bayes classifier is widely used in interactive applications due to its computational efficiency, direct theoretical base, and competitive accuracy. However, its attribute independence assumption can result in sub-optimal accuracy. A number of techniques have explored simple relaxations of the attribute independence assumption in order to increase accuracy. Among these, Lazy Bayesian Rules (LBR) and Tree-Augmented Na?ve-Bayes (TAN) have demonstrated strong prediction accuracy. However, their relative performance has never been evaluated. This paper compares and contrasts these two techniques, finding that they have comparable accuracy and hence should be selected according to computational profile. LBR is desirable when small numbers of objects are to be classified while TAN is desirable when large numbers of objects are to be classified},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The naive Bayes classifier is widely used in interactive applications due to its computational efficiency, direct theoretical base, and competitive accuracy. However, its attribute independence assumption can result in sub-optimal accuracy. A number of techniques have explored simple relaxations of the attribute independence assumption in order to increase accuracy. Among these, Lazy Bayesian Rules (LBR) and Tree-Augmented Na?ve-Bayes (TAN) have demonstrated strong prediction accuracy. However, their relative performance has never been evaluated. This paper compares and contrasts these two techniques, finding that they have comparable accuracy and hence should be selected according to computational profile. LBR is desirable when small numbers of objects are to be classified while TAN is desirable when large numbers of objects are to be classified
Webb, G. I.
Integrating Machine Learning with Knowledge Acquisition
Leondes, C. T. (Ed.): Expert Systems, vol. 3, pp. 937-959, Academic Press, San Diego, CA, 2002.
@incollection{Webb02,
title = {Integrating Machine Learning with Knowledge Acquisition},
author = {G. I. Webb},
editor = {C. T. Leondes},
year = {2002},
date = {2002-01-01},
booktitle = {Expert Systems},
volume = {3},
pages = {937-959},
publisher = {Academic Press},
address = {San Diego, CA},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Webb, G. I.; Boughton, J.; Wang, Z.
Averaged One-Dependence Estimators: Preliminary Results
Simoff, S. J.; Williams, G. J.; Hegland, M. (Ed.): Proceedings of the First Australasian Data Mining Workshop (AusDM02), pp. 65-73, University of Technology, Canberra, Australia, 2002.
@inproceedings{WebbBoughtonWang02,
title = {Averaged One-Dependence Estimators: Preliminary Results},
author = {G. I. Webb and J. Boughton and Z. Wang},
editor = {S. J. Simoff and G. J. Williams and M. Hegland},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the First Australasian Data Mining Workshop (AusDM02)},
pages = {65-73},
publisher = {University of Technology},
address = {Canberra, Australia},
abstract = {Naive Bayes is a simple, computationally efficient and remarkably accurate approach to classification learning. These properties have led to its wide deployment in many online applications. However, it is based on an assumption that all attributes are conditionally independent given the class. This assumption leads to decreased accuracy in some applications. AODE overcomes the attribute independence assumption of naive Bayes by averaging over all models in which all attributes depend upon the class and a single other attribute. The resulting classification learning algorithm for nominal data is computationally efficient and achieves very low error rates.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Naive Bayes is a simple, computationally efficient and remarkably accurate approach to classification learning. These properties have led to its wide deployment in many online applications. However, it is based on an assumption that all attributes are conditionally independent given the class. This assumption leads to decreased accuracy in some applications. AODE overcomes the attribute independence assumption of naive Bayes by averaging over all models in which all attributes depend upon the class and a single other attribute. The resulting classification learning algorithm for nominal data is computationally efficient and achieves very low error rates.
Webb, G. I.; Brain, D.
Generality is Predictive of Prediction Accuracy
Yamaguchi, T.; Hoffmann, A.; Motoda, H.; Compton, P. (Ed.): Proceedings of the 2002 Pacific Rim Knowledge Acquisition Workshop (PKAW'02), pp. 117-130, Japanese Society for Artificial Intelligence, Tokyo, Japan, 2002.
@inproceedings{WebbBrain02,
title = {Generality is Predictive of Prediction Accuracy},
author = {G. I. Webb and D. Brain},
editor = {T. Yamaguchi and A. Hoffmann and H. Motoda and P. Compton},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the 2002 Pacific Rim Knowledge Acquisition Workshop (PKAW'02)},
pages = {117-130},
publisher = {Japanese Society for Artificial Intelligence},
address = {Tokyo, Japan},
abstract = {There has been a dearth of research into the relative impacts of alternative high level learning biases. This paper presents two hypotheses about the expected impact of selecting between classification rules of differing levels of generality in the absence of other evidence about their likely relative performance on unseen data. It is argued that the accuracy on unseen data of the more general rule will tend to be closer to that of a default rule for the class than will that of the more specific rule. It is also argued that the accuracy on unseen cases of the more specific rule will tend to be closer to the accuracy obtained on training data than will the accuracy of the more general rule. Experimental evidence is provided in support of these hypotheses. We argue that these hypotheses can be of use in selecting appropriate learning biases to achieve specific learning objectives.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
There has been a dearth of research into the relative impacts of alternative high level learning biases. This paper presents two hypotheses about the expected impact of selecting between classification rules of differing levels of generality in the absence of other evidence about their likely relative performance on unseen data. It is argued that the accuracy on unseen data of the more general rule will tend to be closer to that of a default rule for the class than will that of the more specific rule. It is also argued that the accuracy on unseen cases of the more specific rule will tend to be closer to the accuracy obtained on training data than will the accuracy of the more general rule. Experimental evidence is provided in support of these hypotheses. We argue that these hypotheses can be of use in selecting appropriate learning biases to achieve specific learning objectives.
Webb, G. I.; Zhang, S.
Removing Trivial Associations in Association Rule Discovery
Proceedings of the First International NAISO Congress on Autonomous Intelligent Systems (ICAIS 2002), NAISO Academic Press, Geelong, Australia, 2002.
@inproceedings{WebbZhang02,
title = {Removing Trivial Associations in Association Rule Discovery},
author = {G. I. Webb and S. Zhang},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the First International NAISO Congress on Autonomous Intelligent Systems (ICAIS 2002)},
publisher = {NAISO Academic Press},
address = {Geelong, Australia},
abstract = {Association rule discovery has become one of the most widely applied data mining strategies. Techniques for association rule discovery have been dominated by the frequent itemset strategy as exemplified by the Apriori algorithm. One limitation of this approach is that it provides little opportunity to detect and remove association rules on the basis of relationships between rules. As a result, the association rules discovered are frequently swamped with large numbers of spurious rules that are of little interest to the user. This paper presents association rule discovery techniques that can detect and discard one form of spurious association rule: trivial associations.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Association rule discovery has become one of the most widely applied data mining strategies. Techniques for association rule discovery have been dominated by the frequent itemset strategy as exemplified by the Apriori algorithm. One limitation of this approach is that it provides little opportunity to detect and remove association rules on the basis of relationships between rules. As a result, the association rules discovered are frequently swamped with large numbers of spurious rules that are of little interest to the user. This paper presents association rule discovery techniques that can detect and discard one form of spurious association rule: trivial associations.
Yang, Y.; Webb, G. I.
Non-Disjoint Discretization for Naive-Bayes Classifiers
Sammut, C.; Hoffmann, A. G. (Ed.): Proceedings of the Nineteenth International Conference on Machine Learning (ICML '02), pp. 666-673, Morgan Kaufmann, Sydney, Australia, 2002.
@inproceedings{YangWebb02b,
title = {Non-Disjoint Discretization for Naive-Bayes Classifiers},
author = {Y. Yang and G. I. Webb},
editor = {C. Sammut and A. G. Hoffmann},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the Nineteenth International Conference on Machine Learning (ICML '02)},
pages = {666-673},
publisher = {Morgan Kaufmann},
address = {Sydney, Australia},
abstract = {Previous discretization techniques have discretized numeric attributes into disjoint intervals. We argue that this is neither necessary nor appropriate for naive-Bayes classifiers. The analysis leads to a new discretization method, Non-Disjoint Discretization (NDD). NDD forms overlapping intervals for a numeric attribute, always locating a value toward the middle of its discretized interval to obtain more reliable probability estimation. It also adjusts the number and size of discretized intervals to the number of training instances, seeking an appropriate trade-off between bias and variance of probability estimation. We justify NDD in theory and test it on a wide cross-section of datasets. Our experimental results suggest that for naive-Bayes classifiers, NDD works better than alternative discretization approaches.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Previous discretization techniques have discretized numeric attributes into disjoint intervals. We argue that this is neither necessary nor appropriate for naive-Bayes classifiers. The analysis leads to a new discretization method, Non-Disjoint Discretization (NDD). NDD forms overlapping intervals for a numeric attribute, always locating a value toward the middle of its discretized interval to obtain more reliable probability estimation. It also adjusts the number and size of discretized intervals to the number of training instances, seeking an appropriate trade-off between bias and variance of probability estimation. We justify NDD in theory and test it on a wide cross-section of datasets. Our experimental results suggest that for naive-Bayes classifiers, NDD works better than alternative discretization approaches.
Yang, Y.; Webb, G. I.
A Comparative Study of Discretization Methods for Naive-Bayes Classifiers
Yamaguchi, T.; Hoffmann, A.; Motoda, H.; Compton, P. (Ed.): Proceedings of the 2002 Pacific Rim Knowledge Acquisition Workshop (PKAW'02), pp. 159-173, Japanese Society for Artificial Intelligence, Tokyo, Japan, 2002.
@inproceedings{YangWebb02a,
title = {A Comparative Study of Discretization Methods for Naive-Bayes Classifiers},
author = {Y. Yang and G. I. Webb},
editor = {T. Yamaguchi and A. Hoffmann and H. Motoda and P. Compton},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the 2002 Pacific Rim Knowledge Acquisition Workshop (PKAW'02)},
pages = {159-173},
publisher = {Japanese Society for Artificial Intelligence},
address = {Tokyo, Japan},
abstract = {Discretization is a popular approach to handling numeric attributes in machine learning. We argue that the requirements for effective discretization differ between naive-Bayes learning and many other learning algorithms. We evaluate the effectiveness with naive-Bayes classifiers of nine discretization methods, equal width discretization (EWD), equal frequency discretization (EFD), fuzzy discretization (FD), entropy minimization discretization (EMD), iterative discretization (ID), proportional k-interval discretization (PKID), lazy discretization (LD), non-disjoint discretization (NDD) and weighted proportional k-interval discretization (WPKID). It is found that in general naive-Bayes classifiers trained on data preprocessed by LD, NDD or WPKID achieve lower classification error than those trained on data preprocessed by the other discretization methods. But LD can not scale to large data. This study leads to a new discretization method, weighted non-disjoint discretization (WNDD) that combines WPKID and NDD's advantages. Our experiments show that among all the rival discretization methods, WNDD best helps naive-Bayes classifiers reduce average classification error.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Discretization is a popular approach to handling numeric attributes in machine learning. We argue that the requirements for effective discretization differ between naive-Bayes learning and many other learning algorithms. We evaluate the effectiveness with naive-Bayes classifiers of nine discretization methods, equal width discretization (EWD), equal frequency discretization (EFD), fuzzy discretization (FD), entropy minimization discretization (EMD), iterative discretization (ID), proportional k-interval discretization (PKID), lazy discretization (LD), non-disjoint discretization (NDD) and weighted proportional k-interval discretization (WPKID). It is found that in general naive-Bayes classifiers trained on data preprocessed by LD, NDD or WPKID achieve lower classification error than those trained on data preprocessed by the other discretization methods. But LD can not scale to large data. This study leads to a new discretization method, weighted non-disjoint discretization (WNDD) that combines WPKID and NDD's advantages. Our experiments show that among all the rival discretization methods, WNDD best helps naive-Bayes classifiers reduce average classification error.
Brain, D.; Webb, G. I.
The Need for Low Bias Algorithms in Classification Learning From Large Data Sets
Lecture Notes in Computer Science 2431: Principles of Data Mining and Knowledge Discovery: Proceedings of the Sixth European Conference (PKDD 2002), pp. 62-73, Springer-Verlag, Helsinki, Finland, 2002.
@inproceedings{BrainWebb02,
title = {The Need for Low Bias Algorithms in Classification Learning From Large Data Sets},
author = {D. Brain and G. I. Webb},
year = {2002},
date = {2002-01-01},
booktitle = {Lecture Notes in Computer Science 2431: Principles of Data Mining and Knowledge Discovery: Proceedings of the Sixth European Conference (PKDD 2002)},
pages = {62-73},
publisher = {Springer-Verlag},
address = {Helsinki, Finland},
abstract = {This paper reviews the appropriateness for application to large data sets of standard machine learning algorithms, which were mainly developed in the context of small data sets. Sampling and parallelization have proved useful means for reducing computation time when learning from large data sets. However, such methods assume that algorithms that were designed for use with what are now considered small data sets are also fundamentally suitable for large data sets. It is plausible that optimal learning from large data sets requires a different type of algorithm to optimal learning from small data sets. This paper investigates one respect in which data set size may affect the requirements of a learning algorithm the bias plus variance decomposition of classification error. Experiments show that learning from large data sets may be more effective when using an algorithm that places greater emphasis on bias management, rather than variance management},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper reviews the appropriateness for application to large data sets of standard machine learning algorithms, which were mainly developed in the context of small data sets. Sampling and parallelization have proved useful means for reducing computation time when learning from large data sets. However, such methods assume that algorithms that were designed for use with what are now considered small data sets are also fundamentally suitable for large data sets. It is plausible that optimal learning from large data sets requires a different type of algorithm to optimal learning from small data sets. This paper investigates one respect in which data set size may affect the requirements of a learning algorithm the bias plus variance decomposition of classification error. Experiments show that learning from large data sets may be more effective when using an algorithm that places greater emphasis on bias management, rather than variance management
Frayman, Y.; Rolfe, B.; Hodgson, P.; Webb, G. I.
Predicting The Rolling Force in Hot Steel Rolling Mill using an Ensemble Model
Proceedings of the Second IASTED International Conference on Artificial Intelligence and Applications (AIA '02), pp. 143-148, ACTA Press, Benalmdena, Spain, 2002.
@inproceedings{FraymanRolfeHodgsonWebb02c,
title = {Predicting The Rolling Force in Hot Steel Rolling Mill using an Ensemble Model},
author = {Y. Frayman and B. Rolfe and P. Hodgson and G. I. Webb},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the Second IASTED International Conference on Artificial Intelligence and Applications (AIA '02)},
pages = {143-148},
publisher = {ACTA Press},
address = {Benalmdena, Spain},
abstract = {Accurate prediction of the roll separating force is critical to assuring the quality of the final product in steel manufacturing. This paper presents an ensemble model that addresses these concerns. A stacked generalisation approach to ensemble modeling is used with two sets of the ensemble model members, the first set being learnt from the current input-output data of the hot rolling finishing mill, while another uses the available information on the previous coil in addition to the current information. Both sets of ensemble members include linear regression, multilayer perceptron, and k-nearest neighbor algorithms. A competitive selection model (multilayer perceptron) is then used to select the output from one of the ensemble members to be the final output of the ensemble model. The ensemble model created by such a stacked generalization is able to achieve extremely high accuracy in predicting the roll separation force with the average relative accuracy being within 1% of the actual measured roll force.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Accurate prediction of the roll separating force is critical to assuring the quality of the final product in steel manufacturing. This paper presents an ensemble model that addresses these concerns. A stacked generalisation approach to ensemble modeling is used with two sets of the ensemble model members, the first set being learnt from the current input-output data of the hot rolling finishing mill, while another uses the available information on the previous coil in addition to the current information. Both sets of ensemble members include linear regression, multilayer perceptron, and k-nearest neighbor algorithms. A competitive selection model (multilayer perceptron) is then used to select the output from one of the ensemble members to be the final output of the ensemble model. The ensemble model created by such a stacked generalization is able to achieve extremely high accuracy in predicting the roll separation force with the average relative accuracy being within 1% of the actual measured roll force.
Frayman, Y.; Rolfe, B.; Webb, G. I.
Solving Regression Problems using Competitive Ensemble Models
McKay, B.; Slaney, J. K. (Ed.): Lecture Notes in Computer Science Vol. 2557: Proceedings of the 15th Australian Joint Conference on Artificial Intelligence (AI 02), pp. 511-522, Springer, Canberra, Australia, 2002.
@inproceedings{FraymanRolfeWebb02,
title = {Solving Regression Problems using Competitive Ensemble Models},
author = {Y. Frayman and B. Rolfe and G. I. Webb},
editor = {B. McKay and J. K. Slaney},
year = {2002},
date = {2002-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 2557: Proceedings of the 15th Australian Joint Conference on Artificial Intelligence (AI 02)},
pages = {511-522},
publisher = {Springer},
address = {Canberra, Australia},
abstract = {The use of ensemble models in many problem domains has increased significantly in the last few years. The ensemble modelling, in particularly boosting, has shown a great promise in improving predictive performance of a model. Combining the ensemble members is normally done in a co-operative fashion where each of the ensemble members performs the same task and their predictions are aggregated to obtain the improved performance. However, it is also possible to combine the ensemble members in a competitive fashion where the best prediction of a relevant ensemble member is selected for a particular input. This option has been previously somewhat overlooked. The aim of this article is to investigate and compare the competitive and co-operative approaches to combining the models in the ensemble. A comparison is made between a competitive ensemble model and that of MARS with bagging, mixture of experts, hierarchical mixture of experts and a neural network ensemble over several public domain regression problems that have a high degree of nonlinearity and noise. The empirical results show a substantial advantage of competitive learning versus the co-operative learning for all the regression problems investigated. The requirements for creating the efficient ensembles and the available guidelines are also discussed.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The use of ensemble models in many problem domains has increased significantly in the last few years. The ensemble modelling, in particularly boosting, has shown a great promise in improving predictive performance of a model. Combining the ensemble members is normally done in a co-operative fashion where each of the ensemble members performs the same task and their predictions are aggregated to obtain the improved performance. However, it is also possible to combine the ensemble members in a competitive fashion where the best prediction of a relevant ensemble member is selected for a particular input. This option has been previously somewhat overlooked. The aim of this article is to investigate and compare the competitive and co-operative approaches to combining the models in the ensemble. A comparison is made between a competitive ensemble model and that of MARS with bagging, mixture of experts, hierarchical mixture of experts and a neural network ensemble over several public domain regression problems that have a high degree of nonlinearity and noise. The empirical results show a substantial advantage of competitive learning versus the co-operative learning for all the regression problems investigated. The requirements for creating the efficient ensembles and the available guidelines are also discussed.
Frayman, Y.; Rolfe, B.; Webb, G. I.
Improving an Inverse Model of Sheet Metal Forming by Neural Network Based Regression
Proceedings of the Design Engineering Technical Conferences and Computer and Information in Engineering Conference (DETC'02/ASME 2002), pp. 1-8, ASME Press, New York, 2002.
@inproceedings{FraymanRolfeWebb02b,
title = {Improving an Inverse Model of Sheet Metal Forming by Neural Network Based Regression},
author = {Y. Frayman and B. Rolfe and G. I. Webb},
year = {2002},
date = {2002-01-01},
booktitle = {Proceedings of the Design Engineering Technical Conferences and Computer and Information in Engineering Conference (DETC'02/ASME 2002)},
pages = {1-8},
publisher = {ASME Press},
address = {New York},
abstract = {The inverse model for a sheet metal forming process aims to determine the initial parameter levels required to form the final formed shape. This is a difficult problem that is usually approached by traditional methods such a finite element analysis. Formulating the problem as a classification problem makes is possible to use a well established classification algorithms such as decision trees. The classification is, however, generally based on a winner-takes-all approach when associating the output value with the corresponding class. On the other hand when formulating the problem as a regression task, all the output values are combined to produce the corresponding class value. For a multi-class problem, this may result in very different associations between the output of the model and the corresponding class. Such formulation makes it possible to use a well known regression algorithms such as neural networks.In this paper, we develop a neural network based inverse model of a sheet forming process, and compare its performance with that of a linear model. Both models are used in two modes: classification mode and a function estimation mode to investigate the advantage of re-formulating the problem as function estimation. This results in large improvements in the recognition rate of set-up parameters of a sheet metal forming process for both models, with a neural network model achieving much more accurate parameters recognition than a linear model},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The inverse model for a sheet metal forming process aims to determine the initial parameter levels required to form the final formed shape. This is a difficult problem that is usually approached by traditional methods such a finite element analysis. Formulating the problem as a classification problem makes is possible to use a well established classification algorithms such as decision trees. The classification is, however, generally based on a winner-takes-all approach when associating the output value with the corresponding class. On the other hand when formulating the problem as a regression task, all the output values are combined to produce the corresponding class value. For a multi-class problem, this may result in very different associations between the output of the model and the corresponding class. Such formulation makes it possible to use a well known regression algorithms such as neural networks.In this paper, we develop a neural network based inverse model of a sheet forming process, and compare its performance with that of a linear model. Both models are used in two modes: classification mode and a function estimation mode to investigate the advantage of re-formulating the problem as function estimation. This results in large improvements in the recognition rate of set-up parameters of a sheet metal forming process for both models, with a neural network model achieving much more accurate parameters recognition than a linear model
Wang, Z.; Webb, G. I.; Dai, H.
Implementation of Lazy Bayesian Rules in the Weka System
Software Technology Catering for 21st Century: Proceedings of the International Symposium on Future Software Technology (ISFST2001), pp. 204-208, Software Engineers Association, Zheng Zhou, China, 2001.
@inproceedings{WangWebbDai01,
title = {Implementation of Lazy Bayesian Rules in the Weka System},
author = {Z. Wang and G. I. Webb and H. Dai},
year = {2001},
date = {2001-01-01},
booktitle = {Software Technology Catering for 21st Century: Proceedings of the International Symposium on Future Software Technology (ISFST2001)},
pages = {204-208},
publisher = {Software Engineers Association},
address = {Zheng Zhou, China},
abstract = {The na?ve Bayesian classification algorithms were shown to be computationally efficient and surprisingly accurate when the conditional independence assumption on which they are based is violated. The lazy Bayesian rule is the application of lazy learning techniques to Bayesian tree induction, which supports a weaker conditional attribute independence assumption. The Weka system is a full, industrial-strength implementation of essentially almost the state-of-the-art machine learning techniques, and it contains a framework, in the form of a Java class library, which supports applications that use embedded machine learning and even the implementation of new learning schemes. In this paper, we mainly discuss the implementation of the algorithm of lazy Bayesian rule in Weka System, and introduce all the methods to be used in the Java class. This is the first lazy learning scheme implemented in Weka System.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The na?ve Bayesian classification algorithms were shown to be computationally efficient and surprisingly accurate when the conditional independence assumption on which they are based is violated. The lazy Bayesian rule is the application of lazy learning techniques to Bayesian tree induction, which supports a weaker conditional attribute independence assumption. The Weka system is a full, industrial-strength implementation of essentially almost the state-of-the-art machine learning techniques, and it contains a framework, in the form of a Java class library, which supports applications that use embedded machine learning and even the implementation of new learning schemes. In this paper, we mainly discuss the implementation of the algorithm of lazy Bayesian rule in Weka System, and introduce all the methods to be used in the Java class. This is the first lazy learning scheme implemented in Weka System.
