A small selection of neural network methods and their statistical connections

Abstract

A tutorial introduction is given to a limited selection. of artificial neural network models. Particular emphasis is laid on multi-layer perceptrons and simple Hopfield associative memories. In the context of perceptrons, two case studies are described, concerning Zip code detection and coin recognition. A simple experiment is reported with a Hopfield net. Some other approaches and applications are briefly listed, and bibliographical remarks are made. Throughout, points of contact with mainstream statistical methodology are highlighted.