John Keane
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
John Keane has a total of 91 co-authors. Here are the top co-authors by number of shared publications:
- Xiao-Jun Zeng
University of Manchester
12 shared publications - Goran Nenadic
Department of Computer Science, University of Manchester, Manchester, UK
4 shared publications - Cameron Shand
The Francis Crick Institute, Software Engineering & AI STP, London, UK
2 shared publications - David Tian
Sheffield Hallam University
2 shared publications - Di Wang
School of Computer Science, University of Manchester, M60 1QD, U.K.
2 shared publications - Edward Abel
University of Manchester, Manchester, United Kingdom
2 shared publications
Latest Publications
- Open Access
HAWKS: Evolving Challenging Benchmark Sets for Cluster Analysis
Comprehensive benchmarking of clustering algorithms is rendered difficult by two key factors: (i)~the elusiveness of a unique mathematical definition of this unsupervised learning approach and (ii)~dependencies between the generating models or clustering criteria adopted by some clustering algorithms and indices for internal cluster validation. Consequently, there is no consensus regarding the best practice for rigorous benchmarking, and whether this is possible at all outside the context of a given application. Here, we argue that synthetic datasets must continue to play an important role in the evaluation of clustering algorithms, but that this necessitates constructing benchmarks that appropriately cover the diverse set of properties that impact clustering algorithm performance. Through our framework, HAWKS, we demonstrate the important role evolutionary algorithms play to support flexible generation of such benchmarks, allowing simple modification and extension. We illustrate two possible uses of our framework: (i)~the evolution of benchmark data consistent with a set of hand-derived properties and (ii)~the generation of datasets that tease out performance differences between a given pair of algorithms. Our work has implications for the design of clustering benchmarks that sufficiently challenge a broad range of algorithms, and for furthering insight into the strengths and weaknesses of specific approaches.
Open AccessHAWKS: Evolving Challenging Benchmark Sets for Cluster Analysis
Comprehensive benchmarking of clustering algorithms is rendered difficult by two key factors: (i)~the elusiveness of a unique mathematical definition of this unsupervised learning approach and (ii)~dependencies between the generating models or clustering criteria adopted by some clustering algorithms and indices for internal cluster validation. Consequently, there is no consensus regarding the best practice for rigorous benchmarking, and whether this is possible at all outside the context of a given application. Here, we argue that synthetic datasets must continue to play an important role in the evaluation of clustering algorithms, but that this necessitates constructing benchmarks that appropriately cover the diverse set of properties that impact clustering algorithm performance. Through our framework, HAWKS, we demonstrate the important role evolutionary algorithms play to support flexible generation of such benchmarks, allowing simple modification and extension. We illustrate two possible uses of our framework: (i)~the evolution of benchmark data consistent with a set of hand-derived properties and (ii)~the generation of datasets that tease out performance differences between a given pair of algorithms. Our work has implications for the design of clustering benchmarks that sufficiently challenge a broad range of algorithms, and for furthering insight into the strengths and weaknesses of specific approaches.
- 21 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Evolutionary Computation
- Vol. 26 (6) , 1206-1220
HAWKS: Evolving Challenging Benchmark Sets for Cluster Analysis
Comprehensive benchmarking of clustering algorithms is rendered difficult by two key factors: (i) the elusiveness of a unique mathematical definition of this unsupervised learning approach and (ii) dependencies between the generating models or clustering criteria adopted by some clustering algorithms and indices for internal cluster validation. Consequently, there is no consensus regarding the best practice for rigorous benchmarking, and whether this is possible at all outside the context of a given application. Here, we argue that synthetic datasets must continue to play an important role in the evaluation of clustering algorithms, but that this necessitates constructing benchmarks that appropriately cover the diverse set of properties that impact clustering algorithm performance. Through our framework, HAWKS, we demonstrate the important role evolutionary algorithms play to support flexible generation of such benchmarks, allowing simple modification and extension. We illustrate two possible uses of our framework: (i) the evolution of benchmark data consistent with a set of hand-derived properties and (ii) the generation of datasets that tease out performance differences between a given pair of algorithms. Our work has implications for the design of clustering benchmarks that sufficiently challenge a broad range of algorithms, and for furthering insight into the strengths and weaknesses of specific approaches.
