COMPARISON OF LEARNING AND GENERALIZATION CAPABILITIES OF THE KAK ANDTHE BACKPROPAGATION ALGORITHMS

This paper investigates the performance of the Kak algorithms for trai ning of feedforward neural networks. It is found that the Backpropagat ion algorithm is much more computation-intensive than the Kak algorith ms that have comparable generalization performance. Results of compute r simulations with standard benchmark problems such as XOR, encoder/de coder, classification, and time series prediction are presented.