Abstract

Recent interest in neural network models for parallel distributed processing is stimulating much interest within the optics community, in particular, on the feasibility of optical associative memories.¹ We illustrate how a set of spatially multiplexed volume holograms recorded in a photorefractive crystal constitutes an optical crossbar switch that globally interconnects a two-layer neural network.² We focus on the use of the network to provide associative memory and show that the mathematical operations involved in distributed memory storage and recall can be optically implemented in a very natural fashion within the framework of photorefractive holography. These operations include vector outer products, linear combinations of outer-product matrices, and matrix-vector multiplication.

© 1988 Optical Society of America