- 21 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Evolutionary Computation
- Vol. 26 (6) , 1206-1220
HAWKS: Evolving Challenging Benchmark Sets for Cluster Analysis
Comprehensive benchmarking of clustering algorithms is rendered difficult by two key factors: (i) the elusiveness of a unique mathematical definition of this unsupervised learning approach and (ii) dependencies between the generating models or clustering criteria adopted by some clustering algorithms and indices for internal cluster validation. Consequently, there is no consensus regarding the best practice for rigorous benchmarking, and whether this is possible at all outside the context of a given application. Here, we argue that synthetic datasets must continue to play an important role in the evaluation of clustering algorithms, but that this necessitates constructing benchmarks that appropriately cover the diverse set of properties that impact clustering algorithm performance. Through our framework, HAWKS, we demonstrate the important role evolutionary algorithms play to support flexible generation of such benchmarks, allowing simple modification and extension. We illustrate two possible uses of our framework: (i) the evolution of benchmark data consistent with a set of hand-derived properties and (ii) the generation of datasets that tease out performance differences between a given pair of algorithms. Our work has implications for the design of clustering benchmarks that sufficiently challenge a broad range of algorithms, and for furthering insight into the strengths and weaknesses of specific approaches.
- Open Access
- 18 November 2020
- Published by Wiley in International Transactions in Operational Research
- Vol. 29 (5) , 3190-3206
Pairwise comparisons or constrained optimization? A usability evaluation of techniques for eliciting decision priorities
Decision support methodologies provide notations for expressing and communicating the priorities that inform a decision. Although a substantial literature has explored the theoretical merits of such notations and methodologies, much less work has investigated their usability in practice, which is of vital importance for their widespread adoption by users. In this paper, we explore the usability of two well‐known preference elicitation techniques, pairwise comparisons and constrained optimization. The techniques were explored through two contrasting crowd worker experiments, a preliminary one evaluating recognition, that is, the ability to identify the most suitable formulation for a given task, and the other synthesis, that is, the ability to construct formulations for a given task. The tasks are based on a case study involving source selection, a well‐known problem in the data integration domain. The results of the empirical evaluation show that, overall, pairwise comparisons resulted in significantly higher performance than constrained optimization, yet there is negligible difference between the usability appraisals for each technique. Furthermore, we observed that the technique that participants perform better with is not necessarily the one that they consider more usable.
- 18 November 2020
- Published by Wiley in International Transactions in Operational Research
- Vol. 29 (5) , 3190-3206
Open AccessPairwise comparisons or constrained optimization? A usability evaluation of techniques for eliciting decision priorities
Decision support methodologies provide notations for expressing and communicating the priorities that inform a decision. Although a substantial literature has explored the theoretical merits of such notations and methodologies, much less work has investigated their usability in practice, which is of vital importance for their widespread adoption by users. In this paper, we explore the usability of two well‐known preference elicitation techniques, pairwise comparisons and constrained optimization. The techniques were explored through two contrasting crowd worker experiments, a preliminary one evaluating recognition, that is, the ability to identify the most suitable formulation for a given task, and the other synthesis, that is, the ability to construct formulations for a given task. The tasks are based on a case study involving source selection, a well‐known problem in the data integration domain. The results of the empirical evaluation show that, overall, pairwise comparisons resulted in significantly higher performance than constrained optimization, yet there is negligible difference between the usability appraisals for each technique. Furthermore, we observed that the technique that participants perform better with is not necessarily the one that they consider more usable.
- Open Access
An End-to-End, Real-Time Solution for Condition Monitoring of Wind Turbine Generators
Data-driven wind generator condition monitoring systems largely rely on multi-stage processing involving feature selection and extraction followed by supervised learning. These stages require expert analysis, are potentially error-prone and do not generalize well between applications. In this paper, we introduce a collection of end-to-end Convolutional Neural Networks for advanced condition monitoring of wind turbine generators. End-to-end models have the benefit of utilizing raw, unstructured signals to make predictions about the parameters of interest. This feature makes it easier to scale an existing collection of models to new predictive tasks (e.g., new failure types) since feature extracting steps are not required. These automated models achieve low Mean Squared Errors in predicting the generator operational state (40.85 for Speed and 0.0018 for Load) and high accuracy in diagnosing rotor demagnetization failures (99.67%) by utilizing only raw current signals. We show how to create, deploy and run the collection of proposed models in a real-time setting using a laptop connected to a test rig via a data acquisition card. Based on a sampling rate of 5 kHz, predictions are stored in an efficient time series database and monitored using a dynamic visualization framework. We further discuss existing options for understanding the decision process behind the predictions made by the models.
