Comparative Aspects of Associative Memory Networks for Modelling

This paper will describe a class of networks called Associative Memory Networks which have many desirable properties for applications within the field of Intelligent Control. This class is defined to include the Albus CMAC neural network, the B-spline neural network and a certain class of Fuzzy Logic networks. These networks will first be described within a common framework which has a natural parallel implementation and then several learning rules will be derived. These are instantaneous gradient descent and error correction adaptive strategies and the sparse internal representation of the networks make them particularly suited to these learning rules. Finally all three networks will be applied to the same nonlinear time series prediction problem, comparing the strengths and weaknesses of each network