Experiments of learning in optical perceptron-like and multilayerneural networks

摘要

An optical multilayer network with backpropagation learningcapability is constructed and tested. The modifiable connection weightsand the unit plane are realized by using a photorefractive crystal and amicrochannel spatial light modulator, respectively. First, in a simpleoptical perceptron-like network, the experiment of learning isconducted. The learning rate is optimally determined by setting theexposure time of the hologram based on the temporal characteristics ofthe photorefractive crystal. Next, the experiment is extended to theoptical three-layer network. The optimal error signal forbackpropagation learning is successfully generated. By incorporating theoptical error signal into the network, the experiment of learning isperformed. Due mainly to optical losses of the system related to theholograms and the unit planes, the projected performance is not fullyrealized. The preliminary experiment is individually conducted by partlydisconnecting the optical interconnections between the optical elements.However, the key performances to a full optical realization ofbackpropagation learning are obtained