38 - Popfnns: Fuzzy Neural Techniques for Rule-Based Identification in Expert Systems

Abstract

Fuzzy neural networks are hybrid intelligent systems that are constructed using both neural network and fuzzy logic techniques. They aim to emulate the human thinking process and can efficaciously serve as an extension of the creative and problem-solving abilities. They have the advantages of both neural networks and fuzzy systems, and alleviate the shortcomings of the respective techniques. Fuzzy neural networks offer a rich environment for producing intelligent applications that can enormously increase productivity and act as intelligent assistants. This chapter focuses on the design and implementation of fuzzy neural networks that use neural network techniques to realize fuzzy rule-based systems under different fuzzy inference models. It presents three novel fuzzy neural network structures and two rule identification algorithms. The fuzzy neural network structures include the pseudo outer-product-based fuzzy neural network using the truth-value restriction method; the pseudo outer-product-based fuzzy neural network using the approximate analogical reasoning schema together with the singleton fuzzifier; and the pseudo outer-product-based fuzzy neural network using the approximate analogical reasoning schema together with the nonsingleton fuzzifier. The two rule identification algorithms are the pseudo outer-product and the Lazy pseudo outer-product learning algorithms. These fuzzy neural networks have been applied efficiently in novel applications for pattern classifications of signatures and fingerprints and, they produce potential results.