Implementation of fuzzy systems using multilayered neural network

Abstract

A synthesis method of a multilayered neural network (NN) for fuzzy systems is presented. Back propagation (BP) makes a multilayered NN an effective implementation technique for a fuzzy system with its adapative capability. The authors' method consists of two stages. First, an initial NN is constructed by a network builder that implements qualitative knowledge about the problem and then the initial NN is trained by BP, using the training data to improve accuracy. Synthesis equations are given for the network builder by generalizing logical functions. It is shown how these synthesis equations can be used to construct the initial NN in function approximation and character recognition problems. >