Piecewise Linear Neural Networks.

摘要

Modern weapons systems often require pattern recognition devices and algorithms. The three-layer neural network--consisting of input layer, hidden layer, and output layer--has great potential as a pattern classifier. Determining the size of the hidden layer and computing the weights are two of the primary design as a function of the input and output dimensions and of the number of pattern prototypes. The neurons are assumed to have piecewise linear transfer functions.