Solving fuzzy relational equations by max-min neural networks

Abstract

The problem of identifying a fuzzy system has been faced from several points of view which include statistical methods, neural networks and relational equation-solving approaches. In this paper, we present the use of a neural network without any activation function in order to identify a fuzzy system through the solution of a fuzzy relational equation from a set of examples. The main contribution of this work is to define a "smooth derivative" to be used in the minimization of the energy function which drives the learning procedure. Some examples show the effectiveness of this new approach.<>