Webb, G. I.; Zhang, S.
Further Pruning for Efficient Association Rule Discovery
Stumptner, M.; Corbett, D.; Brooks, M. J. (Ed.): Lecture Notes in Computer Science Vol. 2256: Proceedings of the 14th Australian Joint Conference on Artificial Intelligence (AI'01), pp. 605-618, Springer, Adelaide, Australia, 2001.
@inproceedings{WebbZhang01,
title = {Further Pruning for Efficient Association Rule Discovery},
author = {G. I. Webb and S. Zhang},
editor = {M. Stumptner and D. Corbett and M. J. Brooks},
year = {2001},
date = {2001-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 2256: Proceedings of the 14th Australian Joint Conference on Artificial Intelligence (AI'01)},
pages = {605-618},
publisher = {Springer},
address = {Adelaide, Australia},
abstract = {The Apriori algorithm's frequent itemset approach has become the standard approach to discovering association rules. However, the computation requirements of the frequent itemset approach are infeasible for dense data and the approach is unable to discover infrequent associations. OPUS_AR is an efficient algorithm for rule discovery that does not utilize frequent itemsets and hence avoids these problems. It can reduce search time by using additional constraints on the search space as well as constraints on itemset frequency. However, the effectiveness of the pruning rules used during search will determine the efficiency of its search. This paper presents and analyzes pruning rules for use with OPUS_AR. We demonstrate that application of OPUS_AR is feasible for a number of datasets for which application of the frequent itemset approach is infeasible and that the new pruning rules can reduce compute time by more than 40%.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The Apriori algorithm's frequent itemset approach has become the standard approach to discovering association rules. However, the computation requirements of the frequent itemset approach are infeasible for dense data and the approach is unable to discover infrequent associations. OPUS_AR is an efficient algorithm for rule discovery that does not utilize frequent itemsets and hence avoids these problems. It can reduce search time by using additional constraints on the search space as well as constraints on itemset frequency. However, the effectiveness of the pruning rules used during search will determine the efficiency of its search. This paper presents and analyzes pruning rules for use with OPUS_AR. We demonstrate that application of OPUS_AR is feasible for a number of datasets for which application of the frequent itemset approach is infeasible and that the new pruning rules can reduce compute time by more than 40%.
Webb, G. I.
Discovering Associations with Numeric Variables
Provost, F.; Srikant, R. (Ed.): Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2001)[short paper], pp. 383-388, The Association for Computing Machinery, San Francisco, CA, 2001.
@inproceedings{Webb01a,
title = {Discovering Associations with Numeric Variables},
author = {G. I. Webb},
editor = {F. Provost and R. Srikant},
url = {http://dl.acm.org/authorize?19861},
year = {2001},
date = {2001-01-01},
booktitle = {Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2001)[short paper]},
pages = {383-388},
publisher = {The Association for Computing Machinery},
address = {San Francisco, CA},
abstract = {This paper further develops Aumann and Lindell's [3] proposal for a variant of association rules for which the consequent is a numeric variable. It is argued that these rules can discover useful interactions with numeric data that cannot be discovered directly using traditional association rules with discretization. Alternative measures for identifying interesting rules are proposed. Efficient algorithms are presented that enable these rules to be discovered for dense data sets for which application of Auman and Lindell's algorithm is infeasible.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper further develops Aumann and Lindell's [3] proposal for a variant of association rules for which the consequent is a numeric variable. It is argued that these rules can discover useful interactions with numeric data that cannot be discovered directly using traditional association rules with discretization. Alternative measures for identifying interesting rules are proposed. Efficient algorithms are presented that enable these rules to be discovered for dense data sets for which application of Auman and Lindell's algorithm is infeasible.
Webb, G. I.
Candidate Elimination Criteria for Lazy Bayesian Rules
Stumptner, M.; Corbett, D.; Brooks, M. J. (Ed.): Lecture Notes in Computer Science Vol. 2256: Proceedings of the 14th Australian Joint Conference on Artificial Intelligence (AI'01), pp. 545-556, Springer, Adelaide, Australia, 2001.
@inproceedings{Webb01b,
title = {Candidate Elimination Criteria for Lazy Bayesian Rules},
author = {G. I. Webb},
editor = {M. Stumptner and D. Corbett and M. J. Brooks},
doi = {10.1007%2F3-540-45656-2_47},
year = {2001},
date = {2001-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 2256: Proceedings of the 14th Australian Joint Conference on Artificial Intelligence (AI'01)},
pages = {545-556},
publisher = {Springer},
address = {Adelaide, Australia},
abstract = {Lazy Bayesian Rules modify naive Bayesian classification to undo elements of the harmful attribute independence assumption. It has been shown to provide classification error comparable to boosting decision trees. This paper explores alternatives to the candidate elimination criterion employed within Lazy Bayesian Rules. Improvements over naive Bayes are consistent so long as the candidate elimination criteria ensures there is sufficient data for accurate probability estimation. However, the original candidate elimination criterion is demonstrated to provide better overall error reduction than the use of a minimum data subset size criterion.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Lazy Bayesian Rules modify naive Bayesian classification to undo elements of the harmful attribute independence assumption. It has been shown to provide classification error comparable to boosting decision trees. This paper explores alternatives to the candidate elimination criterion employed within Lazy Bayesian Rules. Improvements over naive Bayes are consistent so long as the candidate elimination criteria ensures there is sufficient data for accurate probability estimation. However, the original candidate elimination criterion is demonstrated to provide better overall error reduction than the use of a minimum data subset size criterion.
Webb, G. I.; Pazzani, M. J.; Billsus, D.
Machine learning for user modeling
User Modeling and User-Adapted Interaction, vol. 11, pp. 19-20, 2001.
@article{WebbPazzaniBillsus01,
title = {Machine learning for user modeling},
author = {G. I. Webb and M. J. Pazzani and D. Billsus},
doi = {10.1023/A:1011117102175},
year = {2001},
date = {2001-01-01},
journal = {User Modeling and User-Adapted Interaction},
volume = {11},
pages = {19-20},
publisher = {Springer},
address = {Netherlands},
abstract = {At first blush, user modeling appears to be a prime candidate for straight forward application of standard machine learning techniques. Observations of the user's behavior can provide training examples that a machine learning system can use to form a model designed to predict future actions. However, user modeling poses a number of challenges for machine learning that have hindered its application in user modeling, including: the need for large data sets; the need for labelled data; concept drift; and computational complexity. This paper examines each of these issues and reviews approaches to resolving them.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
At first blush, user modeling appears to be a prime candidate for straight forward application of standard machine learning techniques. Observations of the user's behavior can provide training examples that a machine learning system can use to form a model designed to predict future actions. However, user modeling poses a number of challenges for machine learning that have hindered its application in user modeling, including: the need for large data sets; the need for labelled data; concept drift; and computational complexity. This paper examines each of these issues and reviews approaches to resolving them.
Yang, Y.; Webb, G. I.
Proportional K-Interval Discretization for Naive-Bayes Classifiers
DeRaedt, L.; Flach, P. A. (Ed.): Lecture Notes in Computer Science 2167: Proceedings of the 12th European Conference on Machine Learning (ECML'01), pp. 564-575, Springer-Verlag, Freiburg, Germany, 2001.
@inproceedings{YangWebb01,
title = {Proportional K-Interval Discretization for Naive-Bayes Classifiers},
author = {Y. Yang and G. I. Webb},
editor = {L. DeRaedt and P. A. Flach},
year = {2001},
date = {2001-01-01},
booktitle = {Lecture Notes in Computer Science 2167: Proceedings of the 12th European Conference on Machine Learning (ECML'01)},
pages = {564-575},
publisher = {Springer-Verlag},
address = {Freiburg, Germany},
abstract = {This paper argues that two commonly-used discretization approaches, fixed k-interval discretization and entropy-based discretization have sub-optimal characteristics for naive-Bayes classification. This analysis leads to a new discretization method, Proportional k-Interval Discretization (PKID), which adjusts the number and size of discretized intervals to the number of training instances, thus seeks an appropriate trade-off between the bias and variance of the probability estimation for naive-Bayes classifiers. We justify PKID in theory, as well as test it on a wide cross-section of datasets. Our experimental results suggest that in comparison to its alternatives, PKID provides naive-Bayes classifiers competitive classification performance for smaller datasets and better classification performance for larger datasets.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper argues that two commonly-used discretization approaches, fixed k-interval discretization and entropy-based discretization have sub-optimal characteristics for naive-Bayes classification. This analysis leads to a new discretization method, Proportional k-Interval Discretization (PKID), which adjusts the number and size of discretized intervals to the number of training instances, thus seeks an appropriate trade-off between the bias and variance of the probability estimation for naive-Bayes classifiers. We justify PKID in theory, as well as test it on a wide cross-section of datasets. Our experimental results suggest that in comparison to its alternatives, PKID provides naive-Bayes classifiers competitive classification performance for smaller datasets and better classification performance for larger datasets.
Smith, P. A.; Webb, G. I.
The Efficacy of a Low-Level Program Visualization Tool for Teaching Programming Concepts to Novice C Programmers
Journal of Educational Computing Research, vol. 22, no. 2, pp. 187-215, 2000.
@article{SmithWebb00,
title = {The Efficacy of a Low-Level Program Visualization Tool for Teaching Programming Concepts to Novice C Programmers},
author = {P. A. Smith and G. I. Webb},
year = {2000},
date = {2000-01-01},
journal = {Journal of Educational Computing Research},
volume = {22},
number = {2},
pages = {187-215},
publisher = {Baywood Publishing},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Webb, G. I.
Efficient Search for Association Rules
Ramakrishnan, R.; Stolfo, S. (Ed.): Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2000), pp. 99-107, The Association for Computing Machinery, Boston, MA, 2000.
@inproceedings{Webb00b,
title = {Efficient Search for Association Rules},
author = {G. I. Webb},
editor = {R. Ramakrishnan and S. Stolfo},
year = {2000},
date = {2000-01-01},
booktitle = {Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2000)},
pages = {99-107},
publisher = {The Association for Computing Machinery},
address = {Boston, MA},
abstract = {This paper argues that for some applications direct search for association rules can be more efficient than the two stage process of the Apriori algorithm which first finds large item sets which are then used to identify associations. In particular, it is argued, Apriori can impose large computational overheads when the number of frequent itemsets is very large. This will often be the case when association rule analysis is performed on domains other than basket analysis or when it is performed for basket analysis with basket information augmented by other customer information. An algorithm is presented that is computationally efficient for association rule analysis during which the number of rules to be found can be constrained and all data can be maintained in memory.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper argues that for some applications direct search for association rules can be more efficient than the two stage process of the Apriori algorithm which first finds large item sets which are then used to identify associations. In particular, it is argued, Apriori can impose large computational overheads when the number of frequent itemsets is very large. This will often be the case when association rule analysis is performed on domains other than basket analysis or when it is performed for basket analysis with basket information augmented by other customer information. An algorithm is presented that is computationally efficient for association rule analysis during which the number of rules to be found can be constrained and all data can be maintained in memory.
Webb, G. I.
MultiBoosting: A Technique for Combining Boosting and Wagging
Machine Learning, vol. 40, no. 2, pp. 159-196, 2000.
@article{Webb00a,
title = {MultiBoosting: A Technique for Combining Boosting and Wagging},
author = {G. I. Webb},
doi = {10.1023/A:1007659514849},
year = {2000},
date = {2000-01-01},
journal = {Machine Learning},
volume = {40},
number = {2},
pages = {159-196},
publisher = {Springer},
address = {Netherlands},
abstract = {MultiBoosting is an extension to the highly successful AdaBoost technique for forming decision committees. MultiBoosting can be viewed as combining AdaBoost with wagging. It is able to harness both AdaBoost's high bias and variance reduction with wagging's superior variance reduction. Using C4.5 as the base learning algorithm, Multi-boosting is demonstrated to produce decision committees with lower error than either AdaBoost or wagging significantly more often than the reverse over a large representative cross-section of UCI data sets. It offers the further advantage over AdaBoost of suiting parallel execution.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
MultiBoosting is an extension to the highly successful AdaBoost technique for forming decision committees. MultiBoosting can be viewed as combining AdaBoost with wagging. It is able to harness both AdaBoost's high bias and variance reduction with wagging's superior variance reduction. Using C4.5 as the base learning algorithm, Multi-boosting is demonstrated to produce decision committees with lower error than either AdaBoost or wagging significantly more often than the reverse over a large representative cross-section of UCI data sets. It offers the further advantage over AdaBoost of suiting parallel execution.
Zheng, Z.; Webb, G. I.
Lazy Learning of Bayesian Rules
Machine Learning, vol. 41, no. 1, pp. 53-84, 2000.
@article{ZhengWebb00,
title = {Lazy Learning of Bayesian Rules},
author = {Z. Zheng and G. I. Webb},
doi = {10.1023/A:1007613203719},
year = {2000},
date = {2000-01-01},
journal = {Machine Learning},
volume = {41},
number = {1},
pages = {53-84},
publisher = {Springer},
address = {Netherlands},
abstract = {The naive Bayesian classifier provides a simple and effective approach to classifier learning, but its attribute independence assumption is often violated in the real world. A number of approaches have sought to alleviate this problem. A Bayesian tree learning algorithm builds a decision tree, and generates a local naive Bayesian classifier at each leaf. The tests leading to a leaf can alleviate attribute interdependencies for the local naive Bayesian classifier. However, Bayesian tree learning still suffers from the small disjunct problem of tree learning. While inferred Bayesian trees demonstrate low average prediction error rates, there is reason to believe that error rates will be higher for those leaves with few training examples. This paper proposes the application of lazy learning techniques to Bayesian tree induction and presents the resulting lazy Bayesian rule learning algorithm, called LBR. For each test example, it builds a most appropriate rule with a local naive Bayesian classifier as its consequent. It is demonstrated that the computational requirements of LBR are reasonable in a wide cross selection of natural domains. Experiments with these domains show that, on average, this new algorithm obtains lower error rates significantly more often than the reverse in comparison to a naive Bayesian classifier, C4.5, a Bayesian tree learning algorithm, a constructive Bayesian classifier that eliminates attributes and constructs new attributes using Cartesian products of existing nominal attributes, and a lazy decision tree learning algorithm. It also outperforms, although the result is not statistically significant, a selective naive Bayesian classifier.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The naive Bayesian classifier provides a simple and effective approach to classifier learning, but its attribute independence assumption is often violated in the real world. A number of approaches have sought to alleviate this problem. A Bayesian tree learning algorithm builds a decision tree, and generates a local naive Bayesian classifier at each leaf. The tests leading to a leaf can alleviate attribute interdependencies for the local naive Bayesian classifier. However, Bayesian tree learning still suffers from the small disjunct problem of tree learning. While inferred Bayesian trees demonstrate low average prediction error rates, there is reason to believe that error rates will be higher for those leaves with few training examples. This paper proposes the application of lazy learning techniques to Bayesian tree induction and presents the resulting lazy Bayesian rule learning algorithm, called LBR. For each test example, it builds a most appropriate rule with a local naive Bayesian classifier as its consequent. It is demonstrated that the computational requirements of LBR are reasonable in a wide cross selection of natural domains. Experiments with these domains show that, on average, this new algorithm obtains lower error rates significantly more often than the reverse in comparison to a naive Bayesian classifier, C4.5, a Bayesian tree learning algorithm, a constructive Bayesian classifier that eliminates attributes and constructs new attributes using Cartesian products of existing nominal attributes, and a lazy decision tree learning algorithm. It also outperforms, although the result is not statistically significant, a selective naive Bayesian classifier.
Newlands, D.; Webb, G. I.
Convex Hulls in Concept Induction
Zhong, N.; Zhou, L. (Ed.): Lecture Notes in Computer Science 1574: Methodologies for Knowledge Discovery and Data Mining - Proceedings of the Third Pacific-Asia Conference (PAKDD'99), pp. 306-316, Springer-Verlag, Beijing, China, 1999.
@inproceedings{NewlandsWebb99a,
title = {Convex Hulls in Concept Induction},
author = {D. Newlands and G. I. Webb},
editor = {N. Zhong and L. Zhou},
url = {http://link.springer.com/chapter/10.1007/3-540-48912-6_42},
year = {1999},
date = {1999-01-01},
booktitle = {Lecture Notes in Computer Science 1574: Methodologies for Knowledge Discovery and Data Mining - Proceedings of the Third Pacific-Asia Conference (PAKDD'99)},
pages = {306-316},
publisher = {Springer-Verlag},
address = {Beijing, China},
abstract = {This paper investigates modelling concepts as a few, large convex hulls rather than as many, small, axis-orthogonal divisions as is done by systems which currently dominate classification learning. It is argued that this approach produces classifiers which have less strong hypothesis language bias and which, because of the fewness of the concepts induced, are more understandable. The design of such a system is described and its performance is investigated. Convex hulls are shown to be a useful inductive generalisation technique offering rather different biases than well-known systems such as C4.5 and CN2. The types of domains where convex hulls can be usefully employed are described.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper investigates modelling concepts as a few, large convex hulls rather than as many, small, axis-orthogonal divisions as is done by systems which currently dominate classification learning. It is argued that this approach produces classifiers which have less strong hypothesis language bias and which, because of the fewness of the concepts induced, are more understandable. The design of such a system is described and its performance is investigated. Convex hulls are shown to be a useful inductive generalisation technique offering rather different biases than well-known systems such as C4.5 and CN2. The types of domains where convex hulls can be usefully employed are described.
Smith, P. A.; Webb, G. I.
Evaluation of Low-Level Program Visualisation for Teaching Novice C Programmers
Cumming, G.; Okamoto, T.; Gomez, L. (Ed.): Proceedings of the Seventh International Conference on Computers in Education (ICCE '99), pp. 385-392, IOS Press, Chiba, Japan, 1999.
@inproceedings{SmithWebb99,
title = {Evaluation of Low-Level Program Visualisation for Teaching Novice C Programmers},
author = {P. A. Smith and G. I. Webb},
editor = {G. Cumming and T. Okamoto and L. Gomez},
year = {1999},
date = {1999-01-01},
booktitle = {Proceedings of the Seventh International Conference on Computers in Education (ICCE '99)},
volume = {2},
pages = {385-392},
publisher = {IOS Press},
address = {Chiba, Japan},
abstract = {While several program visualisation tools aimed at novice programmers have been developed over the past decade there is little empirical evidence showing that novices actually benefit from their use (Mulholland, 1995). Bradman (Smith \& Webb, 1998) is a low-level program visualisation tool. We present an experiment that tests the efficacy of Bradman in assisting novice programmers learn programming concepts. We show that students with access to this lowlevel program visualisation tool achieved greater understanding of some programming concepts than those without access.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
While several program visualisation tools aimed at novice programmers have been developed over the past decade there is little empirical evidence showing that novices actually benefit from their use (Mulholland, 1995). Bradman (Smith & Webb, 1998) is a low-level program visualisation tool. We present an experiment that tests the efficacy of Bradman in assisting novice programmers learn programming concepts. We show that students with access to this lowlevel program visualisation tool achieved greater understanding of some programming concepts than those without access.
Ting, K. M.; Zheng, Z.; Webb, G. I.
Learning Lazy Rules to Improve the Performance of Classifiers
Coenen, F.; Macintosh, A. (Ed.): Proceedings of the Nineteenth SGES International Conference on Knowledge Based Systems and Applied Artificial Intelligence (ES'99), pp. 122-131, Springer, Peterhouse College, Cambridge, UK, 1999.
@inproceedings{TingZhengWebb99,
title = {Learning Lazy Rules to Improve the Performance of Classifiers},
author = {K. M. Ting and Z. Zheng and G. I. Webb},
editor = {F. Coenen and A. Macintosh},
year = {1999},
date = {1999-01-01},
booktitle = {Proceedings of the Nineteenth SGES International Conference on Knowledge Based Systems and Applied Artificial Intelligence (ES'99)},
pages = {122-131},
publisher = {Springer},
address = {Peterhouse College, Cambridge, UK},
abstract = {Based on an earlier study on lazy Bayesian rule learning, this paper introduces a general lazy learning framework, called LAZYRULE, that begins to learn a rule only when classifying a test case. The objective of the framework is to improve the performance of a base learning algorithm. It has the potential to be used for different types of base learning algorithms. LAZYRULE performs attribute elimination and training case selection using cross-validation to generate the most appropriate rule for each test case. At the consequent of the rule, it applies the base learning algorithm on the selected training subset and the remaining attributes to construct a classifier to make a prediction. This combined action seeks to build a better performing classifier for each test case than the classifier trained using all attributes and all training cases. We show empirically that LAZYRULE improves the performances of naive Bayesian classifiers and majority vote.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Based on an earlier study on lazy Bayesian rule learning, this paper introduces a general lazy learning framework, called LAZYRULE, that begins to learn a rule only when classifying a test case. The objective of the framework is to improve the performance of a base learning algorithm. It has the potential to be used for different types of base learning algorithms. LAZYRULE performs attribute elimination and training case selection using cross-validation to generate the most appropriate rule for each test case. At the consequent of the rule, it applies the base learning algorithm on the selected training subset and the remaining attributes to construct a classifier to make a prediction. This combined action seeks to build a better performing classifier for each test case than the classifier trained using all attributes and all training cases. We show empirically that LAZYRULE improves the performances of naive Bayesian classifiers and majority vote.
Webb, G. I.
Decision Tree Grafting From The All Tests But One Partition
Dean, T. (Ed.): Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence (IJCAI 99), pp. 702-707, Morgan Kaufmann, Stockholm, Sweden, 1999.
@inproceedings{Webb99,
title = {Decision Tree Grafting From The All Tests But One Partition},
author = {G. I. Webb},
editor = {T. Dean},
year = {1999},
date = {1999-01-01},
booktitle = {Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence (IJCAI 99)},
pages = {702-707},
publisher = {Morgan Kaufmann},
address = {Stockholm, Sweden},
abstract = {Decision tree grafting adds nodes to an existing decision tree with the objective of reducing prediction error. A new grafting algorithm is presented that considers one set of training data only for each leaf of the initial decision tree, the set of cases that fail at most one test on the path to the leaf. This new technique is demonstrated to retain the error reduction power of the original grafting algorithm while dramatically reducing compute time and the complexity of the inferred tree. Bias/variance analysis reveal that the original grafting technique operated primarily by variance reduction while the new technique reduces both bias and variance.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Decision tree grafting adds nodes to an existing decision tree with the objective of reducing prediction error. A new grafting algorithm is presented that considers one set of training data only for each leaf of the initial decision tree, the set of cases that fail at most one test on the path to the leaf. This new technique is demonstrated to retain the error reduction power of the original grafting algorithm while dramatically reducing compute time and the complexity of the inferred tree. Bias/variance analysis reveal that the original grafting technique operated primarily by variance reduction while the new technique reduces both bias and variance.
