Brunel University Research Archive (BURA) >
College of Engineering, Design and Physical Sciences >
Dept of Electronic and Computer Engineering >
Dept of Electronic and Computer Engineering Research Papers >

Please use this identifier to cite or link to this item: http://bura.brunel.ac.uk/handle/2438/1499

Title: Percentage-based hybrid pattern training with neural network specific crossover
Authors: Guan, SU
Ramanathan, K
Keywords: Neural network, Genetic algorithm, Initial weight, Training pattern, Hybrid training
Publication Date: 2007
Publisher: Freund & Pettman
Citation: Journal of Intelligent Systems, 16(1): 1 -26
Abstract: In this paper, a new weight-setting method is proposed to improve the training time and generalization accuracy of feed-forward neural networks. This method introduces a percentage-based hybrid Pattern training (PHP) scheme and aims to provide a solution to the problem dependency of other Genetic Algorithm based Neural Network weight-setting methods. A neural network is trained using neural network specific GA until a certain percentage of the training patterns is learnt. The weights thus obtained are used as the initial weights for backpropagation training, which is then applied to complete the network training. Further improvement to the method was looked into and the use of distributed GA in the weight-setting phase was investigated. The final approach derived was tested on four neural network problems and it was observed that as the number of patterns trained using GA approaches 50% of the total number of training patterns, the proposed method is more effective in pulling the networks out of local minima. Also, the networks trained using this method showed as much as 75% improvement in training time and 15% improvement in generalization accuracy.
URI: http://bura.brunel.ac.uk/handle/2438/1499
ISSN: 0334-1860
Appears in Collections:Electronic and Computer Engineering
Dept of Electronic and Computer Engineering Research Papers

Files in This Item:

File Description SizeFormat
Closed Access - Guan et al - percentage based - Journal of Intelligent Systems 2007.txt494 BTextView/Open

Items in BURA are protected by copyright, with all rights reserved, unless otherwise indicated.