Problem Decomposition and Adaptation in Cooperative Neuro-Evolution

<p>One way to train neural networks is to use evolutionary algorithms such as cooperative coevolution - a method that decomposes the network's learnable parameters into subsets, called subcomponents. Cooperative coevolution gains advantage over other methods by evolving particular subcomponents independently from the rest of the network. Its success depends strongly on how the problem decomposition is carried out. This thesis suggests new forms of problem decomposition, based on a novel and intuitive choice of modularity, and examines in detail at what stage and to what extent the different decomposition methods should be used. The new methods are evaluated by training feedforward networks to solve pattern classification tasks, and by training recurrent networks to solve grammatical inference problems. Efficient problem decomposition methods group interacting variables into the same subcomponents. We examine the methods from the literature and provide an analysis of the nature of the neural network optimization problem in terms of interacting variables. We then present a novel problem decomposition method that groups interacting variables and that can be generalized to neural networks with more than a single hidden layer. We then incorporate local search into cooperative neuro-evolution. We present a memetic cooperative coevolution method that takes into account the cost of employing local search across several sub-populations. The optimisation process changes during evolution in terms of diversity and interacting variables. To address this, we examine the adaptation of the problem decomposition method during the evolutionary process. The results in this thesis show that the proposed methods improve performance in terms of optimization time, scalability and robustness. As a further test, we apply the problem decomposition and adaptive cooperative coevolution methods for training recurrent neural networks on chaotic time series problems. The proposed methods show better performance in terms of accuracy and robustness.</p>

Problem Decomposition and Adaptation in Cooperative Neuro-Evolution

<p>One way to train neural networks is to use evolutionary algorithms such as cooperative coevolution - a method that decomposes the network's learnable parameters into subsets, called subcomponents. Cooperative coevolution gains advantage over other methods by evolving particular subcomponents independently from the rest of the network. Its success depends strongly on how the problem decomposition is carried out. This thesis suggests new forms of problem decomposition, based on a novel and intuitive choice of modularity, and examines in detail at what stage and to what extent the different decomposition methods should be used. The new methods are evaluated by training feedforward networks to solve pattern classification tasks, and by training recurrent networks to solve grammatical inference problems. Efficient problem decomposition methods group interacting variables into the same subcomponents. We examine the methods from the literature and provide an analysis of the nature of the neural network optimization problem in terms of interacting variables. We then present a novel problem decomposition method that groups interacting variables and that can be generalized to neural networks with more than a single hidden layer. We then incorporate local search into cooperative neuro-evolution. We present a memetic cooperative coevolution method that takes into account the cost of employing local search across several sub-populations. The optimisation process changes during evolution in terms of diversity and interacting variables. To address this, we examine the adaptation of the problem decomposition method during the evolutionary process. The results in this thesis show that the proposed methods improve performance in terms of optimization time, scalability and robustness. As a further test, we apply the problem decomposition and adaptive cooperative coevolution methods for training recurrent neural networks on chaotic time series problems. The proposed methods show better performance in terms of accuracy and robustness.</p>

Split-Based Algorithm for Weighted Context-Free Grammar Induction

The split-based method in a weighted context-free grammar (WCFG) induction was formalised and verified on a comprehensive set of context-free languages. WCFG is learned using a novel grammatical inference method. The proposed method learns WCFG from both positive and negative samples, whereas the weights of rules are estimated using a novel Inside–Outside Contrastive Estimation algorithm. The results showed that our approach outperforms in terms of F1 scores of other state-of-the-art methods.

Parsing Expression Grammars and Their Induction Algorithm

Grammatical inference (GI), i.e., the task of finding a rule that lies behind given words, can be used in the analyses of amyloidogenic sequence fragments, which are essential in studies of neurodegenerative diseases. In this paper, we developed a new method that generates non-circular parsing expression grammars (PEGs) and compares it with other GI algorithms on the sequences from a real dataset. The main contribution of this paper is a genetic programming-based algorithm for the induction of parsing expression grammars from a finite sample. The induction method has been tested on a real bioinformatics dataset and its classification performance has been compared to the achievements of existing grammatical inference methods. The evaluation of the generated PEG on an amyloidogenic dataset revealed its accuracy when predicting amyloid segments. We show that the new grammatical inference algorithm achieves the best ACC (Accuracy), AUC (Area under ROC curve), and MCC (Mathew’s correlation coefficient) scores in comparison to five other automata or grammar learning methods.