Function approximation by localized basis function neural network

摘要

A transition from the localized basis function description of a probability desity function to the general regression estimator and corresponding neural network model is presented in this article, and its relations to localized basis function neural network are explained. Typical parametric and nonparametric models are described more in detail. Among them, a new elliptical multivariate basis function approach is the most advanced one. In the article, its performance is compared with the radial basis function neural network using various two-dimensional examples.