ABSTRACT
This chapter outlines a simple pattern recognition algorithm which is based on neural networks. The algorithm is self-configuring and utilizes the simple perceptron learning rule. This represents a pleasant alternative to various other approaches which are founded on complicated training algorithms and transfer functions. Evolutionary principles guide the selection of appropriate training sets during network construction. Genetic Algorithms are in general simple adaptive search algorithms, which create new solutions to a given problem by exploiting past performance of older solutions in a manner similar to evolutionary processes as found in nature. The chapter also outlines the Evolutionary Growth Perceptron (EGP) algorithm and presents some empirical findings regarding the choice of crossover operator and its significance. The basic genetic operation made use of by the EGP algorithm can be summarized as follows: Reproduction, Crossover and Mutation. Three approaches to constructing crossover operators for use by EGP. Those are Simple crossover, weighted crossover and Blocked crossover.
