NEURAL NETWORKS - A REVIEW FROM A STATISTICAL PERSPECTIVE

This paper informs a statistical readership about Artificial Neural Ne tworks (ANNs), points out some of the links with statistical methodolo gy and encourages cross-disciplinary research in the directions most l ikely to bear fruit. The areas of statistical interest are briefly out lined, and a series of examples indicates the flavor of ANN models. We then treat various topics in more depth. In each case, we describe th e neural network architectures and training rules and provide a statis tical commentary. The topics treated in this way are perceptrons (from single-unit to multilayer versions), Hopfield-type recurrent networks (including probabilistic versions strongly related to statistical phy sics and Gibbs distributions) and associative memory networks trained by so-called unsupervised learning rules. Perceptrons are shown to hav e strong associations with discriminant analysis and regression, and u nsupervized networks with cluster analysis. The paper concludes with s ome thoughts on the future of the interface between neural networks an d statistics.