Abstract

The incoherent optical neuron (ION) model uses two separate incoherent optical device responses to subtract inhibitory inputs from excitatory inputs for general neural networks. The ION model comprises two elements: an inhibitory element and a nonlinear element. A Hughes liquid crystal light valve (LCLV) can implement both elements. Its nonmonotonic response can be used to approximate a linear, negative slope characteristic for inhibitory inputs as well as a nonlinear positive slope characteristic for the neuron hard-clipping or soft-threshold response. An optical beam is used to provide a suitable neuron threshold. Self-feedback is incorporated to increase the gain and nonlinearity.

© 1989 Optical Society of America