Abstract

Research interest in neural networks has grown recently because of their capabilities of realizing associative content-addressable memories, pattern classifiers, solving optimization problems, and adaptive learning. The characteristics of neural networks such as fault tolerance, massive interconnects. and parallel processing make them attractive candidates for optical implementation. Many implementations of neural networks using optical linear algebra processors with nonlinearity and feedback have been proposed.¹ We extend previous work on noise problems in optical matrix-vector multipliers² and discuss the effects of component and system noise on the performance of both first-and second-order Hopfield-type neural networks.

© 1988 Optical Society of America