Webb, G. I.; Wells, J.; Zheng, Z.
An Experimental Evaluation of Integrating Machine Learning with Knowledge Acquisition
Machine Learning, vol. 35, no. 1, pp. 5-24, 1999.
@article{WebbWellsZheng99,
title = {An Experimental Evaluation of Integrating Machine Learning with Knowledge Acquisition},
author = {G. I. Webb and J. Wells and Z. Zheng},
year = {1999},
date = {1999-01-01},
journal = {Machine Learning},
volume = {35},
number = {1},
pages = {5-24},
publisher = {Springer},
address = {Netherlands},
abstract = {Machine learning and knowledge acquisition from experts have distinct capabilities that appear to complement one another. We report a study that demonstrates the integration of these approaches can both improve the accuracy of the developed knowledge base and reduce development time. In addition, we found that users expected the expert systems created through the integrated approach to have higher accuracy than those created without machine learning and rated the integrated approach less difficult to use. They also provided favorable evaluations of both the specific integrated software, system called The Knowledge Factory, and of the general value of machine learning for knowledge acquisition.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Machine learning and knowledge acquisition from experts have distinct capabilities that appear to complement one another. We report a study that demonstrates the integration of these approaches can both improve the accuracy of the developed knowledge base and reduce development time. In addition, we found that users expected the expert systems created through the integrated approach to have higher accuracy than those created without machine learning and rated the integrated approach less difficult to use. They also provided favorable evaluations of both the specific integrated software, system called The Knowledge Factory, and of the general value of machine learning for knowledge acquisition.
Zheng, Z.; Webb, G. I.
Stochastic Attribute Selection Committees with Multiple Boosting: Learning More Accurate and More Stable Classifier Committees
Zhong, N.; Zhou, L. (Ed.): Lecture Notes in Computer Science 1574: Methodologies for Knowledge Discovery and Data Mining - Proceedings of the Third Pacific-Asia Conference (PAKDD'99), pp. 123-132, Springer-Verlag, Beijing, China, 1999.
@inproceedings{ZhengWebb99b,
title = {Stochastic Attribute Selection Committees with Multiple Boosting: Learning More Accurate and More Stable Classifier Committees},
author = {Z. Zheng and G. I. Webb},
editor = {N. Zhong and L. Zhou},
year = {1999},
date = {1999-01-01},
booktitle = {Lecture Notes in Computer Science 1574: Methodologies for Knowledge Discovery and Data Mining - Proceedings of the Third Pacific-Asia Conference (PAKDD'99)},
pages = {123-132},
publisher = {Springer-Verlag},
address = {Beijing, China},
abstract = {Classifier learning is a key technique for KDD. Approaches to learning classifier committees, including Boosting, Bagging, SASC, and SASCB, have demonstrated great success in increasing the prediction accuracy of decision trees. Boosting and Bagging create different classifiers by modifying the distribution of the training set. SASC adopts a different method. It generates committees by stochastic manipulation of the set of attributes considered at each node during tree induction, but keeping the distribution of the training set unchanged. SASCB, a combination of Boosting and SASC, has shown the ability to further increase, on average, the prediction accuracy of decision trees. It has been found that the performance of SASCB and Boosting is more variable than that of SASC, although SASCB is more accurate than the others on average. In this paper, we present a novel method to reduce variability of SASCB and Boosting, and further increase their average accuracy. It generates multiple committees by incorporating Bagging into SASCB. As well as improving stability and average accuracy, the resulting method is amenable to parallel or distributed processing, while Boosting and SascB are not. This is an important characteristic for datamining in large datasets.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Classifier learning is a key technique for KDD. Approaches to learning classifier committees, including Boosting, Bagging, SASC, and SASCB, have demonstrated great success in increasing the prediction accuracy of decision trees. Boosting and Bagging create different classifiers by modifying the distribution of the training set. SASC adopts a different method. It generates committees by stochastic manipulation of the set of attributes considered at each node during tree induction, but keeping the distribution of the training set unchanged. SASCB, a combination of Boosting and SASC, has shown the ability to further increase, on average, the prediction accuracy of decision trees. It has been found that the performance of SASCB and Boosting is more variable than that of SASC, although SASCB is more accurate than the others on average. In this paper, we present a novel method to reduce variability of SASCB and Boosting, and further increase their average accuracy. It generates multiple committees by incorporating Bagging into SASCB. As well as improving stability and average accuracy, the resulting method is amenable to parallel or distributed processing, while Boosting and SascB are not. This is an important characteristic for datamining in large datasets.
Zheng, Z.; Webb, G. I.; Ting, K. M.
Lazy Bayesian Rules: A Lazy Semi-Naive Bayesian Learning Technique Competitive to Boosting Decision Trees
Bratko, I.; Dzeroski, S. (Ed.): Proceedings of the Sixteenth International Conference on Machine Learning (ICML-99), pp. 493-502, Morgan Kaufmann, Bled, Slovenia, 1999.
@inproceedings{ZhengWebbTing99,
title = {Lazy Bayesian Rules: A Lazy Semi-Naive Bayesian Learning Technique Competitive to Boosting Decision Trees},
author = {Z. Zheng and G. I. Webb and K. M. Ting},
editor = {I. Bratko and S. Dzeroski},
year = {1999},
date = {1999-01-01},
booktitle = {Proceedings of the Sixteenth International Conference on Machine Learning (ICML-99)},
pages = {493-502},
publisher = {Morgan Kaufmann},
address = {Bled, Slovenia},
abstract = {LBR is a lazy semi-naive Bayesian classifier learning technique, designed to alleviate the attribute interdependence problem of naive Bayesian classification. To classify a test example, it creates a conjunctive rule that selects a most appropriate subset of training examples and induces a local naive Bayesian classifier using this subset. LBR can significantly improve the performance of the naive Bayesian classifier. A bias and variance analysis of LBR reveals that it significantly reduces the bias of naive Bayesian classification at a cost of a slight increase in variance. It is interesting to compare this lazy technique with boosting and bagging, two well-known state-of-the-art non-lazy learning techniques. Empirical comparison of LBR with boosting decision trees on discrete valued data shows that LBR has, on average, significantly lower variance and higher bias. As a result of the interaction of these effects, the average prediction error of LBR over a range of learning tasks is at a level directly comparable to boosting. LBR provides a very competitive discrete valued learning technique where error minimization is the primary concern. It is very efficient when a single classifier is to be applied to classify few cases, such as in a typical incremental learning scenario.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
LBR is a lazy semi-naive Bayesian classifier learning technique, designed to alleviate the attribute interdependence problem of naive Bayesian classification. To classify a test example, it creates a conjunctive rule that selects a most appropriate subset of training examples and induces a local naive Bayesian classifier using this subset. LBR can significantly improve the performance of the naive Bayesian classifier. A bias and variance analysis of LBR reveals that it significantly reduces the bias of naive Bayesian classification at a cost of a slight increase in variance. It is interesting to compare this lazy technique with boosting and bagging, two well-known state-of-the-art non-lazy learning techniques. Empirical comparison of LBR with boosting decision trees on discrete valued data shows that LBR has, on average, significantly lower variance and higher bias. As a result of the interaction of these effects, the average prediction error of LBR over a range of learning tasks is at a level directly comparable to boosting. LBR provides a very competitive discrete valued learning technique where error minimization is the primary concern. It is very efficient when a single classifier is to be applied to classify few cases, such as in a typical incremental learning scenario.
Brain, D.; Webb, G. I.
On The Effect of Data Set Size on Bias And Variance in Classification Learning
Richards, D.; Beydoun, G.; Hoffmann, A.; Compton, P. (Ed.): Proceedings of the Fourth Australian Knowledge Acquisition Workshop (AKAW-99), pp. 117-128, The University of New South Wales, Sydney, Australia, 1999.
@inproceedings{BrainWebb99,
title = {On The Effect of Data Set Size on Bias And Variance in Classification Learning},
author = {D. Brain and G. I. Webb},
editor = {D. Richards and G. Beydoun and A. Hoffmann and P. Compton},
year = {1999},
date = {1999-01-01},
booktitle = {Proceedings of the Fourth Australian Knowledge Acquisition Workshop (AKAW-99)},
pages = {117-128},
publisher = {The University of New South Wales},
address = {Sydney, Australia},
abstract = {With the advent of data mining, machine learning has come of age and is now a critical technology in many businesses. However, machine learning evolved in a different research context to that in which it now finds itself employed. A particularly important problem in the data mining world is working effectively with large data sets. However, most machine learning research has been conducted in the context of learning from very small data sets. To date most approaches to scaling up machine learning to large data sets have attempted to modify existing algorithms to deal with large data sets in a more computationally efficient and effective manner. But is this necessarily the best method? This paper explores the possibility of designing algorithms specifically for large data sets. Specifically, the paper looks at how increasing data set size affects bias and variance error decompositions for classification algorithms. Preliminary results of experiments to determine these effects are presented, showing that, as hypothesized variance can be expected to decrease as training set size increases. No clear effect of training set size on bias was observed. These results have profound implications for data mining from large data sets, indicating that developing effective learning algorithms for large data sets is not simply a matter of finding computationally efficient variants of existing learning algorithms.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
With the advent of data mining, machine learning has come of age and is now a critical technology in many businesses. However, machine learning evolved in a different research context to that in which it now finds itself employed. A particularly important problem in the data mining world is working effectively with large data sets. However, most machine learning research has been conducted in the context of learning from very small data sets. To date most approaches to scaling up machine learning to large data sets have attempted to modify existing algorithms to deal with large data sets in a more computationally efficient and effective manner. But is this necessarily the best method? This paper explores the possibility of designing algorithms specifically for large data sets. Specifically, the paper looks at how increasing data set size affects bias and variance error decompositions for classification algorithms. Preliminary results of experiments to determine these effects are presented, showing that, as hypothesized variance can be expected to decrease as training set size increases. No clear effect of training set size on bias was observed. These results have profound implications for data mining from large data sets, indicating that developing effective learning algorithms for large data sets is not simply a matter of finding computationally efficient variants of existing learning algorithms.
Chiu, B. C.; Webb, G. I.
Dual-Model: An Architecture for Utilizing Temporal Information in Student Modeling
Cumming, G.; Okamoto, T.; Gomez, L. (Ed.): Proceedings of the Seventh International Conference on Computers in Education (ICCE '99), pp. 111-118, IOS Press, Chiba, Japan.(Also appeared in the Proceedings of ACAI Workshop W03: Machine Learning in User Modeling, pp 46-53), 1999.
@inproceedings{ChiuWebb99a,
title = {Dual-Model: An Architecture for Utilizing Temporal Information in Student Modeling},
author = {B. C. Chiu and G. I. Webb},
editor = {G. Cumming and T. Okamoto and L. Gomez},
year = {1999},
date = {1999-01-01},
booktitle = {Proceedings of the Seventh International Conference on Computers in Education (ICCE '99)},
volume = {1},
pages = {111-118},
publisher = {IOS Press},
address = {Chiba, Japan.(Also appeared in the Proceedings of ACAI Workshop W03: Machine Learning in User Modeling, pp 46-53)},
abstract = {A modeling system may be required to predict an agent's future actions even when confronted by inadequate or contradictory relevant evidence from observations of past actions. This can result in low prediction accuracy, or otherwise, low prediction rates, leaving a set of cases for which no predictions are made. This raises two issues. First, when maximizing prediction rate is preferable, what mechanisms can be employed such that a system can make more predictions without severely degrading prediction accuracy? Second, for contexts in which accuracy is of primary importance, how can we further improve prediction accuracy? A recently proposed Dual-model approach, which takes models' temporal characteristics into account, suggests a solution to the first problem, but leaves room for further improvement. This paper presents two classes of Dual-model variant. Each aims to achieve one of the above objectives. With the performance of the original system as a baseline, which does not utilize the temporal information, empirical evaluations in the domain of elementary subtraction show that one class of variant outperforms the baseline in prediction rate while the other does so in prediction accuracy, without significantly affecting other overall measures of the original performance. Keywords: Agent modeling, Student modeling, Temporal model, Decision tree.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
A modeling system may be required to predict an agent's future actions even when confronted by inadequate or contradictory relevant evidence from observations of past actions. This can result in low prediction accuracy, or otherwise, low prediction rates, leaving a set of cases for which no predictions are made. This raises two issues. First, when maximizing prediction rate is preferable, what mechanisms can be employed such that a system can make more predictions without severely degrading prediction accuracy? Second, for contexts in which accuracy is of primary importance, how can we further improve prediction accuracy? A recently proposed Dual-model approach, which takes models' temporal characteristics into account, suggests a solution to the first problem, but leaves room for further improvement. This paper presents two classes of Dual-model variant. Each aims to achieve one of the above objectives. With the performance of the original system as a baseline, which does not utilize the temporal information, empirical evaluations in the domain of elementary subtraction show that one class of variant outperforms the baseline in prediction rate while the other does so in prediction accuracy, without significantly affecting other overall measures of the original performance. Keywords: Agent modeling, Student modeling, Temporal model, Decision tree.
Smith, P. A.; Webb, G. I.
Overview of a Low-Level Program Visualisation Tool for Novice Programmers
Proceedings of the Sixth International Conference on Computers in Education (ICCE '98), pp. 213-216, Springer-Verlag, Beijing, 1998.
@inproceedings{SmithWebb98,
title = {Overview of a Low-Level Program Visualisation Tool for Novice Programmers},
author = {P. A. Smith and G. I. Webb},
year = {1998},
date = {1998-01-01},
booktitle = {Proceedings of the Sixth International Conference on Computers in Education (ICCE '98)},
pages = {213-216},
publisher = {Springer-Verlag},
address = {Beijing},
abstract = {As a programming novice attempts to attain expertise in programming she must develop adequate mental models and knowledge structures of the programming process. Unfortunately, many of the computerised tools to which novice programmers have access are designed by expert programmers for experts and as such do not meet the needs of novices. Low-level program visualisation tools make explicit the internal workings of program execution and as such can serve as conceptual models onto which novices can assimilate information about programming. This paper discusses the need for such a tool, what features such a tool may include and gives a brief description of an evaluation of a low-level program visualisation tool developed at Deakin University.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
As a programming novice attempts to attain expertise in programming she must develop adequate mental models and knowledge structures of the programming process. Unfortunately, many of the computerised tools to which novice programmers have access are designed by expert programmers for experts and as such do not meet the needs of novices. Low-level program visualisation tools make explicit the internal workings of program execution and as such can serve as conceptual models onto which novices can assimilate information about programming. This paper discusses the need for such a tool, what features such a tool may include and gives a brief description of an evaluation of a low-level program visualisation tool developed at Deakin University.
Viswanathan, M.; Webb, G. I.
Classification Learning Using All Rules
Nedellec, C.; Rouveiro, C. (Ed.): Lecture Notes in Computer Science 1398: Proceedings of the Tenth European Conference on Machine Learning (ECML'98), pp. 149-159, Springer, Chemnitz, Germany, 1998.
@inproceedings{ViswanathanWebb98,
title = {Classification Learning Using All Rules},
author = {M. Viswanathan and G. I. Webb},
editor = {C. Nedellec and C. Rouveiro},
doi = {10.1007/BFb0026685},
year = {1998},
date = {1998-01-01},
booktitle = {Lecture Notes in Computer Science 1398: Proceedings of the Tenth European Conference on Machine Learning (ECML'98)},
pages = {149-159},
publisher = {Springer},
address = {Chemnitz, Germany},
abstract = {The covering algorithm has been ubiquitous in the induction of classification rules. This approach to machine learning uses heuristic search that seeks to find a minimum number of rules that adequately explains the data. However, recent research has provided evidence that learning redundant classifiers can increase predictive accuracy. Learning all possible classifiers seems to be a plausible form of this nomination of redundant classifiers. This paper presents an algorithm that in effect learns all classifiers. Preliminary investigations by Webb (1996b) suggest that a heuristic covering algorithm in general learns classification rules with higher predictive accuracy than those learned by this new approach. In this paper we present an extensive empirical comparison between the learning-all-rules algorithm and three varied established approaches to inductive learning, namely a covering algorithm, an instance-based learner and a decision tree learner. Empirical evaluation provides strong evidence in support of learning-all-rules as a plausible approach to inductive learning.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The covering algorithm has been ubiquitous in the induction of classification rules. This approach to machine learning uses heuristic search that seeks to find a minimum number of rules that adequately explains the data. However, recent research has provided evidence that learning redundant classifiers can increase predictive accuracy. Learning all possible classifiers seems to be a plausible form of this nomination of redundant classifiers. This paper presents an algorithm that in effect learns all classifiers. Preliminary investigations by Webb (1996b) suggest that a heuristic covering algorithm in general learns classification rules with higher predictive accuracy than those learned by this new approach. In this paper we present an extensive empirical comparison between the learning-all-rules algorithm and three varied established approaches to inductive learning, namely a covering algorithm, an instance-based learner and a decision tree learner. Empirical evaluation provides strong evidence in support of learning-all-rules as a plausible approach to inductive learning.
Webb, G. I.; Kuzmycz, M.
Evaluation Of Data Aging: A Technique For Discounting Old Data During Student Modeling
Goettl, B. P.; Halff, H. M.; Redfield, C.; Shute, V. (Ed.): Lecture Notes in Computer Science Vol. 1452: Proceedings of the Fourth International Conference on Intelligent Tutoring Systems (ITS '98), pp. 384-393, Springer-Verlag, San Antonio, Texas, 1998.
@inproceedings{WebbKuzmycz98,
title = {Evaluation Of Data Aging: A Technique For Discounting Old Data During Student Modeling},
author = {G. I. Webb and M. Kuzmycz},
editor = {B. P. Goettl and H. M. Halff and C. Redfield and V. Shute},
year = {1998},
date = {1998-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 1452: Proceedings of the Fourth International Conference on Intelligent Tutoring Systems (ITS '98)},
pages = {384-393},
publisher = {Springer-Verlag},
address = {San Antonio, Texas},
abstract = {Student modeling systems must operate in an environment in which a student's mastery of a subject matter is likely to change as a lesson progresses. A student model is formed from evaluation of evidence about the student's mastery of the domain. However, given that such mastery will change, older evidence is likely to be less valuable than recent evidence. Data aging addresses this issue by discounting the value of older evidence. This paper provides experimental evaluation of the effects of data aging. While it is demonstrated that data aging can result in statistically significant increases in both the number and accuracy of predictions that a modeling system makes, it is also demonstrated that the reverse can be true. Further, the effects experienced are of only small magnitude. It is argued that these results demonstrate some potential for data aging as a general strategy, but do not warrant employing data aging in its current form.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Student modeling systems must operate in an environment in which a student's mastery of a subject matter is likely to change as a lesson progresses. A student model is formed from evaluation of evidence about the student's mastery of the domain. However, given that such mastery will change, older evidence is likely to be less valuable than recent evidence. Data aging addresses this issue by discounting the value of older evidence. This paper provides experimental evaluation of the effects of data aging. While it is demonstrated that data aging can result in statistically significant increases in both the number and accuracy of predictions that a modeling system makes, it is also demonstrated that the reverse can be true. Further, the effects experienced are of only small magnitude. It is argued that these results demonstrate some potential for data aging as a general strategy, but do not warrant employing data aging in its current form.
Webb, G. I.
The Problem of Missing Values in Decision Tree Grafting
Antoniou, G.; Slaney, J. K. (Ed.): Lecture Notes in Computer Science Vol. 1502: Advanced Topics in Artificial Intelligence, Selected Papers from the Eleventh Australian Joint Conference on Artificial Intelligence (AI '98), pp. 273-283, Springer-Verlag, Brisbane, Australia, 1998.
@inproceedings{Webb98,
title = {The Problem of Missing Values in Decision Tree Grafting},
author = {G. I. Webb},
editor = {G. Antoniou and J. K. Slaney},
year = {1998},
date = {1998-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 1502: Advanced Topics in Artificial Intelligence, Selected Papers from the Eleventh Australian Joint Conference on Artificial Intelligence (AI '98)},
pages = {273-283},
publisher = {Springer-Verlag},
address = {Brisbane, Australia},
abstract = {Decision tree grafting adds nodes to inferred decision trees. Previous research has demonstrated that appropriate grafting techniques can improve predictive accuracy across a wide cross-selection of domains. However, previous decision tree grafting systems are demonstrated to have a serious deficiency for some data sets containing missing values. This problem arises due to the method for handling missing values employed by C4.5, in which the grafting systems have been embedded. This paper provides an explanation of and solution to the problem. Experimental evidence is presented of the efficacy of this solution.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Decision tree grafting adds nodes to inferred decision trees. Previous research has demonstrated that appropriate grafting techniques can improve predictive accuracy across a wide cross-selection of domains. However, previous decision tree grafting systems are demonstrated to have a serious deficiency for some data sets containing missing values. This problem arises due to the method for handling missing values employed by C4.5, in which the grafting systems have been embedded. This paper provides an explanation of and solution to the problem. Experimental evidence is presented of the efficacy of this solution.
Webb, G. I.
Preface to UMUAI Special Issue on Machine Learning for User Modeling
User Modeling and User-Adapted Interaction, vol. 8, no. 1, pp. 1-3, 1998.
@article{Webb98a,
title = {Preface to UMUAI Special Issue on Machine Learning for User Modeling},
author = {G. I. Webb},
year = {1998},
date = {1998-01-01},
journal = {User Modeling and User-Adapted Interaction},
volume = {8},
number = {1},
pages = {1-3},
publisher = {Kluwer Academic Publishers},
address = {Netherlands},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Webb, G. I.; Pazzani, M.
Adjusted Probability Naive Bayesian Induction
Antoniou, G.; Slaney, J. K. (Ed.): Lecture Notes in Computer Science Vol. 1502: Advanced Topics in Artificial Intelligence, Selected Papers from the Eleventh Australian Joint Conference on Artificial Intelligence (AI '98), pp. 285-295, Springer-Verlag, Brisbane, Australia, 1998.
@inproceedings{WebbPazzani98,
title = {Adjusted Probability Naive Bayesian Induction},
author = {G. I. Webb and M. Pazzani},
editor = {G. Antoniou and J. K. Slaney},
year = {1998},
date = {1998-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 1502: Advanced Topics in Artificial Intelligence, Selected Papers from the Eleventh Australian Joint Conference on Artificial Intelligence (AI '98)},
pages = {285-295},
publisher = {Springer-Verlag},
address = {Brisbane, Australia},
abstract = {Naive Bayesian classifiers utilise a simple mathematical model for induction. While it is known that the assumptions on which this model is based are frequently violated, the predictive accuracy obtained in discriminate classification tasks is surprisingly competitive in comparison to more complex induction techniques. Adjusted probability naive Bayesian induction adds a simple extension to the naive Bayesian classifier. A numeric weight is inferred for each class. During discriminate classification, the naive Bayesian probability of a class is multiplied by its weight to obtain an adjusted value. The use of this adjusted value in place of the naive Bayesian probability is shown to significantly improve predictive accuracy.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Naive Bayesian classifiers utilise a simple mathematical model for induction. While it is known that the assumptions on which this model is based are frequently violated, the predictive accuracy obtained in discriminate classification tasks is surprisingly competitive in comparison to more complex induction techniques. Adjusted probability naive Bayesian induction adds a simple extension to the naive Bayesian classifier. A numeric weight is inferred for each class. During discriminate classification, the naive Bayesian probability of a class is multiplied by its weight to obtain an adjusted value. The use of this adjusted value in place of the naive Bayesian probability is shown to significantly improve predictive accuracy.
