Incorporating Fuzzy Logic in an Artificial Neural Network

Abstract

A clustering neural network that uses unsupervised learning to generate clusters is described. A fuzzy logic classifier is used in conjunction with the network to assign membership functions to patterns that describes the degree of membership that the pattern has for a cluster. The input that is fed into the neural network are nonfuzzy (crisp) values. A two-stage clustering neural network that uses unsupervised learning in the first stage and supervised learning in the second stage is to be revised so that the supervised function is disabled. Therefore, the resulting clusters will be generated from the unsupervised stage of the original neural network. The fuzzy logic classifier is developed by implementing a similar algorithm used by fuzzy c-means for generating fuzzy membership functions. By assigning a membership function to the data points, the patterns are allowed to hold membership to more than one cluster. This is in contrast to the crisp membership that is held by the patterns in the original neural network. Data that is ambiguous can be represented and processed using an algorithm such as this. The neural network is written in FORTRAN and the fuzzy logic classifier is written in MATLAB.