Using Taguchi methods to train artificial neural networks in pattern recognition, control and evolutionary applications

Abstract

Taguchi methods are commonly used to optimise industrial systems, particularly in manufacturing. We have shown that they may also be used to optimise neural network weights and therefore train the network. This paper builds on previous work and explains the application of the method to network training in several important areas, including pattern recognition, neurocontrol, evolutionary or genetic networks and nonlinear neurons. Consideration is also given to the training of networks for failure and fault control systems.