Automatic Creation of Functional Sub-Networks Using Genetic Algorithms

Research has shown a strong correlation between the topology and functional capability of neural networks, though difficulties encountered in traditional neural network design and training increase in relation to the size and complexity of the network. Constructing a neural network topology through the integration of modularized, functional sub-networks has been shown to provide a reduction in overall topological complexity and computational requirements. Furthermore, evolutionary computing based optimization techniques may be used to overcome traditional design difficulties. The research presented in this paper outlines a modular neural network approach for the approximation of a Mealy machine example of a sequential logic circuit. The use of a genetic algorithm for the automatic generation of an optimal set of trained functional sub-networks is described. Results concur with the use of evolutionary computing techniques as a method for overcoming traditional neural network design issues and the use of modularization for the reduction of task complexity.