Evolutionary Operations of Interneuron Synaptic Structure for Feed-Forward Multilayer Networks

Abstract

This paper mainly studies evolutionary operations of interneuron synaptic structure search based on the developed structure synthesis algorithm for feed-forward multilayer network trained with a teacher. The decomposition of neural network learning tasks into a series of subtasks is proposed. During evolutionary search the neural network is not converted to a genotype and all the transformations of synaptic structures are performed directly. This differs from a classical approach to genetic algorithms involving coding phenotype of individuals (structure) to genotype (code string).