Zheng, Z.; Webb, G. I.
Stochastic Attribute Selection Committees
Antoniou, G.; Slaney, J. K. (Ed.): Lecture Notes in Computer Science Vol. 1502: Advanced Topics in Artificial Intelligence, Selected Papers from the Eleventh Australian Joint Conference on Artificial Intelligence (AI '98), pp. 321-332, Springer-Verlag, Brisbane, Australia, 1998.
@inproceedings{ZhengWebb98a,
title = {Stochastic Attribute Selection Committees},
author = {Z. Zheng and G. I. Webb},
editor = {G. Antoniou and J. K. Slaney},
year = {1998},
date = {1998-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 1502: Advanced Topics in Artificial Intelligence, Selected Papers from the Eleventh Australian Joint Conference on Artificial Intelligence (AI '98)},
pages = {321-332},
publisher = {Springer-Verlag},
address = {Brisbane, Australia},
abstract = {Classifier committee learning methods generate multiple classifiers to form a committee by repeated application of a single base learning algorithm. The committee members vote to decide the final classification. Two such methods, Bagging and Boosting, have shown great success with decision tree learning. They create different classifiers by modifying the distribution of the training set. This paper studies a different approach: Stochastic Attribute Selection Committee learning of decision trees. It generates classifier committees by stochastically modifying the set of attributes but keeping the distribution of the training set unchanged. An empirical evaluation of a variant of this method, namely Sasc, in a representative collection of natural domains shows that the SASC method can significantly reduce the error rate of decision tree learning. On average Sasc is more accurate than Bagging and less accurate than Boosting, although a one-tailed significance test fails to show that these differences are significant at a level of 0.05. In addition, it is found that, like Bagging, Sasc is more stable than Boosting in terms of less frequently obtaining significantly higher error rates than C4.5 and, when error is raised, producing lower error rate increases. Moreover, like Bagging, Sasc is amenable to parallel and distributed processing while Boosting is not.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Classifier committee learning methods generate multiple classifiers to form a committee by repeated application of a single base learning algorithm. The committee members vote to decide the final classification. Two such methods, Bagging and Boosting, have shown great success with decision tree learning. They create different classifiers by modifying the distribution of the training set. This paper studies a different approach: Stochastic Attribute Selection Committee learning of decision trees. It generates classifier committees by stochastically modifying the set of attributes but keeping the distribution of the training set unchanged. An empirical evaluation of a variant of this method, namely Sasc, in a representative collection of natural domains shows that the SASC method can significantly reduce the error rate of decision tree learning. On average Sasc is more accurate than Bagging and less accurate than Boosting, although a one-tailed significance test fails to show that these differences are significant at a level of 0.05. In addition, it is found that, like Bagging, Sasc is more stable than Boosting in terms of less frequently obtaining significantly higher error rates than C4.5 and, when error is raised, producing lower error rate increases. Moreover, like Bagging, Sasc is amenable to parallel and distributed processing while Boosting is not.
Zheng, Z.; Webb, G. I.
Multiple Boosting: A Combination of Boosting and Bagging
Proceedings of the 1998 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'98), pp. 1133-1140, CSREA Press, Las Vegas, Nevada, 1998.
@inproceedings{ZhengWebb98b,
title = {Multiple Boosting: A Combination of Boosting and Bagging},
author = {Z. Zheng and G. I. Webb},
year = {1998},
date = {1998-01-01},
booktitle = {Proceedings of the 1998 International Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA'98)},
pages = {1133-1140},
publisher = {CSREA Press},
address = {Las Vegas, Nevada},
abstract = {Classifier committee learning approaches have demonstrated great success in increasing the prediction accuracy of classifier learning, which is a key technique for datamining. These approaches generate several classifiers to form a committee by repeated application of a single base learning algorithm. The committee members vote to decide the final classification. It has been shown that Boosting and Bagging, as two representative methods of this type, can significantly decrease the error rate of decision tree learning. Boosting is generally more accurate than Bagging, but the former is more variable than the latter. In addition, bagging is amenable to parallel or distributed processing, while Boosting is not. In this paper, we study a new committee learning algorithm, namely MB (Multiple Boosting). It creates multiple subcommittees by combining Boosting and Bagging. Experimental results in a representative collection of natural domains show that MB is, on average, more accurate than either Bagging or Boosting alone. It is more stable than Boosting, and is amenable to parallel or distributed processing. These characteristics make MB a good choice for parallel datamining.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Classifier committee learning approaches have demonstrated great success in increasing the prediction accuracy of classifier learning, which is a key technique for datamining. These approaches generate several classifiers to form a committee by repeated application of a single base learning algorithm. The committee members vote to decide the final classification. It has been shown that Boosting and Bagging, as two representative methods of this type, can significantly decrease the error rate of decision tree learning. Boosting is generally more accurate than Bagging, but the former is more variable than the latter. In addition, bagging is amenable to parallel or distributed processing, while Boosting is not. In this paper, we study a new committee learning algorithm, namely MB (Multiple Boosting). It creates multiple subcommittees by combining Boosting and Bagging. Experimental results in a representative collection of natural domains show that MB is, on average, more accurate than either Bagging or Boosting alone. It is more stable than Boosting, and is amenable to parallel or distributed processing. These characteristics make MB a good choice for parallel datamining.
Zheng, Z.; Webb, G. I.; Ting, K. M.
Integrating Boosting and Stochastic Attribute Selection Committees for Further Improving The Performance of Decision Tree Learning
Proceedings of the Tenth IEEE International Conference on Tools with Artificial Intelligence (ICTAI-98), pp. 216-223, IEEE Computer Society Press, Taipei, Taiwan, 1998.
@inproceedings{ZhengWebbTing98,
title = {Integrating Boosting and Stochastic Attribute Selection Committees for Further Improving The Performance of Decision Tree Learning},
author = {Z. Zheng and G. I. Webb and K. M. Ting},
year = {1998},
date = {1998-01-01},
booktitle = {Proceedings of the Tenth IEEE International Conference on Tools with Artificial Intelligence (ICTAI-98)},
pages = {216-223},
publisher = {IEEE Computer Society Press},
address = {Taipei, Taiwan},
abstract = {Techniques for constructing classifier committees including boosting and bagging have demonstrated great success, especially boosting for decision tree learning. This type of technique generates several classifiers to form a committee by repeated application of a single base learning algorithm. The committee members vote to decide the final classification. Boosting and bagging create different classifiers by modifying the distribution of the training set. SASC (Stochastic Attribute Selection Committees) uses an alternative approach to generating classifier committees by stochastic manipulation of the set of attributes considered at each node during tree induction, but keeping the distribution of the training set unchanged. We propose a method for improving the performance of boosting. This technique combines boosting and SASC. It builds classifier committees by manipulating both the distribution of the training set and the set of attributes available during induction. In the synergy SASC effectively increases the model diversity of boosting. Experiments with a representative collection of natural domains show that, on average, the combined technique outperforms either boosting or SASC alone in terms of reducing the error rate of decision tree learning.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Techniques for constructing classifier committees including boosting and bagging have demonstrated great success, especially boosting for decision tree learning. This type of technique generates several classifiers to form a committee by repeated application of a single base learning algorithm. The committee members vote to decide the final classification. Boosting and bagging create different classifiers by modifying the distribution of the training set. SASC (Stochastic Attribute Selection Committees) uses an alternative approach to generating classifier committees by stochastic manipulation of the set of attributes considered at each node during tree induction, but keeping the distribution of the training set unchanged. We propose a method for improving the performance of boosting. This technique combines boosting and SASC. It builds classifier committees by manipulating both the distribution of the training set and the set of attributes available during induction. In the synergy SASC effectively increases the model diversity of boosting. Experiments with a representative collection of natural domains show that, on average, the combined technique outperforms either boosting or SASC alone in terms of reducing the error rate of decision tree learning.
Chiu, B. C.; Webb, G. I.
Using Decision Trees For Agent Modelling: Improving Prediction Performance
User Modeling and User-Adapted Interaction, vol. 8, no. 1-2, pp. 131-152, 1998.
@article{ChiuWebb98,
title = {Using Decision Trees For Agent Modelling: Improving Prediction Performance},
author = {B. C. Chiu and G. I. Webb},
year = {1998},
date = {1998-01-01},
journal = {User Modeling and User-Adapted Interaction},
volume = {8},
number = {1-2},
pages = {131-152},
publisher = {Springer},
address = {Netherlands},
abstract = {A modeling system may be required to predict an agent's future actions under constraints of inadequate or contradictory relevant historical evidence. This can result in low prediction accuracy, or otherwise, low prediction rates, leaving a set of cases for which no predictions are made. A previous study that explored techniques for improving prediction rates in the context of modeling students' subtraction skills using Feature Based Modeling showed a tradeoff between prediction rate and predication accuracy. This paper presents research that aims to improve prediction rates without affecting prediction accuracy. The FBM-C4.5 agent modeling system was used in this research. However, the techniques explored are applicable to any Feature Based Modeling system, and the most effective technique developed is applicable to most agent modeling systems. The default FBM-C4.5 system models agents' competencies with a set of decision trees, trained on all historical data. Each tree predicts one particular aspect of the agent's action. Predictions from multiple trees are compared for consensus. FBM-C4.5 makes no prediction when predictions from different trees contradict one another. This strategy trades off reduced prediction rates for increased accuracy. To make predictions in the absence of consensus, three techniques have been evaluated. They include using voting, using a tree quality measure and using a leaf quality measure. An alternative technique that merges multiple decision trees into a single tree provides an advantage of producing models that are more comprehensible. However, all of these techniques demonstrated the previous encountered trade-off between rate of prediction and accuracy of prediction, albeit less pronounced. It was hypothesized that models built on more current observations would outperform models built on earlier observations. Experimental results support this hypothesis. A Dual-model system, which takes this temporal factor into account, has been evaluated. This fifth approach achieved a significant improvement in prediction rate without significantly affecting prediction accuracy.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
A modeling system may be required to predict an agent's future actions under constraints of inadequate or contradictory relevant historical evidence. This can result in low prediction accuracy, or otherwise, low prediction rates, leaving a set of cases for which no predictions are made. A previous study that explored techniques for improving prediction rates in the context of modeling students' subtraction skills using Feature Based Modeling showed a tradeoff between prediction rate and predication accuracy. This paper presents research that aims to improve prediction rates without affecting prediction accuracy. The FBM-C4.5 agent modeling system was used in this research. However, the techniques explored are applicable to any Feature Based Modeling system, and the most effective technique developed is applicable to most agent modeling systems. The default FBM-C4.5 system models agents' competencies with a set of decision trees, trained on all historical data. Each tree predicts one particular aspect of the agent's action. Predictions from multiple trees are compared for consensus. FBM-C4.5 makes no prediction when predictions from different trees contradict one another. This strategy trades off reduced prediction rates for increased accuracy. To make predictions in the absence of consensus, three techniques have been evaluated. They include using voting, using a tree quality measure and using a leaf quality measure. An alternative technique that merges multiple decision trees into a single tree provides an advantage of producing models that are more comprehensible. However, all of these techniques demonstrated the previous encountered trade-off between rate of prediction and accuracy of prediction, albeit less pronounced. It was hypothesized that models built on more current observations would outperform models built on earlier observations. Experimental results support this hypothesis. A Dual-model system, which takes this temporal factor into account, has been evaluated. This fifth approach achieved a significant improvement in prediction rate without significantly affecting prediction accuracy.
Webb, G. I.
Decision Tree Grafting
Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI 97), pp. 846-851, Morgan Kaufmann, Nagoya, Japan, 1997.
@inproceedings{Webb97,
title = {Decision Tree Grafting},
author = {G. I. Webb},
year = {1997},
date = {1997-01-01},
booktitle = {Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI 97)},
pages = {846-851},
publisher = {Morgan Kaufmann},
address = {Nagoya, Japan},
abstract = {This paper extends recent work on decision tree grafting. Grafting is an inductive process that adds nodes to inferred decision trees. This process is demonstrated to frequently improve predictive accuracy. Superficial analysis might suggest that decision tree grafting is the direct reverse of pruning. To the contrary, it is argued that the two processes are complementary. This is because, like standard tree growing techniques, pruning uses only local information, whereas grafting uses non-local information. The use of both pruning and grafting in conjunction is demonstrated to provide the best general predictive accuracy over a representative selection of learning tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper extends recent work on decision tree grafting. Grafting is an inductive process that adds nodes to inferred decision trees. This process is demonstrated to frequently improve predictive accuracy. Superficial analysis might suggest that decision tree grafting is the direct reverse of pruning. To the contrary, it is argued that the two processes are complementary. This is because, like standard tree growing techniques, pruning uses only local information, whereas grafting uses non-local information. The use of both pruning and grafting in conjunction is demonstrated to provide the best general predictive accuracy over a representative selection of learning tasks.
Webb, G. I.; Chiu, B. C.; Kuzmycz, M.
Comparative Evaluation of Alternative Induction Engines for Feature Based Modelling
International Journal of Artificial Intelligence in Education, vol. 8, pp. 97-115, 1997.
@article{WebbChiuKuzmycz97,
title = {Comparative Evaluation of Alternative Induction Engines for Feature Based Modelling},
author = {G. I. Webb and B. C. Chiu and M. Kuzmycz},
doi = {10.1.1.36.3545},
year = {1997},
date = {1997-01-01},
journal = {International Journal of Artificial Intelligence in Education},
volume = {8},
pages = {97-115},
publisher = {IOS Press},
address = {NAmsterdam},
abstract = {Feature Based Modelling has demonstrated the ability to produce agent models with high accuracy in predicting an agent's future actions. There are a number of respects in which this modelling technique is novel. However, there has been no previous analysis of which aspects of the approach are responsible for its performance. One distinctive feature of the approach is a purpose built induction module. This paper presents a study in which the original custom built Feature Based Modelling induction module was replaced by the C4.5 machine learning system. Comparative evaluation shows that the use of C4.5 increases the number of predictions made without significantly altering the accuracy of those predictions. This suggests that it is the general input-output agent modelling methodology used with both systems that has primary responsibility for the high predictive accuracy previously reported for Feature Based Modelling, rather than its initial idiosyncratic induction technique.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Feature Based Modelling has demonstrated the ability to produce agent models with high accuracy in predicting an agent's future actions. There are a number of respects in which this modelling technique is novel. However, there has been no previous analysis of which aspects of the approach are responsible for its performance. One distinctive feature of the approach is a purpose built induction module. This paper presents a study in which the original custom built Feature Based Modelling induction module was replaced by the C4.5 machine learning system. Comparative evaluation shows that the use of C4.5 increases the number of predictions made without significantly altering the accuracy of those predictions. This suggests that it is the general input-output agent modelling methodology used with both systems that has primary responsibility for the high predictive accuracy previously reported for Feature Based Modelling, rather than its initial idiosyncratic induction technique.
Chiu, B. C.; Webb, G. I.; Kuzmycz, M.
A Comparison of First-Order and Zeroth-Order Induction for Input-Output Agent Modelling
Jameson, A.; Paris, C.; Tasso, C. (Ed.): Proceedings of the Sixth International Conference on User Modeling (UM'97), pp. 347-358, Springer, Chia Laguna, Sardinia, 1997.
@inproceedings{ChiuWebbKuzmycz97,
title = {A Comparison of First-Order and Zeroth-Order Induction for Input-Output Agent Modelling},
author = {B. C. Chiu and G. I. Webb and M. Kuzmycz},
editor = {A. Jameson and C. Paris and C. Tasso},
doi = {10.1007/978-3-7091-2670-7_35},
year = {1997},
date = {1997-01-01},
booktitle = {Proceedings of the Sixth International Conference on User Modeling (UM'97)},
pages = {347-358},
publisher = {Springer},
address = {Chia Laguna, Sardinia},
abstract = {Most student modelling systems seek to develop a model of the internal operation of the cognitive system. In contrast, Input-Output Agent Modelling (IOAM) models an agent in terms of relationships between the inputs and outputs of the cognitive system. Previous IOAM systems have demonstrated high predictive accuracy in the domain of elementary subtraction. These systems use zeroth-order induction. Many of the predicates used, however, represent relations. This suggests that first-order induction might perform well in this domain. This paper reports a study in which zeroth-order and first-order induction engines were used to build models of student subtraction skills. Comparative evaluation shows that zeroth-order induction performs better than first-order in detecting regularities indicating misconceptions while first-order induction leads zeroth-order in detecting regularities indicating correct concepts and inducing a more comprehensible student model. This suggests there exists a trade-off between these factors and that there is still scope for improvement.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Most student modelling systems seek to develop a model of the internal operation of the cognitive system. In contrast, Input-Output Agent Modelling (IOAM) models an agent in terms of relationships between the inputs and outputs of the cognitive system. Previous IOAM systems have demonstrated high predictive accuracy in the domain of elementary subtraction. These systems use zeroth-order induction. Many of the predicates used, however, represent relations. This suggests that first-order induction might perform well in this domain. This paper reports a study in which zeroth-order and first-order induction engines were used to build models of student subtraction skills. Comparative evaluation shows that zeroth-order induction performs better than first-order in detecting regularities indicating misconceptions while first-order induction leads zeroth-order in detecting regularities indicating correct concepts and inducing a more comprehensible student model. This suggests there exists a trade-off between these factors and that there is still scope for improvement.
Chiu, B. C.; Webb, G. I.
Using C4.5 as an Induction Engine for Agent Modeling: An Experiment of Optimisation
Proceedings (on-line) of The First Machine Learning for User Modeling Workshop (UM'97), Chia Laguna, Sardinia, 1997.
@inproceedings{ChiuWebb97,
title = {Using C4.5 as an Induction Engine for Agent Modeling: An Experiment of Optimisation},
author = {B. C. Chiu and G. I. Webb},
doi = {10.1023/A:1008296930163},
year = {1997},
date = {1997-01-01},
booktitle = {Proceedings (on-line) of The First Machine Learning for User Modeling Workshop (UM'97)},
address = {Chia Laguna, Sardinia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Chiu, B. C.; Webb, G. I.; Zheng, Z.
Using Decision Trees for Agent Modelling: A Study on Resolving Conflicting Predictions
Sattar, A. (Ed.): Lecture Notes in Computer Science Vol. 1342: Proceedings of the Tenth Australian Joint Conference on Artificial Intelligence (AI'97), pp. 349-358, Springer-Verlag, Perth, Australia, 1997.
@inproceedings{ChiuWebbZheng97,
title = {Using Decision Trees for Agent Modelling: A Study on Resolving Conflicting Predictions},
author = {B. C. Chiu and G. I. Webb and Z. Zheng},
editor = {A. Sattar},
year = {1997},
date = {1997-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 1342: Proceedings of the Tenth Australian Joint Conference on Artificial Intelligence (AI'97)},
pages = {349-358},
publisher = {Springer-Verlag},
address = {Perth, Australia},
abstract = {Input-Output Agent Modelling (IOAM) is an approach to modelling an agent in terms of relationships between the inputs and outputs of the cognitive system. This approach, together with a leading inductive learning algorithm, C4.5, has been adopted to build a subtraction skill modeller, C4.5-IOAM. It models agents' competencies with a set of decision trees. C4.5-IOAM makes no prediction when predictions from different decision trees are contradictory. This paper proposes three techniques for resolving such situations. Two techniques involve selecting the more reliable prediction from a set of competing predictions using a free quality measure and a leaf quality measure. The other technique merges multiple decision trees into a single tree. This has the additional advantage of producing more comprehensible models. Experimental results, in the domain of modelling elementary subtraction skills, showed that the tree quality and the leaf quality of a decision path provided valuable references for resolving contradicting predictions and a single tree model representation performed nearly equally well to the multi-tree model representation.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Input-Output Agent Modelling (IOAM) is an approach to modelling an agent in terms of relationships between the inputs and outputs of the cognitive system. This approach, together with a leading inductive learning algorithm, C4.5, has been adopted to build a subtraction skill modeller, C4.5-IOAM. It models agents' competencies with a set of decision trees. C4.5-IOAM makes no prediction when predictions from different decision trees are contradictory. This paper proposes three techniques for resolving such situations. Two techniques involve selecting the more reliable prediction from a set of competing predictions using a free quality measure and a leaf quality measure. The other technique merges multiple decision trees into a single tree. This has the additional advantage of producing more comprehensible models. Experimental results, in the domain of modelling elementary subtraction skills, showed that the tree quality and the leaf quality of a decision path provided valuable references for resolving contradicting predictions and a single tree model representation performed nearly equally well to the multi-tree model representation.
Webb, G. I.; Kuzmycz, M.
Feature Based Modelling: A Methodology for Producing Coherent, Consistent, Dynamically Changing Models of Agents Competencies
User Modeling and User-Adapted Interaction, vol. 5, no. 2, pp. 117-150, 1996.
@article{WebbKuzmycz96,
title = {Feature Based Modelling: A Methodology for Producing Coherent, Consistent, Dynamically Changing Models of Agents Competencies},
author = {G. I. Webb and M. Kuzmycz},
year = {1996},
date = {1996-01-01},
journal = {User Modeling and User-Adapted Interaction},
volume = {5},
number = {2},
pages = {117-150},
publisher = {Springer},
address = {Netherlands},
abstract = {Feature Based Modelling uses attribute value machine learning techniques to model an agent's competency. This is achieved by creating a model describing the relationships between the features of the agent's actions and of the contexts in which those actions are performed. This paper describes techniques that have been developed for creating these models and for extracting key information therefrom. An overview is provided of previous studies that have evaluated the application of Feature Based Modelling in a number of educational contexts including piano keyboard playing, the unification of Prolog terms and elementary subtraction. These studies have demonstrated that the approach is applicable to a wide spectrum of domains. Classroom use has demonstrated the low computational overheads of the technique. A new study of the application of the approach to modelling elementary subtraction skills is presented. The approach demonstrates accuracy in excess of 90% when predicting student solutions. It also demonstrates the ability to identify and model student's buggy arithmetic procedures.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Feature Based Modelling uses attribute value machine learning techniques to model an agent's competency. This is achieved by creating a model describing the relationships between the features of the agent's actions and of the contexts in which those actions are performed. This paper describes techniques that have been developed for creating these models and for extracting key information therefrom. An overview is provided of previous studies that have evaluated the application of Feature Based Modelling in a number of educational contexts including piano keyboard playing, the unification of Prolog terms and elementary subtraction. These studies have demonstrated that the approach is applicable to a wide spectrum of domains. Classroom use has demonstrated the low computational overheads of the technique. A new study of the application of the approach to modelling elementary subtraction skills is presented. The approach demonstrates accuracy in excess of 90% when predicting student solutions. It also demonstrates the ability to identify and model student's buggy arithmetic procedures.
