All-optical error-signal generation for backpropagation learning in optical multilayer neural networks.

An all-optical error signal based on the generalized delta rule for backpropagation learning in optical multilayer networks is experimentally obtained. Optically addressable microchannel spatial light modulators are used to perform key functional operations for the error signal such as subtraction, soft thresholding, and taking the derivative of the thresholding function. The error signals are readily available to adaptive optical multilayer networks in which photorefractive crystals are used to determine the connection weights between processing units.