Abstract

The effect of discretizing interconnection weight strengths in an optoelectronic learning neural network based on the backpropagation algorithm is investigated. We discuss how discretization arises in such an implementation. Using computer simulations we find that learning performance, as tested on the two-input XOR problem, is poor but that the addition of noise to the system results in substantial improvement.

© 1989 Optical Society of America

Character recognition using a dynamic optoelectronic neural network with unipolar binary weights

Masaya Oita, Masanobu Takahashi, Shuichi Tai, and Kazuo Kyuma
Opt. Lett. 15(21) 1227-1229 (1990)

Weight and neuron output resolution requirements in optoelectronic neural networks

T. J. Allen, K. M. Curtis, and J. W. Orton
Appl. Opt. 34(20) 4136-4139 (1995)

Gain and exposure scheduling to compensate for photorefractive neural-network weight decay

Adam A. Goldstein, Gregory C. Petrisor, and B. Keith Jenkins
Opt. Lett. 20(6) 611-613 (1995)

Architectures for optoelectronic analogs of self-organizing neural networks

Nabil H. Farhat
Opt. Lett. 12(6) 448-450 (1987)

Experimental learning in an optical perceptronlike neural network

Hisao Yoshinaga, Ken-ichi Kitayama, and Terushige Hori
Opt. Lett. 14(14) 716-718 (1989)