An optical learning machine

Abstract

The authors report on what they believe to be the first demonstration of a fully operational optical learning machine. Learning in this machine takes place stochastically in a self-organizing trilayered optoelectronic neural net with plastic connectivity weights that are formed in a programmable nonvolatile spatial light modulator. The net learns by adapting its connectivity weights in accordance with environmental inputs. Learning is driven by error signals derived from state-vector correlation matrices accumulated at the end of fast annealing bursts that are induced by controlled optical injection of noise into the network. Operation of the machine is made possible by two developments: fast annealing by optically induced tremors in the energy landscape of the net, and stochastic learning with binary weights. Details of these developments together with the principal, architecture, structure, and performance evaluation of the machine are given.<>