A SEQUENTIAL LEARNING SCHEME FOR FUNCTION APPROXIMATION USING MINIMALRADIAL BASIS FUNCTION NEURAL NETWORKS

This article presents a sequential learning algorithm for function app roximation and time-series prediction using a minimal radial basis fun ction neural network (RBFNN). The algorithm combines the growth criter ion of the resource-allocating network (RAN) of Platt (1991) with a pr uning strategy based on the relative contribution of each hidden unit to the overall network output. The resulting network leads toward a mi nimal topology for the RBFNN. The performance of the algorithm is comp ared with RAN and the enhanced RAN algorithm of Kadirkamanathan and Ni ranjan (1993) for the following benchmark problems: (1) hearta from th e benchmark problems database PROBEN1, (2) Hermite polynomial, and (3) Mackey-Glass chaotic time series. For these problems, the proposed al gorithm is shown to realize RBFNNs with far fewer hidden neurons with better or same accuracy.