Radial Basis Function Networks and Statistics
David G. Lowe · 2000
Abstract The class of neural network architectures known as radial basis functions has a structure and history which has produced strong links with various areas in the statistical sciences. This chapter reviews some of the more salient aspects, which include the links with conditional and unconditional density estimation (such as Gaussian mixture models), regression problems and links with kernel estimators and linear smooths, supervised feature extraction and discriminant analysis, and topographic feature extraction and links to multidimensional scaling, Sammon mappings and nonlinear principal component analysis.