An Optical Neural Device

"Optical Neural Networks can perform function mapping problems, such as target classification, faster than any other technology available today. However, they are slow at learning new mappings. This is because learning requires changing the optical synapticweights, which us usually done off-line by a conventional computer. It also requires that the new synaptic medium - such as a photographic plate - be created and installed.We propose a solution to this problem called Optical Fixed-Weight Learning Neural Networks, which can learn at full optical speeds. These networks store mapping information in recurrent neural signal loops. They also have their learning algorithm embeddedor wired into their neural circuitry.Instead of having information about how to do a single functional mapping stored in their synaptic weights, Optical Fixed-Weight Learning Neural Networks have information about how to learn any mapping stored in their synapses. Thus, these networks canlearn new mappings without changing the synaptic medium. Very fast learning optical neural networks."