Webb, G. I.; Wells, J.
An Experimental Evaluation of Integrating Machine Learning with Knowledge Acquisition Through Direct Interaction with Domain Experts
Compton, P.; Mizoguchi, R.; Motada, H.; Menzies, T. (Ed.): Proceedings of the 1996 Pacific Knowledge Acquisition Workshop (PKAW'96), pp. 170-189, UNSW Press, Coogee, Sydney, Australia, 1996.
@inproceedings{WebbWells96,
title = {An Experimental Evaluation of Integrating Machine Learning with Knowledge Acquisition Through Direct Interaction with Domain Experts},
author = {G. I. Webb and J. Wells},
editor = {P. Compton and R. Mizoguchi and H. Motada and T. Menzies},
year = {1996},
date = {1996-01-01},
booktitle = {Proceedings of the 1996 Pacific Knowledge Acquisition Workshop (PKAW'96)},
pages = {170-189},
publisher = {UNSW Press},
address = {Coogee, Sydney, Australia},
abstract = {Machine learning and knowledge acquisition from experts have distinct and apparently complementary knowledge acquisition capabilities. This study demonstrates that the integration of these approaches can both improve the accuracy of the knowledge base that is developed and reduce the time taken to develop it. The system studied, called The Knowledge Factory is distinguished by the manner in which it supports direct interaction with domain experts with little or no knowledge engineering expertise. The benefits reported relate to use by such users. In addition to the improved quality of the knowledge base, in questionnaire responses the users provided favourable evaluations of the integration of machine learning with knowledge acquisition within the system.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Machine learning and knowledge acquisition from experts have distinct and apparently complementary knowledge acquisition capabilities. This study demonstrates that the integration of these approaches can both improve the accuracy of the knowledge base that is developed and reduce the time taken to develop it. The system studied, called The Knowledge Factory is distinguished by the manner in which it supports direct interaction with domain experts with little or no knowledge engineering expertise. The benefits reported relate to use by such users. In addition to the improved quality of the knowledge base, in questionnaire responses the users provided favourable evaluations of the integration of machine learning with knowledge acquisition within the system.
Webb, G. I.
Cost Sensitive Specialisation
Foo, N. Y.; Goebel, R. (Ed.): Lecture Notes in Computer Science Vol. 1114. Topics in Artificial Intelligence: Proceedings of the Fourth Pacific Rim International Conference on Artificial Intelligence (PRICAI'96), pp. 23-34, Springer-Verlag, Cairns, Australia, 1996.
@inproceedings{Webb96d,
title = {Cost Sensitive Specialisation},
author = {G. I. Webb},
editor = {N. Y. Foo and R. Goebel},
year = {1996},
date = {1996-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 1114. Topics in Artificial Intelligence: Proceedings of the Fourth Pacific Rim International Conference on Artificial Intelligence (PRICAI'96)},
pages = {23-34},
publisher = {Springer-Verlag},
address = {Cairns, Australia},
abstract = {Cost-sensitive specialization is a generic technique for misclassification cost sensitive induction. This technique involves specializing aspects of a classifier associated with high misclassification costs and generalizing those associated with low misclassification costs. It is widely applicable and simple to implement. It could be used to augment the effect of standard cost-sensitive induction techniques. It should directly extend to test application cost sensitive induction tasks. Experimental evaluation demonstrates consistent positive effects over a range of misclassification cost sensitive learning tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Cost-sensitive specialization is a generic technique for misclassification cost sensitive induction. This technique involves specializing aspects of a classifier associated with high misclassification costs and generalizing those associated with low misclassification costs. It is widely applicable and simple to implement. It could be used to augment the effect of standard cost-sensitive induction techniques. It should directly extend to test application cost sensitive induction tasks. Experimental evaluation demonstrates consistent positive effects over a range of misclassification cost sensitive learning tasks.
Webb, G. I.
Inclusive Pruning: A New Class of Pruning Rule for Unordered Search and its Application to Classification Learning
Ramamohanarao, K. (Ed.): Australian Computer Science Communications Vol. 18 (1): Proceedings of the Nineteenth Australasian Computer Science Conference (ACSC'96), pp. 1-10, ACS, Royal Melbourne Insitute of Technology, Australia, 1996.
@inproceedings{Webb96e,
title = {Inclusive Pruning: A New Class of Pruning Rule for Unordered Search and its Application to Classification Learning},
author = {G. I. Webb},
editor = {K. Ramamohanarao},
year = {1996},
date = {1996-01-01},
booktitle = {Australian Computer Science Communications Vol. 18 (1): Proceedings of the Nineteenth Australasian Computer Science Conference (ACSC'96)},
pages = {1-10},
publisher = {ACS},
address = {Royal Melbourne Insitute of Technology, Australia},
abstract = {This paper presents a new class of pruning rule for unordered search. Previous pruning rules for unordered search identify operators that should not be applied in order to prune nodes reached via those operators. In contrast, the new pruning rules identify operators that should be applied and prune nodes that are not reached via those operators. Specific pruning rules employing both these approaches are identified for classification learning. Experimental results demonstrate that application of the new pruning rules can reduce by more than 60% the number of states from the search space that are considered during classification learning.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper presents a new class of pruning rule for unordered search. Previous pruning rules for unordered search identify operators that should not be applied in order to prune nodes reached via those operators. In contrast, the new pruning rules identify operators that should be applied and prune nodes that are not reached via those operators. Specific pruning rules employing both these approaches are identified for classification learning. Experimental results demonstrate that application of the new pruning rules can reduce by more than 60% the number of states from the search space that are considered during classification learning.
Webb, G. I.
Knowledge-Based Systems, vol. 9, pp. 253-266, 1996.
@article{Webb96a,
title = {Integrating Machine Learning With Knowledge Acquisition Through Direct Interaction With Domain Experts},
author = {G. I. Webb},
doi = {10.1016/0950-7051(96)01033-7},
year = {1996},
date = {1996-01-01},
journal = {Knowledge-Based Systems},
volume = {9},
pages = {253-266},
publisher = {Elsevier},
abstract = {Knowledge elicitation from experts and empirical machine learning are two distinct approaches to knowledge acquisition with differing and mutually complementary capabilities. Learning apprentices have provided environments in which a knowledge engineer may collaborate with a machine learning system allowing, for a synergy between the complementary approaches. The Knowledge Factory is a knowledge acquisition environment that allows a domain expert to collaborate directly with a machine learning system without the need for assistance from a knowledge engineer. This requires a different form of environment to the learning apprentice. This paper describes techniques for supporting such interactions and their implementation in a knowledge acquisition environment called The Knowledge Factory.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Knowledge elicitation from experts and empirical machine learning are two distinct approaches to knowledge acquisition with differing and mutually complementary capabilities. Learning apprentices have provided environments in which a knowledge engineer may collaborate with a machine learning system allowing, for a synergy between the complementary approaches. The Knowledge Factory is a knowledge acquisition environment that allows a domain expert to collaborate directly with a machine learning system without the need for assistance from a knowledge engineer. This requires a different form of environment to the learning apprentice. This paper describes techniques for supporting such interactions and their implementation in a knowledge acquisition environment called The Knowledge Factory.
Webb, G. I.
Further Experimental Evidence Against The Utility Of Occam's Razor
Journal of Artificial Intelligence Research, vol. 4, pp. 397-417, 1996.
@article{Webb96b,
title = {Further Experimental Evidence Against The Utility Of Occam's Razor},
author = {G. I. Webb},
doi = {10.1613/jair.228},
year = {1996},
date = {1996-01-01},
journal = {Journal of Artificial Intelligence Research},
volume = {4},
pages = {397-417},
publisher = {AAAI Press},
address = {Menlo Park, CA},
abstract = {This paper presents new experimental evidence against the utility of Occam's razor. A systematic procedure is presented for post-processing decision trees produced by C4.5. This procedure was derived by rejecting Occam's razor and instead attending to the assumption that similar objects are likely to belong to the same class. It increases a decision tree's complexity without altering the performance of that tree on the training data from which it is inferred. The resulting more complex decision trees are demonstrated to have, on average, for a variety of common learning tasks, higher predictive accuracy than the less complex original decision trees. This result raises considerable doubt about the utility of Occam's razor as it is commonly applied in modern machine learning.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper presents new experimental evidence against the utility of Occam's razor. A systematic procedure is presented for post-processing decision trees produced by C4.5. This procedure was derived by rejecting Occam's razor and instead attending to the assumption that similar objects are likely to belong to the same class. It increases a decision tree's complexity without altering the performance of that tree on the training data from which it is inferred. The resulting more complex decision trees are demonstrated to have, on average, for a variety of common learning tasks, higher predictive accuracy than the less complex original decision trees. This result raises considerable doubt about the utility of Occam's razor as it is commonly applied in modern machine learning.
Webb, G. I.
A Heuristic Covering Algorithm Outperforms Learning All Rules
Proceedings of Information, Statistics and Induction in Science (ISIS '96), pp. 20-30, World Scientific, Melbourne, Australia, 1996.
@inproceedings{Webb96c,
title = {A Heuristic Covering Algorithm Outperforms Learning All Rules},
author = {G. I. Webb},
year = {1996},
date = {1996-01-01},
booktitle = {Proceedings of Information, Statistics and Induction in Science (ISIS '96)},
pages = {20-30},
publisher = {World Scientific},
address = {Melbourne, Australia},
abstract = {The induction of classification rules has been dominated by a single generic technique-the covering algorithm. This approach employs a simple hill-climbing search to learn sets of rules. Such search is subject to numerous widely known deficiencies. Further, there is a growing body of evidence that learning redundant sets of rules can improve predictive accuracy. The ultimate end-point of a move toward learning redundant rule sets would appear to be to learn and employ all possible rules. This paper presents a learning system that does this. An empirical investigation shows that, while the approach often achieves higher predictive accuracy than a covering algorithm, the covering algorithm outperforms induction of all rules significantly more frequently. Preliminary analysis suggests that learning all rules performs well when the training set clearly defines the decision surfaces but that the heuristic covering algorithm performs better when the decision surfaces are not clearly delineated by the training examples.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The induction of classification rules has been dominated by a single generic technique-the covering algorithm. This approach employs a simple hill-climbing search to learn sets of rules. Such search is subject to numerous widely known deficiencies. Further, there is a growing body of evidence that learning redundant sets of rules can improve predictive accuracy. The ultimate end-point of a move toward learning redundant rule sets would appear to be to learn and employ all possible rules. This paper presents a learning system that does this. An empirical investigation shows that, while the approach often achieves higher predictive accuracy than a covering algorithm, the covering algorithm outperforms induction of all rules significantly more frequently. Preliminary analysis suggests that learning all rules performs well when the training set clearly defines the decision surfaces but that the heuristic covering algorithm performs better when the decision surfaces are not clearly delineated by the training examples.
Newlands, D.; Webb, G. I.
Polygonal Inductive Generalisation System
Forsyth, G.; Ali, M. (Ed.): Proceedings of the Eighth International Conference on Industrial and Engineering Applications of Artificial Intelligence and Expert Systems (IEA/AIE '95), pp. 587-592, Gordon and Breach Science Publishers, Inc, Melbourne, Australia, 1995.
@inproceedings{NewlandsWebb95,
title = {Polygonal Inductive Generalisation System},
author = {D. Newlands and G. I. Webb},
editor = {G. Forsyth and M. Ali},
year = {1995},
date = {1995-01-01},
booktitle = {Proceedings of the Eighth International Conference on Industrial and Engineering Applications of Artificial Intelligence and Expert Systems (IEA/AIE '95)},
pages = {587-592},
publisher = {Gordon and Breach Science Publishers, Inc},
address = {Melbourne, Australia},
abstract = {Knowledge acquisition remains one of the primary constraints on the development of expert systems. A number of researchers have explored methods for allowing a machine learning system to assist a knowledge engineer in knowledge acquisition. In contrast, we are exploring methods for enabling an expert to directly interact with a machine learning system to collaborate during knowledge acquisition. We report recent extensions to our methodology encompassing a revised model of the role of machine learning in knowledge acquisition; techniques for communication between a machine learning system and a domain expert and novel forms of assistance that a machine learning system may provide to an expert.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Knowledge acquisition remains one of the primary constraints on the development of expert systems. A number of researchers have explored methods for allowing a machine learning system to assist a knowledge engineer in knowledge acquisition. In contrast, we are exploring methods for enabling an expert to directly interact with a machine learning system to collaborate during knowledge acquisition. We report recent extensions to our methodology encompassing a revised model of the role of machine learning in knowledge acquisition; techniques for communication between a machine learning system and a domain expert and novel forms of assistance that a machine learning system may provide to an expert.
Smith, P. A.; Webb, G. I.
Transparency Debugging with Explanations for Novice Programmers
Ducassa, M. (Ed.): Proceedings of the Second International Workshop on Automated and Algorithmic Debugging (AADEBUG'95), IRISA-CNRS, Saint-Malo, France, 1995.
@inproceedings{SmithWebb95a,
title = {Transparency Debugging with Explanations for Novice Programmers},
author = {P. A. Smith and G. I. Webb},
editor = {M. Ducassa},
year = {1995},
date = {1995-01-01},
booktitle = {Proceedings of the Second International Workshop on Automated and Algorithmic Debugging (AADEBUG'95)},
publisher = {IRISA-CNRS},
address = {Saint-Malo, France},
abstract = {Novice programmers often find programming to be a difficult and frustrating task. Because of their lack of experience in programming novices have different needs to experts when it comes to debugging assistants. One way a debugging assistant could be tailored to novices, as proposed by Eisenstadt, is to provide them with an explicit model of how their program works and, hence encourage them to find errors for themselves. We discuss such a transparency debugger, Bradman, that we have been developing to assist novice programmers understand and debug their C programs. We also present the results of an experiment, conducted on volunteer novice programmers, in which approximately half of the students had access to an explanation of each statement as it was executed and the other half did not. We show that access to such explanations provided beneficial results for a significant number of students.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Novice programmers often find programming to be a difficult and frustrating task. Because of their lack of experience in programming novices have different needs to experts when it comes to debugging assistants. One way a debugging assistant could be tailored to novices, as proposed by Eisenstadt, is to provide them with an explicit model of how their program works and, hence encourage them to find errors for themselves. We discuss such a transparency debugger, Bradman, that we have been developing to assist novice programmers understand and debug their C programs. We also present the results of an experiment, conducted on volunteer novice programmers, in which approximately half of the students had access to an explanation of each statement as it was executed and the other half did not. We show that access to such explanations provided beneficial results for a significant number of students.
Smith, P. A.; Webb, G. I.
Reinforcing a Generic Computer Model for Novice Programmers
Proceedings of the Seventh Australian Society for Computers in Learning in Tertiary Education Conference (ASCILITE '95), ASCILITE, Melbourne, Australia, 1995.
@inproceedings{SmithWebb95b,
title = {Reinforcing a Generic Computer Model for Novice Programmers},
author = {P. A. Smith and G. I. Webb},
year = {1995},
date = {1995-01-01},
booktitle = {Proceedings of the Seventh Australian Society for Computers in Learning in Tertiary Education Conference (ASCILITE '95)},
publisher = {ASCILITE},
address = {Melbourne, Australia},
abstract = {Novices often find learning their first programming language to be a frustrating and difficult process. They have difficulties in developing and debugging their programs. One of their problems is that their mental model of how the computer works is inadequate. In this paper we discuss a programming assistant, called Bradman, which we are currently developing. It is aimed at novice programmers and designed to reinforce a concrete mental model of how the computer works as a program is executed. It shows explicitly how program states change as statements in the procedural language C are executed. It does this by means of graphical display together with contextualised verbal explanations of each statement.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Novices often find learning their first programming language to be a frustrating and difficult process. They have difficulties in developing and debugging their programs. One of their problems is that their mental model of how the computer works is inadequate. In this paper we discuss a programming assistant, called Bradman, which we are currently developing. It is aimed at novice programmers and designed to reinforce a concrete mental model of how the computer works as a program is executed. It shows explicitly how program states change as statements in the procedural language C are executed. It does this by means of graphical display together with contextualised verbal explanations of each statement.
Webb, G. I.
OPUS: An Efficient Admissible Algorithm For Unordered Search
Journal of Artificial Intelligence Research, vol. 3, pp. 431-465, 1995.
@article{Webb95,
title = {OPUS: An Efficient Admissible Algorithm For Unordered Search},
author = {G. I. Webb},
doi = {10.1613/jair.227},
year = {1995},
date = {1995-01-01},
journal = {Journal of Artificial Intelligence Research},
volume = {3},
pages = {431-465},
publisher = {AAAI Press},
address = {Menlo Park, CA},
abstract = {OPUS is a branch and bound search algorithm that enables efficient admissible search through spaces for which the order of search operator application is not significant. The algorithm's search efficiency is demonstrated with respect to very large machine learning search spaces. The use of admissible search is of potential value to the machine learning community as it means that the exact learning biases to be employed for complex learning tasks can be precisely specified and manipulated. OPUS also has potential for application in other areas of artificial intelligence, notably, truth maintenance.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
OPUS is a branch and bound search algorithm that enables efficient admissible search through spaces for which the order of search operator application is not significant. The algorithm's search efficiency is demonstrated with respect to very large machine learning search spaces. The use of admissible search is of potential value to the machine learning community as it means that the exact learning biases to be employed for complex learning tasks can be precisely specified and manipulated. OPUS also has potential for application in other areas of artificial intelligence, notably, truth maintenance.
Webb, G. I.; Wells, J.
Recent Progress in Machine-Expert Collaboration for Knowledge Acquisition
Yao, X. (Ed.): Proceedings of the Eighth Australian Joint Conference on Artificial Intelligence (AI'95), pp. 291-298, World Scientific, Canberra, Australia, 1995.
@inproceedings{WebbWells95,
title = {Recent Progress in Machine-Expert Collaboration for Knowledge Acquisition},
author = {G. I. Webb and J. Wells},
editor = {X. Yao},
year = {1995},
date = {1995-01-01},
booktitle = {Proceedings of the Eighth Australian Joint Conference on Artificial Intelligence (AI'95)},
pages = {291-298},
publisher = {World Scientific},
address = {Canberra, Australia},
abstract = {Knowledge acquisition remains one of the primary constraints on the development of expert systems. A number of researchers have explored methods for allowing a machine learning system to assist a knowledge engineer in knowledge acquisition. In contrast, we are exploring methods for enabling an expert to directly interact with a machine learning system to collaborate during knowledge acquisition. We report recent extensions to our methodology encompassing a revised model of the role of machine learning in knowledge acquisition; techniques for communication between a machine learning system and a domain expert and novel forms of assistance that a machine learning system may provide to an expert.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Knowledge acquisition remains one of the primary constraints on the development of expert systems. A number of researchers have explored methods for allowing a machine learning system to assist a knowledge engineer in knowledge acquisition. In contrast, we are exploring methods for enabling an expert to directly interact with a machine learning system to collaborate during knowledge acquisition. We report recent extensions to our methodology encompassing a revised model of the role of machine learning in knowledge acquisition; techniques for communication between a machine learning system and a domain expert and novel forms of assistance that a machine learning system may provide to an expert.
Webb, G. I.
Generality Is More Significant Then Complexity: Toward An Alternative To Occams Razor
Zhang, C.; Debenham, J.; Lukose, D. (Ed.): Artificial Intelligence: Sowing the Seeds for the Future, Proceedings of Seventh Australian Joint Conference on Artificial Intelligence (AI'94), pp. 60-67, World Scientific, Armidale,NSW, Australia, 1994.
@inproceedings{Webb94b,
title = {Generality Is More Significant Then Complexity: Toward An Alternative To Occams Razor},
author = {G. I. Webb},
editor = {C. Zhang and J. Debenham and D. Lukose},
year = {1994},
date = {1994-01-01},
booktitle = {Artificial Intelligence: Sowing the Seeds for the Future, Proceedings of Seventh Australian Joint Conference on Artificial Intelligence (AI'94)},
pages = {60-67},
publisher = {World Scientific},
address = {Armidale,NSW, Australia},
abstract = {Occam's Razor is widely employed in machine learning to select between classifiers with equal empirical support. This paper presents the theorem of decreasing inductive power: that, all other things being equal, if two classifiers a and b cover identical cases from the training set and a is a generalisation of b, a has higher probability than b of misclassifying a previously unsighted case. This theorem suggests that, to the contrary of Occam's Razor, generality, not complexity, should be used to select between classifiers with equal empirical support. Two studies are presented. The first study demonstrates that the theorem of decreasing inductive power holds for a number of commonly studied learning problems and for a number of different means of manipulating classifier generality. The second study demonstrates that generality provides a more consistent indicator of predictive accuracy in the context of a default rule than does complexity. These results suggest that the theorem of decreasing predictive power provides a suitable theoretical framework for the development of learning biases for use in selecting between classifiers with identical empirical support},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Occam's Razor is widely employed in machine learning to select between classifiers with equal empirical support. This paper presents the theorem of decreasing inductive power: that, all other things being equal, if two classifiers a and b cover identical cases from the training set and a is a generalisation of b, a has higher probability than b of misclassifying a previously unsighted case. This theorem suggests that, to the contrary of Occam's Razor, generality, not complexity, should be used to select between classifiers with equal empirical support. Two studies are presented. The first study demonstrates that the theorem of decreasing inductive power holds for a number of commonly studied learning problems and for a number of different means of manipulating classifier generality. The second study demonstrates that generality provides a more consistent indicator of predictive accuracy in the context of a default rule than does complexity. These results suggest that the theorem of decreasing predictive power provides a suitable theoretical framework for the development of learning biases for use in selecting between classifiers with identical empirical support
Webb, G. I.
Recent Progress in Learning Decision Lists by Prepending Inferred Rules
Proceedings of the Second Singapore International Conference on Intelligent Systems (SPICIS-94), pp. 280-285, Asia Computer Weekly, Singapore, 1994.
@inproceedings{Webb94a,
title = {Recent Progress in Learning Decision Lists by Prepending Inferred Rules},
author = {G. I. Webb},
year = {1994},
date = {1994-01-01},
booktitle = {Proceedings of the Second Singapore International Conference on Intelligent Systems (SPICIS-94)},
volume = {B},
pages = {280-285},
publisher = {Asia Computer Weekly},
address = {Singapore},
abstract = {This paper describes a new algorithm for learning decision lists that operates by prepending successive rules to the front of the list under construction. By contrast, the classic algorithm operates by appending successive rules to the end of the decision list under construction. The new algorithm is demonstrated in the majority of cases to produce smaller classifiers that provide improved predictive accuracy in less time than the classic algorithm.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes a new algorithm for learning decision lists that operates by prepending successive rules to the front of the list under construction. By contrast, the classic algorithm operates by appending successive rules to the end of the decision list under construction. The new algorithm is demonstrated in the majority of cases to produce smaller classifiers that provide improved predictive accuracy in less time than the classic algorithm.
