Abstract

The basic functions of neural network models are vector-matrix multiplication and nonlinear function. In recent years, many kind of systems to achieve these functions by optical parallel technique have been proposed. In most of these systems, however, the learning processes have not been done by optical systems. A rudimentary system to achieve learning with optical outer-product technique was presented in Ref.1. But the device to memorize and display weight matrices in the system had poor characteristics and the system did not show enough performances. In this paper we present the system that performs learning process by optical parallel calculations using a Pockels readout optical modulator (PROM) as a key device.

© 1997 Optical Society of America