Design of a fuzzy logic-based bidirectional mapping for Kohonen networks

Abstract

In control engineering identifying the model of a system as well as inverting this model is an important task. Neural networks have recently shown great success in identifying a model and fuzzy logic allows in some cases to invert the model and to describe it in a way that human beings can understand it. An algorithm is described that combines the advantages of both techniques. It is a self-organising locally interpolating map (SOLIM) that allows a bidirectional continuous mapping with extrapolation, the use of a-priori knowledge and a good visualisation of the data.