Yip, S.; Webb, G. I.
Incorporating Canonical Discriminate Attributes in Classification Learning
Elio, R. (Ed.): Proceedings of the Tenth Biennial Canadian Artificial Intelligence Conference(AI-94), pp. 63-70, Morgan Kaufmann, Banff, Canada, 1994.
@inproceedings{YipWebb94b,
title = {Incorporating Canonical Discriminate Attributes in Classification Learning},
author = {S. Yip and G. I. Webb},
editor = {R. Elio},
year = {1994},
date = {1994-01-01},
booktitle = {Proceedings of the Tenth Biennial Canadian Artificial Intelligence Conference(AI-94)},
pages = {63-70},
publisher = {Morgan Kaufmann},
address = {Banff, Canada},
abstract = {This paper describes a method for incorporating canonical discriminant attributes in classification machine learning. Though decision trees and rules have semantic appeal when building expert systems, the merits of discriminant analysis are well documented. For data sets on which discriminant analysis obtains significantly better predictive accuracy than symbolic machine learning, the incorporation of canonical discriminant attributes can benefit machine learning. The process starts by applying canonical discriminant analysis to the training set. The canonical discriminant attributes are included as additional attributes. The expanded data set is then subjected to machine learning. This enables linear combinations of numeric attributes to be incorporated in the classifiers that are learnt. Evaluation on the data sets on which discriminant analysis performs better than most machine learning systems, such as the Iris flowers and Waveform data sets, shows that incorporating the power of discriminant analysis in machine classification learning can significantly improve the predictive accuracy and reduce the complexity of classifiers induced by machine learning systems.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes a method for incorporating canonical discriminant attributes in classification machine learning. Though decision trees and rules have semantic appeal when building expert systems, the merits of discriminant analysis are well documented. For data sets on which discriminant analysis obtains significantly better predictive accuracy than symbolic machine learning, the incorporation of canonical discriminant attributes can benefit machine learning. The process starts by applying canonical discriminant analysis to the training set. The canonical discriminant attributes are included as additional attributes. The expanded data set is then subjected to machine learning. This enables linear combinations of numeric attributes to be incorporated in the classifiers that are learnt. Evaluation on the data sets on which discriminant analysis performs better than most machine learning systems, such as the Iris flowers and Waveform data sets, shows that incorporating the power of discriminant analysis in machine classification learning can significantly improve the predictive accuracy and reduce the complexity of classifiers induced by machine learning systems.
Yip, S.; Webb, G. I.
Empirical Function Attribute Construction in Classification Learning
Zhang, C.; Debenham, J.; Lukose, D. (Ed.): Artificial Intelligence: Sowing the Seeds for the Future, Proceedings of Seventh Australian Joint Conference on Artificial Intelligence (AI'94), pp. 29-36, World Scientific, Armidale,NSW, Australia, 1994.
@inproceedings{YipWebb94a,
title = {Empirical Function Attribute Construction in Classification Learning},
author = {S. Yip and G. I. Webb},
editor = {C. Zhang and J. Debenham and D. Lukose},
year = {1994},
date = {1994-01-01},
booktitle = {Artificial Intelligence: Sowing the Seeds for the Future, Proceedings of Seventh Australian Joint Conference on Artificial Intelligence (AI'94)},
pages = {29-36},
publisher = {World Scientific},
address = {Armidale,NSW, Australia},
abstract = {The merits of incorporating feature construction to assist selective induction in learning hard concepts are well documented. This paper introduces the notion of function attributes and reports a method of incorporating functional regularities in classifiers. Training sets are preprocessed with this method before submission to a selective induction classification learning system. The method, referred to as FAFA (function attribute finding), is characterised by finding bivariate functions that contribute to the discrimination between classes and then transforming them to function attributes as additional attributes of the data set. The value of each function attribute equals the deviation of each example from the value obtained by applying that function to the example. The expanded data set is then submitted to classification learning. Evaluation with published and artificial data shows that this method can improve classifiers in terms of predictive accuracy and complexity.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The merits of incorporating feature construction to assist selective induction in learning hard concepts are well documented. This paper introduces the notion of function attributes and reports a method of incorporating functional regularities in classifiers. Training sets are preprocessed with this method before submission to a selective induction classification learning system. The method, referred to as FAFA (function attribute finding), is characterised by finding bivariate functions that contribute to the discrimination between classes and then transforming them to function attributes as additional attributes of the data set. The value of each function attribute equals the deviation of each example from the value obtained by applying that function to the example. The expanded data set is then submitted to classification learning. Evaluation with published and artificial data shows that this method can improve classifiers in terms of predictive accuracy and complexity.
Webb, G. I.
DLGref2: Techniques for Inductive Rule Refinement
Proceedings of the 1993 IJCAI Workshop W16: Machine Learning and Knowledge Acquisition, pp. 236-252, Chambery, France, 1993.
@inproceedings{Webb93d,
title = {DLGref2: Techniques for Inductive Rule Refinement},
author = {G. I. Webb},
year = {1993},
date = {1993-01-01},
booktitle = {Proceedings of the 1993 IJCAI Workshop W16: Machine Learning and Knowledge Acquisition},
pages = {236-252},
address = {Chambery, France},
abstract = {This paper describes and evaluates machine learning techniques for knowledge-base refinement. These techniques are central to Einstein, a knowledge acquisition system that enables a human expert to collaborate with a machine learning system at all stages of the knowledge-acquisition cycle. Experimental evaluation demonstrates that the knowledge-base refinement techniques are able to significantly increase the accuracy of nontrivial expert systems in a wide variety of domains.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes and evaluates machine learning techniques for knowledge-base refinement. These techniques are central to Einstein, a knowledge acquisition system that enables a human expert to collaborate with a machine learning system at all stages of the knowledge-acquisition cycle. Experimental evaluation demonstrates that the knowledge-base refinement techniques are able to significantly increase the accuracy of nontrivial expert systems in a wide variety of domains.
Webb, G. I.
Control, Capabilities and Communication: Three Key Issues for Machine-Expert Collaborative Knowledge Acquisition
Aussenac, N.; Boy, G.; Gaines, B.; Linster, M.; Ganascia, J. G.; Kodratoff, Y. (Ed.): Proceedings (Complement) of the Seventh European Workshop on Knowledge Acquisition for Knowledge-based Systems (EWKA'93), pp. 263-275, Springer-Verlag, Toulouse, France, 1993.
@inproceedings{Webb93e,
title = {Control, Capabilities and Communication: Three Key Issues for Machine-Expert Collaborative Knowledge Acquisition},
author = {G. I. Webb},
editor = {N. Aussenac and G. Boy and B. Gaines and M. Linster and J. G. Ganascia and Y. Kodratoff},
year = {1993},
date = {1993-01-01},
booktitle = {Proceedings (Complement) of the Seventh European Workshop on Knowledge Acquisition for Knowledge-based Systems (EWKA'93)},
pages = {263-275},
publisher = {Springer-Verlag},
address = {Toulouse, France},
abstract = {Machine learning and knowledge elicitation are different but complementary approaches to knowledge acquisition. On the face of it there are large potential gains to be reaped from the integration of these two knowledge acquisition techniques. Machine-expert collaborative knowledge acquisition combines these approaches by placing the machine learning system and the human expert as partners in the knowledge-acquisition task. This paper examines three key issues facing machine-expert collaborative knowledge-acquisition where should control reside, what capabilities should each partner bring to the task and how should the partners communicate?},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Machine learning and knowledge elicitation are different but complementary approaches to knowledge acquisition. On the face of it there are large potential gains to be reaped from the integration of these two knowledge acquisition techniques. Machine-expert collaborative knowledge acquisition combines these approaches by placing the machine learning system and the human expert as partners in the knowledge-acquisition task. This paper examines three key issues facing machine-expert collaborative knowledge-acquisition where should control reside, what capabilities should each partner bring to the task and how should the partners communicate?
Webb, G. I.
Feature Based Modelling: A Methodology for Producing Coherent, Consistent, Dynamically Changing Models of Agents Competency
Brna, P.; Ohlsson, S.; Pain, H. (Ed.): Proceedings of the 1993 World Conference on Artificial Intelligence in Education (AI-ED'93), pp. 497-504, AACE, Edinburgh, Scotland. Also published in User Modeling and User-Adapted Interaction. 5: 117-150, 1996, 1993.
@inproceedings{Webb93c,
title = {Feature Based Modelling: A Methodology for Producing Coherent, Consistent, Dynamically Changing Models of Agents Competency},
author = {G. I. Webb},
editor = {P. Brna and S. Ohlsson and H. Pain},
year = {1993},
date = {1993-01-01},
booktitle = {Proceedings of the 1993 World Conference on Artificial Intelligence in Education (AI-ED'93)},
pages = {497-504},
publisher = {AACE},
address = {Edinburgh, Scotland. Also published in User Modeling and User-Adapted Interaction. 5: 117-150, 1996},
abstract = {Feature Based Modelling uses attribute value machine learning techniques to model an agent's competency. This is achieved by creating a model describing the relationships between the features of the agent's actions and of the contexts in which those actions are performed. This paper describes techniques that have been developed for creating these models and for extracting key information therefrom. An overview is provided of previous studies that have evaluated the application of Feature Based Modelling in a number of educational contexts including piano keyboard playing, the unification of Prolog terms and elementary subtraction. These studies have demonstrated that the approach is applicable to a wide spectrum of domains. Classroom use has demonstrated the low computational overheads of the technique. A new study of the application of the approach to modelling elementary subtraction skills is presented. The approach demonstrates accuracy in excess of 90% when predicting student solutions. It also demonstrates the ability to identify and model students' buggy arithmetic procedures.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Feature Based Modelling uses attribute value machine learning techniques to model an agent's competency. This is achieved by creating a model describing the relationships between the features of the agent's actions and of the contexts in which those actions are performed. This paper describes techniques that have been developed for creating these models and for extracting key information therefrom. An overview is provided of previous studies that have evaluated the application of Feature Based Modelling in a number of educational contexts including piano keyboard playing, the unification of Prolog terms and elementary subtraction. These studies have demonstrated that the approach is applicable to a wide spectrum of domains. Classroom use has demonstrated the low computational overheads of the technique. A new study of the application of the approach to modelling elementary subtraction skills is presented. The approach demonstrates accuracy in excess of 90% when predicting student solutions. It also demonstrates the ability to identify and model students' buggy arithmetic procedures.
Webb, G. I.
Systematic Search for Categorical Attribute-Value Data-Driven Machine Learning
Rowles, C.; Liu, H.; Foo, N. (Ed.): Proceedings of the Sixth Australian Joint Conference on Artificial Intelligence (AI'93), pp. 342-347, World Scientific, Melbourne, Australia, 1993.
@inproceedings{Webb93a,
title = {Systematic Search for Categorical Attribute-Value Data-Driven Machine Learning},
author = {G. I. Webb},
editor = {C. Rowles and H. Liu and N. Foo},
year = {1993},
date = {1993-01-01},
booktitle = {Proceedings of the Sixth Australian Joint Conference on Artificial Intelligence (AI'93)},
pages = {342-347},
publisher = {World Scientific},
address = {Melbourne, Australia},
abstract = {Optimal Pruning for Unordered Search is a search algorithm that enables complete search through the space of possible disjuncts at the inner level of a covering algorithm. This algorithm takes as inputs an evaluation function, e, a training set, t, and a set of specialisation operators, o. It outputs a set of operators from o that creates a classifier that maximises e with respect to t. While OPUS has exponential worst case time complexity, the algorithm is demonstrated to reach solutions for complex real world domains within reasonable time frames. Indeed, for some domains, the algorithm exhibits greater computational efficiency than common heuristic search algorithms.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Optimal Pruning for Unordered Search is a search algorithm that enables complete search through the space of possible disjuncts at the inner level of a covering algorithm. This algorithm takes as inputs an evaluation function, e, a training set, t, and a set of specialisation operators, o. It outputs a set of operators from o that creates a classifier that maximises e with respect to t. While OPUS has exponential worst case time complexity, the algorithm is demonstrated to reach solutions for complex real world domains within reasonable time frames. Indeed, for some domains, the algorithm exhibits greater computational efficiency than common heuristic search algorithms.
Webb, G. I.; Brkic, N.
Learning Decision Lists by Prepending Inferred Rules
Sestito, S. (Ed.): Proceedings of the AI 93 Workshop on Machine Learning and Hybrid Systems, pp. 6-10, Melbourne, Australia, 1993.
@inproceedings{WebbBrkic93,
title = {Learning Decision Lists by Prepending Inferred Rules},
author = {G. I. Webb and N. Brkic},
editor = {S. Sestito},
year = {1993},
date = {1993-01-01},
booktitle = {Proceedings of the AI 93 Workshop on Machine Learning and Hybrid Systems},
pages = {6-10},
address = {Melbourne, Australia},
abstract = {This paper describes a new algorithm for learning decision lists that operates by prepending successive rules to front of the list under construction. This contrasts with the original decision list induction algorithm which operates by appending successive rules to end of the list under construction. The new algorithm is demonstrated in the majority of cases to produce smaller classifiers that provide improved predictive accuracy than those produced by the original decision list induction algorithm.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes a new algorithm for learning decision lists that operates by prepending successive rules to front of the list under construction. This contrasts with the original decision list induction algorithm which operates by appending successive rules to end of the list under construction. The new algorithm is demonstrated in the majority of cases to produce smaller classifiers that provide improved predictive accuracy than those produced by the original decision list induction algorithm.
Smith, P. A.; Webb, G. I.
Recent progress in the Development of a Debugging Assistant for Computer Programs
Chia, B.; Pennell, R.; Sims, R. (Ed.): A Future Promised: Proceedings of the Fifth Australian Society for Computers in Learning in Tertiary Education Conference (ASCILITE '92), pp. 351-356, Sydney, Australia, 1992.
@inproceedings{SmithWebb92,
title = {Recent progress in the Development of a Debugging Assistant for Computer Programs},
author = {P. A. Smith and G. I. Webb},
editor = {B. Chia and R. Pennell and R. Sims},
year = {1992},
date = {1992-01-01},
booktitle = {A Future Promised: Proceedings of the Fifth Australian Society for Computers in Learning in Tertiary Education Conference (ASCILITE '92)},
pages = {351-356},
address = {Sydney, Australia},
abstract = {We present recent progress in the development of a debugging assistant for helping novices debug their computer programs. Bradman, which is still in the implementation phase, is an interactive system which builds two models of the user's program - one reflecting what the program actually does and the other reflecting what the programmer intended to do. Conflicts between these two models are used by Bradman to find bugs in the program.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We present recent progress in the development of a debugging assistant for helping novices debug their computer programs. Bradman, which is still in the implementation phase, is an interactive system which builds two models of the user's program - one reflecting what the program actually does and the other reflecting what the programmer intended to do. Conflicts between these two models are used by Bradman to find bugs in the program.
Webb, G. I.
Man-Machine Collaboration for Knowledge Acquisition
Adams, A.; L., Sterling. (Ed.): Proceedings of the Fifth Australian Joint Conference on Artificial Intelligence (AI'92), pp. 329-334, World Scientific, Hobart, Tas., Australia, 1992.
@inproceedings{Webb92,
title = {Man-Machine Collaboration for Knowledge Acquisition},
author = {G. I. Webb},
editor = {A. Adams and Sterling. L.},
year = {1992},
date = {1992-01-01},
booktitle = {Proceedings of the Fifth Australian Joint Conference on Artificial Intelligence (AI'92)},
pages = {329-334},
publisher = {World Scientific},
address = {Hobart, Tas., Australia},
abstract = {Both machine learning and knowledge elicitation from human experts have unique strengths and weaknesses. Man-machine collaboration for knowledge acquisition allows both knowledge acquisition techniques to be employed hand- in-hand. The strengths of each can alleviate the other's weaknesses. This has the potential to both reduce the time taken to develop an expert system while increasing the quality of the finished product. This paper discusses techniques for man-machine collaboration for knowledge acquisition and describes Einstein, a computer system that implements those techniques},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Both machine learning and knowledge elicitation from human experts have unique strengths and weaknesses. Man-machine collaboration for knowledge acquisition allows both knowledge acquisition techniques to be employed hand- in-hand. The strengths of each can alleviate the other's weaknesses. This has the potential to both reduce the time taken to develop an expert system while increasing the quality of the finished product. This paper discusses techniques for man-machine collaboration for knowledge acquisition and describes Einstein, a computer system that implements those techniques
Webb, G. I.; Agar, J.
Inducing Diagnostic Rules For Glomerular Disease With The DLG Machine Learning Algorithm
Artificial Intelligence in Medicine, vol. 4, no. 6, pp. 419-430, 1992.
@article{WebbAgar92,
title = {Inducing Diagnostic Rules For Glomerular Disease With The DLG Machine Learning Algorithm},
author = {G. I. Webb and J. Agar},
doi = {10.1016/0933-3657(92)90010-M},
year = {1992},
date = {1992-01-01},
journal = {Artificial Intelligence in Medicine},
volume = {4},
number = {6},
pages = {419-430},
publisher = {Elsevier},
abstract = {A pilot study has applied the DLG machine learning algorithm to create expert systems for the assessment and interpretation of clinical and laboratory data in glomerular disease. Despite the limited size of the data-set and major deficiencies in the information recorded therein, promising results have been obtained. On average, 100 expert systems developed from different subsets of the database, had a diagnostic accuracy of 54.70% when applied to cases that had not been used in their development. This compares with an average diagnostic accuracy of 48.31% obtained by four expert clinicians and of 3.23% obtained by random diagnosis. The expert systems demonstrated increased accuracy (62.90% on average) when cases of diseases represented by less than twenty examples were discarded. These results suggest that database expansion may enable the induction of diagnostic rules that provide accurate non-invasive diagnosis of specific categories of glomerular disease.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
A pilot study has applied the DLG machine learning algorithm to create expert systems for the assessment and interpretation of clinical and laboratory data in glomerular disease. Despite the limited size of the data-set and major deficiencies in the information recorded therein, promising results have been obtained. On average, 100 expert systems developed from different subsets of the database, had a diagnostic accuracy of 54.70% when applied to cases that had not been used in their development. This compares with an average diagnostic accuracy of 48.31% obtained by four expert clinicians and of 3.23% obtained by random diagnosis. The expert systems demonstrated increased accuracy (62.90% on average) when cases of diseases represented by less than twenty examples were discarded. These results suggest that database expansion may enable the induction of diagnostic rules that provide accurate non-invasive diagnosis of specific categories of glomerular disease.
Yip, S.; Webb, G. I.
Discriminate Attribute Finding in Classification Learning
Adams, A.; Sterling, L. (Ed.): Proceedings of the Fifth Australian Joint Conference on Artificial Intelligence (AI'92), pp. 374-379, World Scientific, Hobart, Tas., Australia, 1992.
@inproceedings{YipWebb92b,
title = {Discriminate Attribute Finding in Classification Learning},
author = {S. Yip and G. I. Webb},
editor = {A. Adams and L. Sterling},
year = {1992},
date = {1992-01-01},
booktitle = {Proceedings of the Fifth Australian Joint Conference on Artificial Intelligence (AI'92)},
pages = {374-379},
publisher = {World Scientific},
address = {Hobart, Tas., Australia},
abstract = {This paper describes a method for extending domain models in classification learning by deriving new attributes from existing ones. The process starts by examining examples of different classes which have overlapping ranges in all of their numeric attribute values. Based on existing attributes, new attributes which enhance the distinguishability of a class are created. These additional attributes are then used in the subsequent classification learning process. The research revealed that this method can enable relationships between attributes to be incorporated in the classification procedures and, depending on the nature of data, significantly increase the coverage of class descriptions, improve the accuracy of classifying novel instances and reduce the number of clauses in class description when compared to classification learning alone. Evaluation with the data on iris flower classification showed that the classification accuracy is slightly improved and the number of clauses in the class description is significantly reduced.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes a method for extending domain models in classification learning by deriving new attributes from existing ones. The process starts by examining examples of different classes which have overlapping ranges in all of their numeric attribute values. Based on existing attributes, new attributes which enhance the distinguishability of a class are created. These additional attributes are then used in the subsequent classification learning process. The research revealed that this method can enable relationships between attributes to be incorporated in the classification procedures and, depending on the nature of data, significantly increase the coverage of class descriptions, improve the accuracy of classifying novel instances and reduce the number of clauses in class description when compared to classification learning alone. Evaluation with the data on iris flower classification showed that the classification accuracy is slightly improved and the number of clauses in the class description is significantly reduced.
Yip, S.; Webb, G. I.
Function Finding in Classification Learning
Proceedings of the Second Pacific Rim International Conference on Artificial Intelligence (PRICAI '92), pp. 555-561, Springer-Verlag, Seoul, Korea, 1992.
@inproceedings{YipWebb92a,
title = {Function Finding in Classification Learning},
author = {S. Yip and G. I. Webb},
year = {1992},
date = {1992-01-01},
booktitle = {Proceedings of the Second Pacific Rim International Conference on Artificial Intelligence (PRICAI '92)},
pages = {555-561},
publisher = {Springer-Verlag},
address = {Seoul, Korea},
abstract = {The paper describes a method for extending domain models in classification learning by deriving new attributes from existing attributes. The process starts by finding functional regularities within each class. Such regularities are then treated as additional attributes in the subsequent classification learning process. The research revealed that these techniques can reduce the number of clauses required to describe each class, enable functional regularities between attributes to be incorporated in the classification procedures and, depending on the nature of data, significantly increase the coverage of class descriptions and improve the accuracy of classifying novel instances when compared to classification learning alone.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The paper describes a method for extending domain models in classification learning by deriving new attributes from existing attributes. The process starts by finding functional regularities within each class. Such regularities are then treated as additional attributes in the subsequent classification learning process. The research revealed that these techniques can reduce the number of clauses required to describe each class, enable functional regularities between attributes to be incorporated in the classification procedures and, depending on the nature of data, significantly increase the coverage of class descriptions and improve the accuracy of classifying novel instances when compared to classification learning alone.
Agar, J.; Webb, G. I.
Application Of Machine Learning To A Renal Biopsy Data-Base
Nephrology, Dialysis and Transplantation, vol. 7, pp. 472-478, 1992.