Open AccessAn End-to-End, Real-Time Solution for Condition Monitoring of Wind Turbine Generators
Data-driven wind generator condition monitoring systems largely rely on multi-stage processing involving feature selection and extraction followed by supervised learning. These stages require expert analysis, are potentially error-prone and do not generalize well between applications. In this paper, we introduce a collection of end-to-end Convolutional Neural Networks for advanced condition monitoring of wind turbine generators. End-to-end models have the benefit of utilizing raw, unstructured signals to make predictions about the parameters of interest. This feature makes it easier to scale an existing collection of models to new predictive tasks (e.g., new failure types) since feature extracting steps are not required. These automated models achieve low Mean Squared Errors in predicting the generator operational state (40.85 for Speed and 0.0018 for Load) and high accuracy in diagnosing rotor demagnetization failures (99.67%) by utilizing only raw current signals. We show how to create, deploy and run the collection of proposed models in a real-time setting using a laptop connected to a test rig via a data acquisition card. Based on a sampling rate of 5 kHz, predictions are stored in an efficient time series database and monitored using a dynamic visualization framework. We further discuss existing options for understanding the decision process behind the predictions made by the models.
- 15 July 2020
- Published by Elsevier in Expert Systems with Applications
- Vol. 163, 113720
A novel dynamic asset allocation system using Feature Saliency Hidden Markov models for smart beta investing
- 15 July 2020
- Published by Elsevier in Expert Systems with Applications
- Vol. 163, 113720
A novel dynamic asset allocation system using Feature Saliency Hidden Markov models for smart beta investing
- Open Access
- 15 April 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 7 (1) , 169-186
Incorporating Data Context to Cost-Effectively Automate End-to-End Data Wrangling
The process of preparing potentially large and complex data sets for further analysis or manual examination is often called data wrangling. In classical warehousing environments, the steps in such a process are carried out using Extract-Transform-Load platforms, with significant manual involvement in specifying, configuring or tuning many of them. In typical big data applications, we need to ensure that all wrangling steps, including web extraction, selection, integration and cleaning, benefit from automation wherever possible. Towards this goal, in the paper we: (i) introduce a notion of data context, which associates portions of a target schema with extensional data of types that are commonly available; (ii) define a scalable methodology to bootstrap an end-to-end data wrangling process based on data profiling; (iii) describe how data context is used to inform automation in several steps within wrangling, specifically, matching, value format transformation, data repair, and mapping generation and selection to optimise the accuracy, consistency and relevance of the result; and (iv) we evaluate the approach with real estate data and financial data, showing substantial improvements in the results of automated wrangling.
- 15 April 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 7 (1) , 169-186
Open AccessIncorporating Data Context to Cost-Effectively Automate End-to-End Data Wrangling
The process of preparing potentially large and complex data sets for further analysis or manual examination is often called data wrangling. In classical warehousing environments, the steps in such a process are carried out using Extract-Transform-Load platforms, with significant manual involvement in specifying, configuring or tuning many of them. In typical big data applications, we need to ensure that all wrangling steps, including web extraction, selection, integration and cleaning, benefit from automation wherever possible. Towards this goal, in the paper we: (i) introduce a notion of data context, which associates portions of a target schema with extensional data of types that are commonly available; (ii) define a scalable methodology to bootstrap an end-to-end data wrangling process based on data profiling; (iii) describe how data context is used to inform automation in several steps within wrangling, specifically, matching, value format transformation, data repair, and mapping generation and selection to optimise the accuracy, consistency and relevance of the result; and (iv) we evaluate the approach with real estate data and financial data, showing substantial improvements in the results of automated wrangling.
