Real-Time Implementation of Nonlinear Optical Data Processing Functions

摘要

In this final report for work performed in the period March 1988 to June 1990 wedescribe our efforts toward optical implementations of neural network models. The parallelism of optics will allow the hardware implementation of very large numbers of neurons with high throughputs. Under this effort we have begun development of SPONN (Stimulated Photorefractive Optical Neural Network), a hybrid optoelectronic system for programmable, adaptive, and fully parallel direct physical implementations of neural network models which is based on self-pumped photorefractive phase conjugate mirrors.