@article{AgarWebb92,
title = {Application Of Machine Learning To A Renal Biopsy Data-Base},
author = {J. Agar and G. I. Webb},
url = {http://ndt.oxfordjournals.org/content/7/6/472.abstract},
year = {1992},
date = {1992-01-01},
journal = {Nephrology, Dialysis and Transplantation},
volume = {7},
pages = {472-478},
publisher = {Oxford University Press},
address = {Oxford UK},
abstract = {This pilot study has applied machine learning (artificial intelligence derived qualitative analysis procedures) to yield non-invasive techniques for the assessment and interpretation of clinical and laboratory data in glomerular disease. To evaluate the appropriateness of these techniques, they were applied to subsets of a small database of 284 case histories and the resulting procedures evaluated against the remaining cases. Over such evaluations, the following average diagnostic accuracies were obtained: microscopic polyarteritis, 95.37%; minimal lesion nephrotic syndrome, 96.50%; immunoglobulin A nephropathy, 81.26%; minor changes, 93.66%; lupus nephritis, 96.27%; focal glomerulosclerosis, 92.06%; mesangial proliferative glomerulonephritis, 92.56%; and membranous nephropathy, 92.56%. Although in general the new diagnostic system is not yet as accurate as the histological evaluation of renal biopsy specimens, it shows promise of adding a further dimension to the diagnostic process. When the machine learning techniques are applied to a larger database, greater diagnostic accuracy should be obtained. It may allow accurate non- invasive diagnosis of some cases of glomerular disease without the need for renal biopsy. This may reduce both the cost and the morbidity of the investigation of glomerular disease and may be of particular value in situations where renal biopsy is considered hazardous or contraindicated.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This pilot study has applied machine learning (artificial intelligence derived qualitative analysis procedures) to yield non-invasive techniques for the assessment and interpretation of clinical and laboratory data in glomerular disease. To evaluate the appropriateness of these techniques, they were applied to subsets of a small database of 284 case histories and the resulting procedures evaluated against the remaining cases. Over such evaluations, the following average diagnostic accuracies were obtained: microscopic polyarteritis, 95.37%; minimal lesion nephrotic syndrome, 96.50%; immunoglobulin A nephropathy, 81.26%; minor changes, 93.66%; lupus nephritis, 96.27%; focal glomerulosclerosis, 92.06%; mesangial proliferative glomerulonephritis, 92.56%; and membranous nephropathy, 92.56%. Although in general the new diagnostic system is not yet as accurate as the histological evaluation of renal biopsy specimens, it shows promise of adding a further dimension to the diagnostic process. When the machine learning techniques are applied to a larger database, greater diagnostic accuracy should be obtained. It may allow accurate non- invasive diagnosis of some cases of glomerular disease without the need for renal biopsy. This may reduce both the cost and the morbidity of the investigation of glomerular disease and may be of particular value in situations where renal biopsy is considered hazardous or contraindicated.
Kuzmycz, M.; Webb, G. I.
Evaluation of Feature Based Modelling in Subtraction
Frasson, C.; Gauthier, G.; McCalla, G. I. (Ed.): Lecture Notes in Computer Science Vol. 608: Proceedings of the Second International Conference on Intelligent Tutoring Systems (ITS'92), pp. 269-276, Springer-Verlag, Montrial, Canada, 1992.
@inproceedings{KuzmyczWebb92,
title = {Evaluation of Feature Based Modelling in Subtraction},
author = {M. Kuzmycz and G. I. Webb},
editor = {C. Frasson and G. Gauthier and G. I. McCalla},
year = {1992},
date = {1992-01-01},
booktitle = {Lecture Notes in Computer Science Vol. 608: Proceedings of the Second International Conference on Intelligent Tutoring Systems (ITS'92)},
pages = {269-276},
publisher = {Springer-Verlag},
address = {Montrial, Canada},
abstract = {One aim of intelligent tutoring systems is to tailor lessons to each individual student's needs. To do this a tutoring system requires a model of the student's knowledge. Cognitive modelling aims to produce a detailed explanation of the student's progress. Feature Based Modelling forms a cognitive model of the student by creating aspects of problem descriptions and of students' responses. This paper will discuss Feature Based Modelling and show the results of an evaluation carried out in the domain of elemental subtraction.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
One aim of intelligent tutoring systems is to tailor lessons to each individual student's needs. To do this a tutoring system requires a model of the student's knowledge. Cognitive modelling aims to produce a detailed explanation of the student's progress. Feature Based Modelling forms a cognitive model of the student by creating aspects of problem descriptions and of students' responses. This paper will discuss Feature Based Modelling and show the results of an evaluation carried out in the domain of elemental subtraction.
Smith, P. A.; Webb, G. I.
Debugging Using Partial Models
Godfrey, R. (Ed.): Simulation & Academic Gaming in Tertiary Education, The Proceedings of the Eighth Annual Conference of ASCILITE (ASCILITE '91), pp. 581-590, University of Tasmania, Launceston, TAS, Australia, 1991.
@inproceedings{SmithWebb91,
title = {Debugging Using Partial Models},
author = {P. A. Smith and G. I. Webb},
editor = {R. Godfrey},
year = {1991},
date = {1991-01-01},
booktitle = {Simulation \& Academic Gaming in Tertiary Education, The Proceedings of the Eighth Annual Conference of ASCILITE (ASCILITE '91)},
pages = {581-590},
publisher = {University of Tasmania},
address = {Launceston, TAS, Australia},
abstract = {We present initial work on an expert system that will help users to debug their Pascal programs. This system asks the user questions concerning attempts to build a 'partial model' of the program - a model of those aspects of the program likely to relate to the error. This contrasts with previous systems in which a complete model of the user's program is built and compared to templates of correct versions of the program. The advantages of this approach are greater flexibility, greater student involvement in the debugging process and lower computational overheads.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We present initial work on an expert system that will help users to debug their Pascal programs. This system asks the user questions concerning attempts to build a 'partial model' of the program - a model of those aspects of the program likely to relate to the error. This contrasts with previous systems in which a complete model of the user's program is built and compared to templates of correct versions of the program. The advantages of this approach are greater flexibility, greater student involvement in the debugging process and lower computational overheads.
Webb, G. I.
An Attribute-Value Machine Learning Approach To Student Modelling
Kay, J.; Quilici, A. (Ed.): Proceedings of the IJCAI Workshop W.4: Agent Modelling for Intelligent Interaction, pp. 128-136, Sydney, Australia, 1991.
@inproceedings{Webb91e,
title = {An Attribute-Value Machine Learning Approach To Student Modelling},
author = {G. I. Webb},
editor = {J. Kay and A. Quilici},
year = {1991},
date = {1991-01-01},
booktitle = {Proceedings of the IJCAI Workshop W.4: Agent Modelling for Intelligent Interaction},
pages = {128-136},
address = {Sydney, Australia},
abstract = {This paper describes an application of machine learning to student modelling. Unlike previous machine learning approaches to student modelling, the new approach is based on attribute-value machine learning. In contrast to many previous approaches it is not necessary for the lesson author to identify all forms of error that may be detected or to identify the possible approaches to problem solving in the domain that may be adopted. Rather, the lesson author need only identify the relevant attributes both of the tasks to be performed by the student and of the student's actions. The values of these attributes are automatically processed by the student modeler to produce the student model.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes an application of machine learning to student modelling. Unlike previous machine learning approaches to student modelling, the new approach is based on attribute-value machine learning. In contrast to many previous approaches it is not necessary for the lesson author to identify all forms of error that may be detected or to identify the possible approaches to problem solving in the domain that may be adopted. Rather, the lesson author need only identify the relevant attributes both of the tasks to be performed by the student and of the student's actions. The values of these attributes are automatically processed by the student modeler to produce the student model.
Webb, G. I.
Inside the Unification Tutor: The Architecture of an Intelligent Educational System
Godfrey, R. (Ed.): Simulation & Academic Gaming in Tertiary Education, The Proceedings of the Eighth Annual Conference of ASCILITE (ASCILITE '91), pp. 677-684, University of Tasmania, Launceston, 1991.
@inproceedings{Webb91b,
title = {Inside the Unification Tutor: The Architecture of an Intelligent Educational System},
author = {G. I. Webb},
editor = {R. Godfrey},
year = {1991},
date = {1991-01-01},
booktitle = {Simulation \& Academic Gaming in Tertiary Education, The Proceedings of the Eighth Annual Conference of ASCILITE (ASCILITE '91)},
pages = {677-684},
publisher = {University of Tasmania},
address = {Launceston},
abstract = {The Unification Tutor provides practice and tuition on the unification of terms from the Prolog programming language. It integrates multiple knowledge sources encompassing both performance and declarative knowledge. A key feature of the tutor is the use of a detailed student model. It has been used since 1989 in Computer Science courses at Deakin and La Trobe Universities. Previous papers have examined the student modelling component of this system. This paper investigates the internal operation of the Unification Tutor, the sub-systems it incorporates and their interaction.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The Unification Tutor provides practice and tuition on the unification of terms from the Prolog programming language. It integrates multiple knowledge sources encompassing both performance and declarative knowledge. A key feature of the tutor is the use of a detailed student model. It has been used since 1989 in Computer Science courses at Deakin and La Trobe Universities. Previous papers have examined the student modelling component of this system. This paper investigates the internal operation of the Unification Tutor, the sub-systems it incorporates and their interaction.
Webb, G. I.
Data Driven Inductive Refinement of Production Rules
Quinlan, R. (Ed.): Proceedings of the First Australian Workshop on Knowledge Acquisition for Knowledge-Based Systems (AKAW '91), pp. 44-52, University of Sydney Press., Pokolbin, NSW, Australia, 1991.
@inproceedings{Webb91d,
title = {Data Driven Inductive Refinement of Production Rules},
author = {G. I. Webb},
editor = {R. Quinlan},
year = {1991},
date = {1991-01-01},
booktitle = {Proceedings of the First Australian Workshop on Knowledge Acquisition for Knowledge-Based Systems (AKAW '91)},
pages = {44-52},
publisher = {University of Sydney Press.},
address = {Pokolbin, NSW, Australia},
abstract = {This paper presents algorithms for inductive refinement of production rules based on the DLG data-driven machine learning algorithm. These algorithms modify the input production rules with reference to a set of examples so as to ensure that all positive examples are covered and no negative examples are covered. The input production rules may either have been previously learnt by a machine learning system or be extracted from an existing expert system.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper presents algorithms for inductive refinement of production rules based on the DLG data-driven machine learning algorithm. These algorithms modify the input production rules with reference to a set of examples so as to ensure that all positive examples are covered and no negative examples are covered. The input production rules may either have been previously learnt by a machine learning system or be extracted from an existing expert system.
Webb, G. I.
Einstein: An Interactive Inductive Knowledge-Acquisition Tool
Proceedings of the Sixth Banff Knowledge Acquisition for Knowledge-Based Systems Workshop, pp. (36)1-16, Banff, Canada, 1991.
@inproceedings{Webb91c,
title = {Einstein: An Interactive Inductive Knowledge-Acquisition Tool},
author = {G. I. Webb},
year = {1991},
date = {1991-01-01},
booktitle = {Proceedings of the Sixth Banff Knowledge Acquisition for Knowledge-Based Systems Workshop},
pages = {(36)1-16},
address = {Banff, Canada},
abstract = {Einstein is a knowledge acquisition system that incorporates data-driven inductive rule development and refinement in a user driven production rule development and evaluation environment. This allows the user and the induction system to interact as a cooperative knowledge acquisition team. Unique features of this system include efficient automated inductive refinement of existing production rules, interactive user management of machine learning facilities, including local and global guidance, interactive specification of key examples and counter-examples and interactive case-based rule assessment.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Einstein is a knowledge acquisition system that incorporates data-driven inductive rule development and refinement in a user driven production rule development and evaluation environment. This allows the user and the induction system to interact as a cooperative knowledge acquisition team. Unique features of this system include efficient automated inductive refinement of existing production rules, interactive user management of machine learning facilities, including local and global guidance, interactive specification of key examples and counter-examples and interactive case-based rule assessment.
Webb, G. I.; Agar, J.
The Application of Machine Learning to the Diagnosis of Glomerular Disease
Sarmeinto, C. (Ed.): Proceedings of the IJCAI Workshop W.15: Representing Knowledge in Medical Decision Support Systems, pp. 8.1-8.8, Sydney, Australia, 1991.
@inproceedings{WebbAgar91,
title = {The Application of Machine Learning to the Diagnosis of Glomerular Disease},
author = {G. I. Webb and J. Agar},
editor = {C. Sarmeinto},
year = {1991},
date = {1991-01-01},
booktitle = {Proceedings of the IJCAI Workshop W.15: Representing Knowledge in Medical Decision Support Systems},
pages = {8.1-8.8},
address = {Sydney, Australia},
abstract = {A pilot study has applied the DLG machine learning algorithm to create expert systems for the assessment and interpretation of clinical and laboratory data in glomerular disease. Despite the limited size of the data-set and major deficiencies in the information recorded therein, for one of the conditions examined in this study, microscopic polyarteritis, a consistent diagnostic accuracy of 100% was obtained. With expansion of the data base, it is possible that techniques will be derived that provide accurate non-invasive diagnosis of some cases of glomerular disease, thus obviating the need for renal biopsy. Success in this project will result in significant reductions in both the cost and the morbidity associated with the investigation of glomerular disease.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
A pilot study has applied the DLG machine learning algorithm to create expert systems for the assessment and interpretation of clinical and laboratory data in glomerular disease. Despite the limited size of the data-set and major deficiencies in the information recorded therein, for one of the conditions examined in this study, microscopic polyarteritis, a consistent diagnostic accuracy of 100% was obtained. With expansion of the data base, it is possible that techniques will be derived that provide accurate non-invasive diagnosis of some cases of glomerular disease, thus obviating the need for renal biopsy. Success in this project will result in significant reductions in both the cost and the morbidity associated with the investigation of glomerular disease.
Kuzmycz, M.; Webb, G. I.
Modelling Elementary Subtraction: Intelligent Warfare Against Bugs
Godfrey, R. (Ed.): Simulation & Academic Gaming in Tertiary Education, The Proceedings of the Eighth Annual Conference of ASCILITE (ASCILITE '91), pp. 367-376, University of Tasmania, Launceston, TAS, Australia, 1991.
@inproceedings{KuzmyczWebb91,
title = {Modelling Elementary Subtraction: Intelligent Warfare Against Bugs},
author = {M. Kuzmycz and G. I. Webb},
editor = {R. Godfrey},
year = {1991},
date = {1991-01-01},
booktitle = {Simulation \& Academic Gaming in Tertiary Education, The Proceedings of the Eighth Annual Conference of ASCILITE (ASCILITE '91)},
pages = {367-376},
publisher = {University of Tasmania},
address = {Launceston, TAS, Australia},
abstract = {This paper discusses an intelligent system .that uses Input/ Output Cognitive Modelling (IOCM) techniques to form a model of the student. The paper describes FBM, an IOCM system that uses features to represent the inputs and outputs of the tasks being presented to the student and forms a relationship which describes in essence the knowledge the student has in the domain. Also presented is ASPMoRe, an intelligent tool that takes the model of the student and adapts the lesson to both refine the model and give the student practice in weak areas of his knowledge. Results have shown that the system can be an effective tool for educational purposes.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper discusses an intelligent system .that uses Input/ Output Cognitive Modelling (IOCM) techniques to form a model of the student. The paper describes FBM, an IOCM system that uses features to represent the inputs and outputs of the tasks being presented to the student and forms a relationship which describes in essence the knowledge the student has in the domain. Also presented is ASPMoRe, an intelligent tool that takes the model of the student and adapts the lesson to both refine the model and give the student practice in weak areas of his knowledge. Results have shown that the system can be an effective tool for educational purposes.
Sanzogni, L.; Surruwerra, F.; Webb, G. I.
Improving the Efficiency of Rule Based Expert Systems by Rule Activation
Journal of Experimental and Theoretical Artificial Intelligence, vol. 2, pp. 369-380, 1990.
@article{SanzogniSurruwerraWebb90,
title = {Improving the Efficiency of Rule Based Expert Systems by Rule Activation},
author = {L. Sanzogni and F. Surruwerra and G. I. Webb},
year = {1990},
date = {1990-01-01},
journal = {Journal of Experimental and Theoretical Artificial Intelligence},
volume = {2},
pages = {369-380},
publisher = {Taylor and Francis},
abstract = {In this paper we test a hypothesis that has shown promise in enhancing the efficiency (run-time) of rule-based systems. The results of our experiments suggest that the use of rule activation plays an active part in improving the performance of rule bases containing conflict sets.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In this paper we test a hypothesis that has shown promise in enhancing the efficiency (run-time) of rule-based systems. The results of our experiments suggest that the use of rule activation plays an active part in improving the performance of rule bases containing conflict sets.
Webb, G. I.
Rule Optimisation and Theory Optimisation: Heuristic Search Strategies for Data-Driven Machine Learning
Motada, H.; Mizoguchi, R.; Boose, J.; Gaines, B. (Ed.): Proceedings of the First Japanese Knowledge Acquisition for Knowledge-Based Systems Workshop (JKAW'90), pp. 219-232, IOS Press, Kyoto, Japan, 1990.
@inproceedings{Webb90,
title = {Rule Optimisation and Theory Optimisation: Heuristic Search Strategies for Data-Driven Machine Learning},
author = {G. I. Webb},
editor = {H. Motada and R. Mizoguchi and J. Boose and B. Gaines},
year = {1990},
date = {1990-01-01},
booktitle = {Proceedings of the First Japanese Knowledge Acquisition for Knowledge-Based Systems Workshop (JKAW'90)},
pages = {219-232},
publisher = {IOS Press},
address = {Kyoto, Japan},
abstract = {Previous implementations of the Aq algorithm have used rule optimisation search strategies to attempt to develop optimal classification procedures. These strategies involve generating successive characteristic descriptions each of which is individually of maximal value. This is contrasted with theory optimisation search strategies which, instead, generate successive complete classification procedures from which those with the maximal value are selected. These two strategies have been applied to the domain of the diagnosis of Immunoglobulin A Nephropathy disease. The theory optimisation strategy was observed to out perform the rule optimisation strategy.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Previous implementations of the Aq algorithm have used rule optimisation search strategies to attempt to develop optimal classification procedures. These strategies involve generating successive characteristic descriptions each of which is individually of maximal value. This is contrasted with theory optimisation search strategies which, instead, generate successive complete classification procedures from which those with the maximal value are selected. These two strategies have been applied to the domain of the diagnosis of Immunoglobulin A Nephropathy disease. The theory optimisation strategy was observed to out perform the rule optimisation strategy.
Webb, G. I.; Cumming, G.; Richards, T.; Yum, K-K.
Educational Evaluation of Feature Based Modelling in a Problem Solving Domain
Lewis, R.; Otsuki, S. (Ed.): Proceedings of the IFIP TC3 International Conference on Advanced Research on Computers in Education (ARCE'90), pp. 101-108, Elsevier, Tokyo, Japan, 1990.
@inproceedings{WebbCummingRichardsYum90,
title = {Educational Evaluation of Feature Based Modelling in a Problem Solving Domain},
author = {G. I. Webb and G. Cumming and T. Richards and K-K. Yum},
editor = {R. Lewis and S. Otsuki},
year = {1990},
date = {1990-01-01},
booktitle = {Proceedings of the IFIP TC3 International Conference on Advanced Research on Computers in Education (ARCE'90)},
pages = {101-108},
publisher = {Elsevier},
address = {Tokyo, Japan},
abstract = {Feature-Based Modelling is a machine learning based cognitive modelling methodology. An intelligent educational system has been implemented, for the purpose of evaluating the methodology, which helps students learn about the unification of terms from the Prolog programming language. The system has been used by Third Year Computer Science students at La Trobe University during September 1989. Students were randomly allocated to an Experimental condition, in which FBM modelling was used to select tasks, and give extra comments, or to a Control condition in which similar tasks and comments were given, but without FBM tailoring to the individual. Ratings of task appropriateness, and comment usefulness, were collected on-line as the students worked with the tutor; overall ratings were obtained by questionnaire at the end; and semester exam results were examined. Despite the fact that only a minority of students showed sufficient misunderstanding for FBM to have potential value, of the ten comparisons chat relate most directly to the aims of the Tutor, while in no case reaching significance, seven were in favour of the Tutor, and only two against. These preliminary results are very encouraging for the FBM principles of the Tutor.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Feature-Based Modelling is a machine learning based cognitive modelling methodology. An intelligent educational system has been implemented, for the purpose of evaluating the methodology, which helps students learn about the unification of terms from the Prolog programming language. The system has been used by Third Year Computer Science students at La Trobe University during September 1989. Students were randomly allocated to an Experimental condition, in which FBM modelling was used to select tasks, and give extra comments, or to a Control condition in which similar tasks and comments were given, but without FBM tailoring to the individual. Ratings of task appropriateness, and comment usefulness, were collected on-line as the students worked with the tutor; overall ratings were obtained by questionnaire at the end; and semester exam results were examined. Despite the fact that only a minority of students showed sufficient misunderstanding for FBM to have potential value, of the ten comparisons chat relate most directly to the aims of the Tutor, while in no case reaching significance, seven were in favour of the Tutor, and only two against. These preliminary results are very encouraging for the FBM principles of the Tutor.
Webb, G. I.
A Machine Learning Approach to Student Modelling
Proceedings of the Third Australian Joint Conference on Artificial Intelligence (AI 89), pp. 195-205, Melbourne, Australia, 1989.
@inproceedings{Webb89a,
title = {A Machine Learning Approach to Student Modelling},
author = {G. I. Webb},
year = {1989},
date = {1989-01-01},
booktitle = {Proceedings of the Third Australian Joint Conference on Artificial Intelligence (AI 89)},
pages = {195-205},
address = {Melbourne, Australia},
abstract = {This paper describes an application of established machine learning principles to student modelling. Unlike previous machine learning based approaches to student modelling, the new approach is based on attribute-value machine learning. In contrast to many previous approaches it is not necessary for the lesson author to identify all forms of error that may be detected. Rather, the lesson author need only identify the relevant attributes both of the tasks to be performed by the student and of the student's actions. The values of these attributes are automatically processed by the student modeler to produce the student model.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes an application of established machine learning principles to student modelling. Unlike previous machine learning based approaches to student modelling, the new approach is based on attribute-value machine learning. In contrast to many previous approaches it is not necessary for the lesson author to identify all forms of error that may be detected. Rather, the lesson author need only identify the relevant attributes both of the tasks to be performed by the student and of the student's actions. The values of these attributes are automatically processed by the student modeler to produce the student model.
Webb, G. I.
Courseware Abstraction: Reducing Development Costs While Producing Qualitative Improvements in CAL
Journal of Computer Assisted Learning, vol. 5, pp. 103-113, 1989.
