Radial basis function networks with FIR/IIR synapses

摘要

A learning algorithm for a special class of Radial Basis Functions (RBF) networks is proposed. The novelty aspect of the architecture consists in replacing the usual scalar values of the output weights with linear FIR or IIR filters transfer functions. Simulation results for approximating a lowpass Butterworth filter are presented and other potential temporal processing applications are outlined.