Elliptical basis function networks for classification tasks

Abstract

In this paper we compare variants of elliptical basis function networks for classification tasks. The networks are introduced as density estimators and then modified towards RBF networks. Node reduction is accomplished by a genetic algorithm(GA). Two different kinds of node connections are compared. As a second degree of freedom different types of basis functions are investigated. On an artificial test set of the time series domain the impact of dimensionality is considered.