Determination and adaptive alteration of artificial neural network structures by a genetic algorithm with a controlled genotype-phenotype mapping

A method is proposed for determination and adaptive alteration of artificial neural network (ANN) structures. Not only the weights but also the structure is altered adaptively. From the engineering viewpoint, such an adaptation will be beneficial, for example, for accommodation of faults which require different structures of the ANN. The salient point of the proposed method is that it can alter the ANN structure by adjusting a single parameter; it is not necessary to consider which particular nodes are to be removed or where and how many nodes are to be added. The method is based on genetic algorithms (GA) which emulate evolution. A fitness function and a coding system of ANN structures on chromosomes are also proposed which are appropriate for optimization of the ANN structures.

[1]  Peter M. Todd,et al.  Designing Neural Networks using Genetic Algorithms , 1989, ICGA.

[2]  Scott E. Fahlman,et al.  The Recurrent Cascade-Correlation Architecture , 1990, NIPS.

[3]  Kotaro Hirasawa,et al.  Neural Network Structure Design Using Genetic Algorithm , 1998 .

[4]  Russell Reed,et al.  Pruning algorithms-a survey , 1993, IEEE Trans. Neural Networks.

[5]  D. B. Fogel,et al.  AN INFORMATION CRITERION FOR OPTIMAL NEURAL NETWORK SELECTION , 1990, 1990 Conference Record Twenty-Fourth Asilomar Conference on Signals, Systems and Computers, 1990..

[6]  Xin Yao,et al.  The Evolution of Connectionist Networks , 1994 .

[7]  Peter J. Angeline,et al.  An evolutionary algorithm that constructs recurrent neural networks , 1994, IEEE Trans. Neural Networks.