Abstract

In recent years, much work has been going on in the optical implementation of artificial neural network systems. The parallel and crosstalk free interconnection characteristics of optical systems are well suited to exploit fully the desired parallel characteristics of artificial neural networks. The application of these systems as associative memories has been explored in many cases.^1-4 To facilitate optical implementation of these neural systems various modifications to the original Hopfield model have been proposed.⁵ It has been shown both theoretically and experimentally, that the storage and recall capacity of neural systems based on the Hopfield model are not significantly reduced when only the inhibitory (negative) interconnections are used.^6,7

© 1991 Optical Society of America