Wavelet networks for functional learning

Abstract

A wavelet-based neural network is described. The network is similar to the radial basis function (RBF) network, except that the RBF's are replaced by orthonormal scaling functions. It has been shown that the wavelet network has universal and L/sup 2/ approximation properties and is a consistent function estimator. Convergence rates, which avoid the "curse of dimensionality," are obtained for certain function classes. The network also compared favorably to the MLP and RBF networks in the experiments.