Benjamin Fowler | Memorial University of Newfoundland (original) (raw)

Papers by Benjamin Fowler

2015 Brazilian Conference on Intelligent Systems (BRACIS), 2015

We develop a tree-based genetic programming system capable of modelling evolvability during evolu... more We develop a tree-based genetic programming system capable of modelling evolvability during evolution through machine learning algorithms, and exploiting those models to increase the efficiency and final fitness. Existing methods of determining evolvability require too much computational time to be effective in any practical sense. By being able to model evolvability instead, computational time may be reduced. This will be done first by demonstrating the effectiveness of modelling these properties a priori, before expanding the system to show its effectiveness as evolution occurs.

The visual analysis of large movement data sets can be a challenging task. This study proposes an... more The visual analysis of large movement data sets can be a challenging task. This study proposes an approach for identifying interesting movement patterns that combines human knowledge and decision making with a hybrid clustering-classification method. Rather than performing an unsupervised clustering of the entire data set, a stratified random sample of the full data set is used to identify initial clusters that are verified and labelled by the analyst, and then used as input patterns for classifying the remainder of the data set using an iterative genetic program. Classifications suggested after each iteration are presented to the analyst for refinement based on their knowledge and experience. A geovisual analytics environment is provided to both show the outcomes of the clustering and classification, and to obtain the analyst’s input during the hybrid clustering-classification process. Our approach allows data to be classified without a priori specification of classification patterns. Instead, the process takes advantage of human decision making within the automatic analysis of the data. The approach was tested with fishing vessel movement data in Eastern Canada.

Machine lifelong learning (ML3) is concerned with machines capable of learning and retaining know... more Machine lifelong learning (ML3) is concerned with machines capable of learning and retaining knowledge over time, and exploiting previous knowledge to assist new learning. An ML3 system requires effective task retention, and effective consolidation of new tasks. This thesis presents an ML3 system using a context-sensitive multiple task learning (csMTL) neural network that functions as a consolidated domain knowledge store. csMTL was developed in response to structural limitations of multiple task learning (MTL) for ML3. Instead of additional outputs for each task csMTL uses additional context inputs that indicate the associated task. The csMTL-based system is analyzed empirically using synthetic and real domains. The experiments focus on the effective retention of knowledge and the effective consolidation of new knowledge, using independent test set accuracy as a measure of effectiveness. The studies indicate that the methodology results in effective task retention when appropriate learning parameters are used. New task consolidation efficacy suffered using the same learning parameters. Experimentation also suggests that virtual instances (input-output pairs constructed from the consolidated domain knowledge) that correspond to real instances improves efficacy of retention and new task consolidation. Experimentation also indicates that representational transfer allows more effective retention, but at the cost of less effective new task consolidation.

We present context-sensitive Multiple Task Learning, or csMTL as a method of inductive transfer e... more We present context-sensitive Multiple Task Learning, or csMTL as a method of inductive transfer embedded in the well known WEKA machine learning suite. csMTL uses a single output neural network and additional contextual inputs for learning multiple tasks. Inductive transfer occurs from secondary tasks to the model for the primary task so as to improve its predictive performance. The WEKA multi-layer perceptron algorithm is modified to accept csMTL encoded multiple tasks examples. Testing on three domains of tasks demonstrates that this WEKA-based version of csMTL provides modest but beneficial performance increases. Our on-going objective is to increase the availability of transfer learning systems to students, researchers and practitioners.

2015 Brazilian Conference on Intelligent Systems (BRACIS), 2015

We develop a tree-based genetic programming system capable of modelling evolvability during evolu... more We develop a tree-based genetic programming system capable of modelling evolvability during evolution through machine learning algorithms, and exploiting those models to increase the efficiency and final fitness. Existing methods of determining evolvability require too much computational time to be effective in any practical sense. By being able to model evolvability instead, computational time may be reduced. This will be done first by demonstrating the effectiveness of modelling these properties a priori, before expanding the system to show its effectiveness as evolution occurs.

The visual analysis of large movement data sets can be a challenging task. This study proposes an... more The visual analysis of large movement data sets can be a challenging task. This study proposes an approach for identifying interesting movement patterns that combines human knowledge and decision making with a hybrid clustering-classification method. Rather than performing an unsupervised clustering of the entire data set, a stratified random sample of the full data set is used to identify initial clusters that are verified and labelled by the analyst, and then used as input patterns for classifying the remainder of the data set using an iterative genetic program. Classifications suggested after each iteration are presented to the analyst for refinement based on their knowledge and experience. A geovisual analytics environment is provided to both show the outcomes of the clustering and classification, and to obtain the analyst’s input during the hybrid clustering-classification process. Our approach allows data to be classified without a priori specification of classification patterns. Instead, the process takes advantage of human decision making within the automatic analysis of the data. The approach was tested with fishing vessel movement data in Eastern Canada.

Machine lifelong learning (ML3) is concerned with machines capable of learning and retaining know... more Machine lifelong learning (ML3) is concerned with machines capable of learning and retaining knowledge over time, and exploiting previous knowledge to assist new learning. An ML3 system requires effective task retention, and effective consolidation of new tasks. This thesis presents an ML3 system using a context-sensitive multiple task learning (csMTL) neural network that functions as a consolidated domain knowledge store. csMTL was developed in response to structural limitations of multiple task learning (MTL) for ML3. Instead of additional outputs for each task csMTL uses additional context inputs that indicate the associated task. The csMTL-based system is analyzed empirically using synthetic and real domains. The experiments focus on the effective retention of knowledge and the effective consolidation of new knowledge, using independent test set accuracy as a measure of effectiveness. The studies indicate that the methodology results in effective task retention when appropriate learning parameters are used. New task consolidation efficacy suffered using the same learning parameters. Experimentation also suggests that virtual instances (input-output pairs constructed from the consolidated domain knowledge) that correspond to real instances improves efficacy of retention and new task consolidation. Experimentation also indicates that representational transfer allows more effective retention, but at the cost of less effective new task consolidation.

We present context-sensitive Multiple Task Learning, or csMTL as a method of inductive transfer e... more We present context-sensitive Multiple Task Learning, or csMTL as a method of inductive transfer embedded in the well known WEKA machine learning suite. csMTL uses a single output neural network and additional contextual inputs for learning multiple tasks. Inductive transfer occurs from secondary tasks to the model for the primary task so as to improve its predictive performance. The WEKA multi-layer perceptron algorithm is modified to accept csMTL encoded multiple tasks examples. Testing on three domains of tasks demonstrates that this WEKA-based version of csMTL provides modest but beneficial performance increases. Our on-going objective is to increase the availability of transfer learning systems to students, researchers and practitioners.