Chapter 4 - Diffuse Neural and Neuro-Fuzzy Networks Training Algorithms

Abstract

This chapter deals with the problem of multilayer neural and neuro-fuzzy networks training with simultaneous estimation of the hidden and output layers parameters. The hidden layer parameters probable values and their possible deviations are assumed to be known. A priori information about the output layer weights is absent and in one initialization of the Gauss–Newton method (the GN method) they are assumed to be random variables with zero expectations and a covariance matrix proportional to a large parameter and in the other option they are either unknown constants or random variables with unknown statistical characteristics. Training algorithms based on the GN method with linearization about the latest estimate are proposed and studied. The theoretical results are illustrated with the examples of pattern recognition and identification of nonlinear static and dynamic plants.