Fuzzy Neural Models Based on Some New Fuzzy* Arithmetic Operations
Petr Musilekl and Madan M. Gupta
Intelligent Systems Research Laboratory College of Engineering, University of Saskatchewan, Saskatoon, Saskatchewan, Canada S7N 5A9
This paper introduces a novel approach to fuzzy arithmetic computation in fuzzy neural networks. The first part provides an overview of the standard fuzzy arithmetic operations and limitations of their use in fuzzy arithmetic based neural models. Consequently, alternative fuzzy arithmetic operations are developed and their aspects for the neural models are discussed in more detail. Originality of our approach lies in the treatment of neural inputs and weights as interactive variables which allows control of uncertainty growth in neural processing. Besides the detailed theoretical description of these operations, corresponding implementation algorithms are given as well. Combination of the alternative fuzzy arithmetic operations is briefly shown on two particular fuzzy arithmetic neurons providing fuzzy extensions of common crisp neural models. Finally, an example of a simple fuzzy neural structure for pattern classification is given.