Abstract

Artificial neural networks have been successfully applied to a wide variety of practical tasks ranging from the recognition of speech and handwriting to the prediction of future stock prices. Highly refined learning procedures allow these multilayer, nonlinear networks to construct internal representations from examples, thus removing the need for explicit programming. The power of neural networks (NN) derives from their ability to learn nonlinear relationships and to generalize from specific training situations. They are advantageous where a small number of decisions have to be made from a large amount of data, and where nonlinear or very complex mappings must be acquired.

© 1996 Optical Society of America