Abstract

In a system with many variables, finding their optimum values is, in general, a computationally intensive process that scales badly with the number of variables. The potential benefit of accelerating the solution of such problems with optical computing techniques is therefore large. Considerable success has been reported [1] with the use of Hopfield neural networks for optimisation and neural network architectures lend themselves to parallel operation in hardware. Optics is appropriate for implementing the large number of connections required, which it can provide very efficiently when there is any regularity in the interconnect pattern.

© 1997 Optical Society of America