- 1 April 2019
- Published by Elsevier in Neurocomputing
- Vol. 339, 105-115
Learning from data streams using kernel least-mean-square with multiple kernel-sizes and adaptive step-size
- 1 April 2019
- Published by Elsevier in Neurocomputing
- Vol. 339, 105-115
Learning from data streams using kernel least-mean-square with multiple kernel-sizes and adaptive step-size
- Open Access
- 12 April 2018
- Published by Springer Nature in Journal of Biomedical Semantics
- Vol. 9 (1) , 1-24
Deep learning meets ontologies: experiments to anchor the cardiovascular disease ontology in the biomedical literature
Automatic identification of term variants or acceptable alternative free-text terms for gene and protein names from the millions of biomedical publications is a challenging task. Ontologies, such as the Cardiovascular Disease Ontology (CVDO), capture domain knowledge in a computational form and can provide context for gene/protein names as written in the literature. This study investigates: 1) if word embeddings from Deep Learning algorithms can provide a list of term variants for a given gene/protein of interest; and 2) if biological knowledge from the CVDO can improve such a list without modifying the word embeddings created. We have manually annotated 105 gene/protein names from 25 PubMed titles/abstracts and mapped them to 79 unique UniProtKB entries corresponding to gene and protein classes from the CVDO. Using more than 14 M PubMed articles (titles and available abstracts), word embeddings were generated with CBOW and Skip-gram. We setup two experiments for a synonym detection task, each with four raters, and 3672 pairs of terms (target term and candidate term) from the word embeddings created. For Experiment I, the target terms for 64 UniProtKB entries were those that appear in the titles/abstracts; Experiment II involves 63 UniProtKB entries and the target terms are a combination of terms from PubMed titles/abstracts with terms (i.e. increased context) from the CVDO protein class expressions and labels. In Experiment I, Skip-gram finds term variants (full and/or partial) for 89% of the 64 UniProtKB entries, while CBOW finds term variants for 67%. In Experiment II (with the aid of the CVDO), Skip-gram finds term variants for 95% of the 63 UniProtKB entries, while CBOW finds term variants for 78%. Combining the results of both experiments, Skip-gram finds term variants for 97% of the 79 UniProtKB entries, while CBOW finds term variants for 81%. This study shows performance improvements for both CBOW and Skip-gram on a gene/protein synonym detection task by adding knowledge formalised in the CVDO and without modifying the word embeddings created. Hence, the CVDO supplies context that is effective in inducing term variability for both CBOW and Skip-gram while reducing ambiguity. Skip-gram outperforms CBOW and finds more pertinent term variants for gene/protein names annotated from the scientific literature.
- 12 April 2018
- Published by Springer Nature in Journal of Biomedical Semantics
- Vol. 9 (1) , 1-24
Open AccessDeep learning meets ontologies: experiments to anchor the cardiovascular disease ontology in the biomedical literature
Automatic identification of term variants or acceptable alternative free-text terms for gene and protein names from the millions of biomedical publications is a challenging task. Ontologies, such as the Cardiovascular Disease Ontology (CVDO), capture domain knowledge in a computational form and can provide context for gene/protein names as written in the literature. This study investigates: 1) if word embeddings from Deep Learning algorithms can provide a list of term variants for a given gene/protein of interest; and 2) if biological knowledge from the CVDO can improve such a list without modifying the word embeddings created. We have manually annotated 105 gene/protein names from 25 PubMed titles/abstracts and mapped them to 79 unique UniProtKB entries corresponding to gene and protein classes from the CVDO. Using more than 14 M PubMed articles (titles and available abstracts), word embeddings were generated with CBOW and Skip-gram. We setup two experiments for a synonym detection task, each with four raters, and 3672 pairs of terms (target term and candidate term) from the word embeddings created. For Experiment I, the target terms for 64 UniProtKB entries were those that appear in the titles/abstracts; Experiment II involves 63 UniProtKB entries and the target terms are a combination of terms from PubMed titles/abstracts with terms (i.e. increased context) from the CVDO protein class expressions and labels. In Experiment I, Skip-gram finds term variants (full and/or partial) for 89% of the 64 UniProtKB entries, while CBOW finds term variants for 67%. In Experiment II (with the aid of the CVDO), Skip-gram finds term variants for 95% of the 63 UniProtKB entries, while CBOW finds term variants for 78%. Combining the results of both experiments, Skip-gram finds term variants for 97% of the 79 UniProtKB entries, while CBOW finds term variants for 81%. This study shows performance improvements for both CBOW and Skip-gram on a gene/protein synonym detection task by adding knowledge formalised in the CVDO and without modifying the word embeddings created. Hence, the CVDO supplies context that is effective in inducing term variability for both CBOW and Skip-gram while reducing ambiguity. Skip-gram outperforms CBOW and finds more pertinent term variants for gene/protein names annotated from the scientific literature.