@article{Webb89b,
title = {Courseware Abstraction: Reducing Development Costs While Producing Qualitative Improvements in CAL},
author = {G. I. Webb},
year = {1989},
date = {1989-01-01},
journal = {Journal of Computer Assisted Learning},
volume = {5},
pages = {103-113},
publisher = {Blackwell Publishing},
abstract = {Courseware abstraction is an approach to CAL whereby the lesson author creates a general parametetized CAL lesson that is then applied to many concrete examples. This approach has the following advantages over alternative approaches to lesson development: it is cost efficient; it facilitates lesson verification; it encourages the provision of as many examples as are desirable; it simplifies the selection of appropriate examples for presentation to each student; it provides a convenient framework for student evaluation, and it supports the development of factually exhaustive lessons. In short it provides qualitative improvements, while at the same time reducing lesson development costs. Although widely used, courseware abstraction has not previously been identified as an important CAL technique and its relative merits have never received attention. In particular, there has been a failure to recognize that generative CAL derives most of its power from the use of courseware abstraction.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Courseware abstraction is an approach to CAL whereby the lesson author creates a general parametetized CAL lesson that is then applied to many concrete examples. This approach has the following advantages over alternative approaches to lesson development: it is cost efficient; it facilitates lesson verification; it encourages the provision of as many examples as are desirable; it simplifies the selection of appropriate examples for presentation to each student; it provides a convenient framework for student evaluation, and it supports the development of factually exhaustive lessons. In short it provides qualitative improvements, while at the same time reducing lesson development costs. Although widely used, courseware abstraction has not previously been identified as an important CAL technique and its relative merits have never received attention. In particular, there has been a failure to recognize that generative CAL derives most of its power from the use of courseware abstraction.
Webb, G. I.; Cumming, G.; Richards, T.; Yum, K-K.
The Unification Tutor: An Intelligent Educational System in the Classroom
Bishop, G.; Baker, J. (Ed.): Proceedings of the Seventh Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE '89), pp. 408-420, Bond University, Gold Coast, QLD, Australia, 1989.
@inproceedings{WebbCummingRichardsYum89,
title = {The Unification Tutor: An Intelligent Educational System in the Classroom},
author = {G. I. Webb and G. Cumming and T. Richards and K-K. Yum},
editor = {G. Bishop and J. Baker},
year = {1989},
date = {1989-01-01},
booktitle = {Proceedings of the Seventh Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE '89)},
pages = {408-420},
publisher = {Bond University},
address = {Gold Coast, QLD, Australia},
abstract = {The Unification Tutor is experimental Intelligent Tutoring System for the domain of the unification of Prolog terms. It demonstrates the interactive use of Feature-Based Modelling - an approach to cognitive modelling that has been presented at previous ASCILITE Conferences (Webb, 1988b.) The Unification Tutor has been used by Third Year Computer Science students at La Trobe University during September 1989. This paper describes the Unification Tutor and evaluates its performance at La Trobe.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The Unification Tutor is experimental Intelligent Tutoring System for the domain of the unification of Prolog terms. It demonstrates the interactive use of Feature-Based Modelling - an approach to cognitive modelling that has been presented at previous ASCILITE Conferences (Webb, 1988b.) The Unification Tutor has been used by Third Year Computer Science students at La Trobe University during September 1989. This paper describes the Unification Tutor and evaluates its performance at La Trobe.
Richards, T.; Webb, G. I.; Craske, N.
Object-oriented Control for Intelligent Computer Assisted Learning Systems
Ercoli, P.; Lewis, R. (Ed.): Proceedings of the IFIP TC3 Working Conference on Artificial Intelligence Tools in Education, pp. 203-219, Elsevier, Frascati, Italy, 1988.
@inproceedings{RichardsWebbCraske88,
title = {Object-oriented Control for Intelligent Computer Assisted Learning Systems},
author = {T. Richards and G. I. Webb and N. Craske},
editor = {P. Ercoli and R. Lewis},
year = {1988},
date = {1988-01-01},
booktitle = {Proceedings of the IFIP TC3 Working Conference on Artificial Intelligence Tools in Education},
pages = {203-219},
publisher = {Elsevier},
address = {Frascati, Italy},
abstract = {This paper investigates an approach to providing a general-purpose authoring/tutoring shell for intelligent computer assisted learning systems. The approach is to outline an object-oriented representation of task/goal hierarchies, then to consider the ways in which domain expertise, student information and teacher expertise can be made to interact with such hierarchies. The result is a skeleton in terms of which exploratory lessons are being constructed; and in terms of which further research on the domain expert system, student modelling, educational expertise modelling, and user interfacing can more concretely be developed.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper investigates an approach to providing a general-purpose authoring/tutoring shell for intelligent computer assisted learning systems. The approach is to outline an object-oriented representation of task/goal hierarchies, then to consider the ways in which domain expertise, student information and teacher expertise can be made to interact with such hierarchies. The result is a skeleton in terms of which exploratory lessons are being constructed; and in terms of which further research on the domain expert system, student modelling, educational expertise modelling, and user interfacing can more concretely be developed.
Webb, G. I.
Techniques for Efficient Empirical Induction
Barter, C. J.; Brooks, M. J. (Ed.): Lecture Notes in Artificial Intelligence Vol. 406: Proceedings of the Second Australian Joint Conference on Artificial Intelligence (AI'88), pp. 225-239, Springer-Verlag, Adelaide, S.A., Australia, 1988.
@inproceedings{Webb88b,
title = {Techniques for Efficient Empirical Induction},
author = {G. I. Webb},
editor = {C. J. Barter and M. J. Brooks},
year = {1988},
date = {1988-01-01},
booktitle = {Lecture Notes in Artificial Intelligence Vol. 406: Proceedings of the Second Australian Joint Conference on Artificial Intelligence (AI'88)},
pages = {225-239},
publisher = {Springer-Verlag},
address = {Adelaide, S.A., Australia},
abstract = {This paper describes the LEI algorithm for empirical induction. The LEI algorithm provides efficient empirical induction for discrete attribute value data. It derives a classification procedure in the form of a set of predicate logic classification rules. This contrasts with the only other efficient approach to exhaustive empirical induction, the derivatives of the CLS algorithm, which present their classification procedures in the form of a decision tree. The LEI algorithm will always find the simplest non-disjunctive rule that correctly classifies all examples of a single class where such a rule exists.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper describes the LEI algorithm for empirical induction. The LEI algorithm provides efficient empirical induction for discrete attribute value data. It derives a classification procedure in the form of a set of predicate logic classification rules. This contrasts with the only other efficient approach to exhaustive empirical induction, the derivatives of the CLS algorithm, which present their classification procedures in the form of a decision tree. The LEI algorithm will always find the simplest non-disjunctive rule that correctly classifies all examples of a single class where such a rule exists.
Webb, G. I.
Cognitive Diagnosis Using Student Attributions
Fielden, K.; Hicks, F.; Scott, N. (Ed.): Computers in Learning in Tertiary education: Proceedings of the Sixth Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE-88), pp. 502-514, Canberra, Australia, 1988.
@inproceedings{Webb88a,
title = {Cognitive Diagnosis Using Student Attributions},
author = {G. I. Webb},
editor = {K. Fielden and F. Hicks and N. Scott},
year = {1988},
date = {1988-01-01},
booktitle = {Computers in Learning in Tertiary education: Proceedings of the Sixth Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE-88)},
pages = {502-514},
address = {Canberra, Australia},
abstract = {This paper details an approach to cognitive diagnosis that enables the inference of detailed models of a student's conceptualisation of a domain. This model is constructed be examining the attributes of the problems that the student has tackled and the student's performance while tackling those problems. A feature network is used to represent educationally relevant domain knowledge. This approach has low implementation and operational overheads; It provides a detailed model of the student's conceptualisation of the subject domain in terms of elements of knowledge from that domain; Student models are not restricted to overlays of predefined correct and/or incorrect knowledge; It does not require that the instructional designer anticipate the possible forms of error that may occur; It is robust in the face of partial evaluation of student performance; It is also robust in the face of the instructional designer's failure to incorporate relevant aspects of the subject domain in the knowledge-base; The student models can be executed; It supports accurate diagnosis of multiple viewpoints of the domain even when those viewpoints are not anticipated by the instructional designer; It can support multiple teaching styles in the one lesson.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper details an approach to cognitive diagnosis that enables the inference of detailed models of a student's conceptualisation of a domain. This model is constructed be examining the attributes of the problems that the student has tackled and the student's performance while tackling those problems. A feature network is used to represent educationally relevant domain knowledge. This approach has low implementation and operational overheads; It provides a detailed model of the student's conceptualisation of the subject domain in terms of elements of knowledge from that domain; Student models are not restricted to overlays of predefined correct and/or incorrect knowledge; It does not require that the instructional designer anticipate the possible forms of error that may occur; It is robust in the face of partial evaluation of student performance; It is also robust in the face of the instructional designer's failure to incorporate relevant aspects of the subject domain in the knowledge-base; The student models can be executed; It supports accurate diagnosis of multiple viewpoints of the domain even when those viewpoints are not anticipated by the instructional designer; It can support multiple teaching styles in the one lesson.
Webb, G. I.
A Knowledge-Based Approach To Computer-Aided Learning
International Journal of Man-Machine Studies, vol. 29, pp. 257-285, 1988.
@article{Webb88c,
title = {A Knowledge-Based Approach To Computer-Aided Learning},
author = {G. I. Webb},
year = {1988},
date = {1988-01-01},
journal = {International Journal of Man-Machine Studies},
volume = {29},
pages = {257-285},
publisher = {Academic Press},
abstract = {This paper describes a methodology for the creation of knowledge-based computer- aided learning lessons. Unlike previous approaches, the knowledge base is utilized only for restricted aspects of the lesson - both for the management of flow of control through a body of instructional materials and for the evaluation of the student's understanding of the subject matter. This has many advantages. While the approach has lower developmental and operational overheads than alternatives it is also able to perform far more flexible evaluations of the student's performance. As flow of control is managed by a knowledge-based component with reference to a detailed analysis of the student's understanding of the subject matter, lessons adapt to each student's individual understanding and aptitude within a domain.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper describes a methodology for the creation of knowledge-based computer- aided learning lessons. Unlike previous approaches, the knowledge base is utilized only for restricted aspects of the lesson - both for the management of flow of control through a body of instructional materials and for the evaluation of the student's understanding of the subject matter. This has many advantages. While the approach has lower developmental and operational overheads than alternatives it is also able to perform far more flexible evaluations of the student's performance. As flow of control is managed by a knowledge-based component with reference to a detailed analysis of the student's understanding of the subject matter, lessons adapt to each student's individual understanding and aptitude within a domain.
Webb, G. I.
Generative CAL and Courseware Abstraction
Barrett, J.; Hedberg, J. (Ed.): Using computers intelligently in Tertiary Education: Proceedings of the Fifth Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE-87), pp. 257-285, Sydney, Australia, 1987.
@inproceedings{Webb87a,
title = {Generative CAL and Courseware Abstraction},
author = {G. I. Webb},
editor = {J. Barrett and J. Hedberg},
year = {1987},
date = {1987-01-01},
booktitle = {Using computers intelligently in Tertiary Education: Proceedings of the Fifth Annual Conference of the Australian Society for Computers in Learning in Tertiary Education (ASCILITE-87)},
pages = {257-285},
address = {Sydney, Australia},
abstract = {Courseware abstraction is an approach to CAL whereby the lesson author creates a general parameterised CAL lesson that is then applied to many concrete examples. This approach has the following advantages: 1. it provides a powerful framework within which to adapt tuition to a student's knowledge and aptitude; 2. it encourages the development of detailed treatments of the subject matter; it reduces the cost of lesson development as a ratio to student lesson time; 3. and it enables large numbers of examples to be made available for individual students. Generative CAL is an example of courseware abstraction. It is argued that the advantages of generative CAL do not arise directly from the generation of the examples to be examined but rather can be directly attributed to the use of courseware abstraction.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Courseware abstraction is an approach to CAL whereby the lesson author creates a general parameterised CAL lesson that is then applied to many concrete examples. This approach has the following advantages: 1. it provides a powerful framework within which to adapt tuition to a student's knowledge and aptitude; 2. it encourages the development of detailed treatments of the subject matter; it reduces the cost of lesson development as a ratio to student lesson time; 3. and it enables large numbers of examples to be made available for individual students. Generative CAL is an example of courseware abstraction. It is argued that the advantages of generative CAL do not arise directly from the generation of the examples to be examined but rather can be directly attributed to the use of courseware abstraction.
Webb, G. I.
Domain and Tutoring Knowledge in Computer-Aided Learning
Gero, J.; Sudweeks, F. (Ed.): Proceedings of the First Australian Joint Conference on Artificial Intelligence (AI'87), pp. 488-502, The University of Sydney Printing Service, Sydney, Australia, 1987.
@inproceedings{Webb87b,
title = {Domain and Tutoring Knowledge in Computer-Aided Learning},
author = {G. I. Webb},
editor = {J. Gero and F. Sudweeks},
year = {1987},
date = {1987-01-01},
booktitle = {Proceedings of the First Australian Joint Conference on Artificial Intelligence (AI'87)},
pages = {488-502},
publisher = {The University of Sydney Printing Service},
address = {Sydney, Australia},
abstract = {Previous approaches to the utilisation of expertise in computer-aided learning have emphasised expertise in the subject domain. By contrast, this paper details an approach that emphasises tutoring expertise and only relies on minimal domain expertise. This has several advantages. The intelligent use of restricted domain expertise enables the detailed evaluation of the students' understanding of the domain. This permits the provision of very flexible tuition that uniquely adjusts to each student's understanding of the domain. Further, due to the restricted nature of the domain knowledge that is required, the developmental overheads associated with a lesson are minimal. Finally, the type of domain knowledge required has a well defined semantics further enabling its intelligent manipulation. The approach described is domain independent. This paper describes the general system architecture, the knowledge representation formalism used and the tutoring strategies that are employed.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Previous approaches to the utilisation of expertise in computer-aided learning have emphasised expertise in the subject domain. By contrast, this paper details an approach that emphasises tutoring expertise and only relies on minimal domain expertise. This has several advantages. The intelligent use of restricted domain expertise enables the detailed evaluation of the students' understanding of the domain. This permits the provision of very flexible tuition that uniquely adjusts to each student's understanding of the domain. Further, due to the restricted nature of the domain knowledge that is required, the developmental overheads associated with a lesson are minimal. Finally, the type of domain knowledge required has a well defined semantics further enabling its intelligent manipulation. The approach described is domain independent. This paper describes the general system architecture, the knowledge representation formalism used and the tutoring strategies that are employed.
Webb, G. I.
Knowledge Based Flow of Control in Computer-Aided Learning
Proceedings of the First Australian Artificial Intelligence Congress (1AAIC'86), pp. B: 1-7, Melbourne, Australia, 1986.
@inproceedings{Webb86,
title = {Knowledge Based Flow of Control in Computer-Aided Learning},
author = {G. I. Webb},
year = {1986},
date = {1986-01-01},
booktitle = {Proceedings of the First Australian Artificial Intelligence Congress (1AAIC'86)},
pages = {B: 1-7},
address = {Melbourne, Australia},
abstract = {In this paper I examine the utilisation of knowledge representation in Computer-Aided Learning (CAL) with the aim of establishing knowledge-based CAL techniques that are best suited to current technology. Most existing knowledge-based CAL systems attempt to generate the entire instructional sequence directly from a domain knowledge base. Such systems suffer from several limitations. These limitations include: 1. It is questionable whether the techniques exist to produce such systems for any but a highly restricted set of domains. 2. Even for those domains in which such systems can be produced the overheads are prohibitive for most purposes. Given these limitations, I argue that knowledge representation should be utilised in CAL only for those aspects of the instructional process for which it results in substantial gains without prohibitive overheads. I demonstrate that one aspect of CAL for which this holds is for managing flow of control within instructional material. I provide a detailed description of feature networks. These are a variant of M.A.K. Halliday's system network formalism. Feature networks are a knowledge representation formalism that efficiently encodes exactly the knowledge that is required for knowledge-based flow of control. It is shown that computer based lessons that utilise feature networks for control flow of control are extremely economic in terms of both authoring time and computer resources while providing highly responsive tuition. DABIS, a system that embodies the methodology outlined above, has been implemented and is described. DABIS demonstrates the feasibility of this methodology. There are no fundamental restrictions to the domains to which it can be applied. A lesson created under the DABIS system has been found to have an authoring to student time ration of approximately 12:1. Lessons created by the system also demonstrate the sensitivity to a student's knowledge and abilities within a domain that results from the intelligent use of knowledge-based CAL.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
In this paper I examine the utilisation of knowledge representation in Computer-Aided Learning (CAL) with the aim of establishing knowledge-based CAL techniques that are best suited to current technology. Most existing knowledge-based CAL systems attempt to generate the entire instructional sequence directly from a domain knowledge base. Such systems suffer from several limitations. These limitations include: 1. It is questionable whether the techniques exist to produce such systems for any but a highly restricted set of domains. 2. Even for those domains in which such systems can be produced the overheads are prohibitive for most purposes. Given these limitations, I argue that knowledge representation should be utilised in CAL only for those aspects of the instructional process for which it results in substantial gains without prohibitive overheads. I demonstrate that one aspect of CAL for which this holds is for managing flow of control within instructional material. I provide a detailed description of feature networks. These are a variant of M.A.K. Halliday's system network formalism. Feature networks are a knowledge representation formalism that efficiently encodes exactly the knowledge that is required for knowledge-based flow of control. It is shown that computer based lessons that utilise feature networks for control flow of control are extremely economic in terms of both authoring time and computer resources while providing highly responsive tuition. DABIS, a system that embodies the methodology outlined above, has been implemented and is described. DABIS demonstrates the feasibility of this methodology. There are no fundamental restrictions to the domains to which it can be applied. A lesson created under the DABIS system has been found to have an authoring to student time ration of approximately 12:1. Lessons created by the system also demonstrate the sensitivity to a student's knowledge and abilities within a domain that results from the intelligent use of knowledge-based CAL.
Webb, G. I.
The Domain-Analysis Based Instruction System
Bishop, G.; vanLint, W. (Ed.): Proceedings of the Fourth Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'86), pp. 295-302, University of Adelaide, Adelaide, Australia, 1986.
@inproceedings{Webb86a,
title = {The Domain-Analysis Based Instruction System},
author = {G. I. Webb},
editor = {G. Bishop and W. vanLint},
year = {1986},
date = {1986-01-01},
booktitle = {Proceedings of the Fourth Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'86)},
pages = {295-302},
publisher = {University of Adelaide},
address = {Adelaide, Australia},
abstract = {At the past two CALITE conferences I have described a methodology for creating knowledge-based CAL. This paper describes how that methodology has evolved. The Domain-Analysis Based Instruction System, a CAL system that utilises the methodology is then described in detail.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
At the past two CALITE conferences I have described a methodology for creating knowledge-based CAL. This paper describes how that methodology has evolved. The Domain-Analysis Based Instruction System, a CAL system that utilises the methodology is then described in detail.
Richards, T.; Webb, G. I.
ECCLES An Expert System for CAL
Garrett, H. (Ed.): Proceedings of the Tenth Western Educational Computing Conference (WECC'85), pp. 151-157, : Western Periodicals Company, Oakland, CA, 1985.
@inproceedings{RichardsWebb85,
title = {ECCLES An Expert System for CAL},
author = {T. Richards and G. I. Webb},
editor = {H. Garrett},
year = {1985},
date = {1985-01-01},
booktitle = {Proceedings of the Tenth Western Educational Computing Conference (WECC'85)},
pages = {151-157},
publisher = {: Western Periodicals Company},
address = {Oakland, CA},
abstract = {An authoring and lesson management system is described for Computer Assisted Learning in which lesson questioning and control flow arising from student response are generated at lesson time from an internal model of the lesson topic area. This approach permits the rapid authoring of conceptually complex lesson material. A languageless menu-driven authoring system rninimises the system familiarization time for lesson authors and enforces the construction of logically complete lessons. From the student's point of view, the system appears as an expert system in the lesson subject area, with precise and detailed knowledge of the topic taught and of the causes of the student's own errors.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
An authoring and lesson management system is described for Computer Assisted Learning in which lesson questioning and control flow arising from student response are generated at lesson time from an internal model of the lesson topic area. This approach permits the rapid authoring of conceptually complex lesson material. A languageless menu-driven authoring system rninimises the system familiarization time for lesson authors and enforces the construction of logically complete lessons. From the student's point of view, the system appears as an expert system in the lesson subject area, with precise and detailed knowledge of the topic taught and of the causes of the student's own errors.
Webb, G. I.
Student Control Under the Feature-Network Based Courseware Design Methodology
Bowden, J. A.; Lichtenstein, S. (Ed.): Student Control of Learning: Proceedings of the Third Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'85), pp. 27-34, University of Melbourne, Melbourne, Australia, 1985.
@inproceedings{Webb85a,
title = {Student Control Under the Feature-Network Based Courseware Design Methodology},
author = {G. I. Webb},
editor = {J. A. Bowden and S. Lichtenstein},
year = {1985},
date = {1985-01-01},
booktitle = {Student Control of Learning: Proceedings of the Third Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'85)},
pages = {27-34},
publisher = {University of Melbourne},
address = {Melbourne, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Richards, T.; Webb, G. I.; Bodnar, S.
ECCLES An Intelligent C.A.L. System
Russell, R. (Ed.): Proceedings of the Second Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'84), pp. 232-235, University of Queensland, Brisbane, Australia, 1984.
@inproceedings{RichardsWebbBodnar84,
title = {ECCLES An Intelligent C.A.L. System},
author = {T. Richards and G. I. Webb and S. Bodnar},
editor = {R. Russell},
year = {1984},
date = {1984-01-01},
booktitle = {Proceedings of the Second Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'84)},
pages = {232-235},
publisher = {University of Queensland},
address = {Brisbane, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Webb, G. I.
A Methodology for Intermediate Level Knowledge Representation in CAL
Russell, R. (Ed.): Proceedings of the Second Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'84),, pp. 288-303, University of Queensland, Brisbane, Australia, 1984.
@inproceedings{Webb84,
title = {A Methodology for Intermediate Level Knowledge Representation in CAL},
author = {G. I. Webb},
editor = {R. Russell},
year = {1984},
date = {1984-01-01},
booktitle = {Proceedings of the Second Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'84),},
pages = {288-303},
publisher = {University of Queensland},
address = {Brisbane, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Richards, T.; Hooke, R.; Webb, G. I.
Automatic Authoring of Complex and Analytical Question-Answer Lessons
Russell, R. (Ed.): Proceedings of the First Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'83), pp. 282-293, University of Queensland, Brisbane, Australia, 1983.
@inproceedings{RichardsHookeWebb83,
title = {Automatic Authoring of Complex and Analytical Question-Answer Lessons},
author = {T. Richards and R. Hooke and G. I. Webb},
editor = {R. Russell},
year = {1983},
date = {1983-01-01},
booktitle = {Proceedings of the First Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'83)},
pages = {282-293},
publisher = {University of Queensland},
address = {Brisbane, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Webb, G. I.
A Computer Based Language Instructional Expert
Russell, R. (Ed.): Proceedings of the First Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'83), pp. 391-402, University of Queensland, Brisbane, Australia, 1983.
@inproceedings{Webb83,
title = {A Computer Based Language Instructional Expert},
author = {G. I. Webb},
editor = {R. Russell},
year = {1983},
date = {1983-01-01},
booktitle = {Proceedings of the First Annual Computer-Assisted Learning in Tertiary Education Conference (CALITE'83)},
pages = {391-402},
publisher = {University of Queensland},
address = {Brisbane, Australia},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}