FEN (fuzzy expert network) learning architecture; generalization of self adjusting fuzzy modeling on event-driven acyclic neural networks using expert networks backpropagation learning

FEN (fuzzy expert network) is a new network architecture of neural objects for fuzzy modeling. The neural objects process information through node functions that are different from a typical sigmoidal node processor for an analog perceptron. By connecting a few types of node processors on an event driven acyclic (feedforward) neural network, FEN represents the fuzzy modeling with self adjustment. Weights on this network imply fuzzy parameters to be adjusted with no restriction of layered topology by learning. FEN offers automated tuning from input-output data for membership functions on which the performance of fuzzy modeling depends. And especially using the enhanced idea of a dynamic backward error assignment for learning, FEN is effective for tuning parameters for nonsmooth membership functions, for example, symmetric triangular functions of an antecedent part. Results of testing FEN are presented to demonstrate learning performance and adaptability.