A Polynomial Time Algorithm for Generating Neural Networks for Pattern Classification: Its Stability Properties and Some Test Results

Abstract

Polynomial time training and network design are two major issues for the neural network community. A new algorithm has been developed that can learn in polynomial time and also design an appropriate network. The algorithm is for classification problems and uses linear programing models to design and train the network. This paper summarizes the new algorithm, proves its stability properties, and provides some computational results to demonstrate its potential.