A New Algorithm for Training RBF Networks

A. Klautau

DOI: 10.14209/sbrt.2003.681

Evento: XX Simpósio Brasileiro de Telecomunicações (SBrT2003)

Keywords: Pattern recognition radial basis function discriminative training

Abstract

The radial basis function (RBF) network is the main practical alternative to the multi-layer perceptron for non-linear modeling. This work presents a new discriminative algorithm for the first training-stage of classifiers consisting of RBF networks with diagonal covariance Gaussians as basis functions. The experimental results show that the algorithm leads to improved performance when compared to the conventional expectation-maximization algorithm.

Download