Evolution of a dynamical modular neural network and its application to associative memories

This paper presents an evolutionary approach to architecture design of dynamical modular neural networks. As one of the modular neural networks, we adopt Cross-Coupled Hopfield Nets (CCHN) in which Hopfield networks are coupled to each other. The architecture of CCHN is represented by some structural parameters such as the number of modules, the numbers of module units, module connectivity, and so forth. In this paper, these structural parameters are treated as the pheno-type of an individual, and a suitable modular architecture is searched by using genetic algorithms. To verify the usefulness of the proposed architecture design algorithm we apply CCHN to associative memories.