- Open Access
- 22 February 2018
- Published by Springer Nature in Molecular Autism
- Vol. 9 (1) , 1-13
Randomised controlled trial of simvastatin treatment for autism in young children with neurofibromatosis type 1 (SANTA)
Neurofibromatosis 1 (NF1) is a monogenic model for syndromic autism. Statins rescue the social and cognitive phenotype in animal knockout models, but translational trials with subjects > 8 years using cognition/behaviour outcomes have shown mixed results. This trial breaks new ground by studying statin effects for the first time in younger children with NF1 and co-morbid autism and by using multiparametric imaging outcomes. A single-site triple-blind RCT of simvastatin vs. placebo was done. Assessment (baseline and 12-week endpoint) included peripheral MAPK assay, awake magnetic resonance imaging spectroscopy (MRS; GABA and glutamate+glutamine (Glx)), arterial spin labelling (ASL), apparent diffusion coefficient (ADC), resting state functional MRI, and autism behavioural outcomes (Aberrant Behaviour Checklist and Clinical Global Impression). Thirty subjects had a mean age of 8.1 years (SD 1.8). Simvastatin was well tolerated. The amount of imaging data varied by test. Simvastatin treatment was associated with (i) increased frontal white matter MRS GABA (t(12) = − 2.12, p = .055), GABA/Glx ratio (t(12) = − 2.78, p = .016), and reduced grey nuclei Glx (ANCOVA p < 0.05, Mann-Whitney p < 0.01); (ii) increased ASL perfusion in ventral diencephalon (Mann-Whitney p < 0.01); and (iii) decreased ADC in cingulate gyrus (Mann-Whitney p < 0.01). Machine-learning classification of imaging outcomes achieved 79% (p < .05) accuracy differentiating groups at endpoint against chance level (64%, p = 0.25) at baseline. Three of 12 (25%) simvastatin cases compared to none in placebo met ‘clinical responder’ criteria for behavioural outcome. We show feasibility of peripheral MAPK assay and autism symptom measurement, but the study was not powered to test effectiveness. Multiparametric imaging suggests possible simvastatin effects in brain areas previously associated with NF1 pathophysiology and the social brain network. EU Clinical Trial Register (EudraCT) 2012-005742-38 ( www.clinicaltrialsregister.eu )
- 22 February 2018
- Published by Springer Nature in Molecular Autism
- Vol. 9 (1) , 1-13
Open AccessRandomised controlled trial of simvastatin treatment for autism in young children with neurofibromatosis type 1 (SANTA)
Neurofibromatosis 1 (NF1) is a monogenic model for syndromic autism. Statins rescue the social and cognitive phenotype in animal knockout models, but translational trials with subjects > 8 years using cognition/behaviour outcomes have shown mixed results. This trial breaks new ground by studying statin effects for the first time in younger children with NF1 and co-morbid autism and by using multiparametric imaging outcomes. A single-site triple-blind RCT of simvastatin vs. placebo was done. Assessment (baseline and 12-week endpoint) included peripheral MAPK assay, awake magnetic resonance imaging spectroscopy (MRS; GABA and glutamate+glutamine (Glx)), arterial spin labelling (ASL), apparent diffusion coefficient (ADC), resting state functional MRI, and autism behavioural outcomes (Aberrant Behaviour Checklist and Clinical Global Impression). Thirty subjects had a mean age of 8.1 years (SD 1.8). Simvastatin was well tolerated. The amount of imaging data varied by test. Simvastatin treatment was associated with (i) increased frontal white matter MRS GABA (t(12) = − 2.12, p = .055), GABA/Glx ratio (t(12) = − 2.78, p = .016), and reduced grey nuclei Glx (ANCOVA p < 0.05, Mann-Whitney p < 0.01); (ii) increased ASL perfusion in ventral diencephalon (Mann-Whitney p < 0.01); and (iii) decreased ADC in cingulate gyrus (Mann-Whitney p < 0.01). Machine-learning classification of imaging outcomes achieved 79% (p < .05) accuracy differentiating groups at endpoint against chance level (64%, p = 0.25) at baseline. Three of 12 (25%) simvastatin cases compared to none in placebo met ‘clinical responder’ criteria for behavioural outcome. We show feasibility of peripheral MAPK assay and autism symptom measurement, but the study was not powered to test effectiveness. Multiparametric imaging suggests possible simvastatin effects in brain areas previously associated with NF1 pathophysiology and the social brain network. EU Clinical Trial Register (EudraCT) 2012-005742-38 ( www.clinicaltrialsregister.eu )
- 1 November 2017
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 75, S28-S33
Learning to identify Protected Health Information by integrating knowledge- and data-driven algorithms: A case study on psychiatric evaluation notes
- 1 November 2017
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 75, S28-S33
Learning to identify Protected Health Information by integrating knowledge- and data-driven algorithms: A case study on psychiatric evaluation notes