SOME NEURAL-NET REALIZATIONS OF FUZZY-REASONING

In this paper we analyze the neural network implementation of fuzzy lo gic proposed by Keller et al. [Fuzzy Sets Syst., 45, 1-12 (1992)], der ive a learning algorithm for obtaining an optimal alpha for the net, a nd, for a special case, we show how one can directly (avoiding trainin g) compute the optimal alpha. We address how training data can be gene rated for such a system. Effectiveness of the optimal alpha is then es tablished through numerical examples. In this regard, several indices for performance evaluation are discussed. Finally, we propose a new ar chitecture and demonstrate its effectiveness with numerical examples. (C) 1998 John Wiley & Sons, Inc.