A method for designing and training neural networks using genetic al-gorithms is proposed,  with the aim of getting the optimal structure of the network and the optimized parameter set  simultaneously. For this purpose, a fitness function depending on both the output errors and  sim-pleness in the structure of the network is introduced. The validity of this method is checked  by experiments on four logical operation prob-lems: XOR, 6XOR, 4XOR-2AND, and 2XOR 2AND-2OR; and on two other problems: 4-bit pattern copying and an 8 8-encoder/decoder. It is  concluded that, although this method is less powerful for disconnected networks, it is useful for  connected ones.

Published in

Complex Systems