WAV-a weight adaptation algorithm for normalized radial basis function networks

A weight adaptation algorithm for normalized radial basis function networks is discussed. This simple algorithm-WAV-is based on the idea of Weighted AVeraging, and is inspired by the functional equivalence of normalized radial basis function networks and probabilistic fuzzy inference systems. We compare WAV with other weight adaptation algorithms including least mean square algorithm (LMS) and recursive least square algorithm (RLS) for tuning the weights of a normalized radial basis function network in the time series prediction problem with Box and Jenkins' data. Simulation shows that WAV achieves close to RLS mean square error with computation complexity close to LMS. Further study is suggested.