Abstract

A neurooptic processor (NOP) is the optical analog of a recurrently interconnected neural network. The basic NOP architecture consists of an optical cavity defined at one end by a fully adaptive reflective optical interconnect matrix (ROIM) and at the other end by an amplifying phase conjugate mirror (APCM). Elements of the ROIM function as optical synapses and the APCM functions as an array of optical neurons. Information is fed into and extracted from a NOP via an intracavity beam splitter. Functional behavior of a NOP is governed by the way in which the APCM is pumped, by the structure of the ROIM, and by the manner in which outside stimuli are fed into the NOP. Intelligent functions which may be realized by appropriately configured NOPs include feature extraction,¹ associative recall,¹ novelty filtering,¹ inference,¹ generalization,¹ disambiguation,¹ and problem optimization.² In this paper we discuss solutions of the canonical equations of motion which govern the modal evolution of various NOP architectures. Particular attention is directed toward understanding the effects of different stimuli and intermodal driving terms.

© 1986 Optical Society of America