Abstract:
The results are given of recent investigations of the feasibility of using photorefractive crystals in the construction of optical neural computers and of associative memories. The physical basis of the interaction of laser radiation with photorefractive crystals is given and the principles governing the formation of an all-optical neuron and of links (synapses) between neurons in such crystals are discussed. An analysis is made of the learning capabilities of various models of neural networks (Boltzmann machine, perceptron, associatron, neural networks with competition, etc.) and optical systems implementing these models with the aid of photorefractive crystals are described. Extensive experimental data are reported and the results are given of modeling of various tasks (multidimensional optimization, image recognition, etc.) by optical neural computers.