Abstract

The proposed hybrid electrooptical implementation of the Huberman-Hogg (H-H) model has several major components.¹ The H-H model specifies that the neuron connections are between nearest diagonal neighbors and that the output of a neuron is a nonlinear function of the neuron memory and a difference operation performed on the neuron inputs. The first component is a liquid crystal TV which serves as an input device that converts electrical signals to a 2-D array of optical signals. This 2-D array simulates one layer of neurons. The second component is an optical system that specifies the interconnections between the neurons that are appropriate for the H-H model. The third component is a CCD camera that serves as an output device which transforms the 2-D optical interconnection signals to electrical signals. The fourth component is a digitizer processor that stores the internal state of each neuron in digital form and performs the output calculations and logic operations required for each neuron. The system is arranged in a loop so that the outputs of one 2-D array, or layer of neurons, become the inputs to the next layer of neurons.

© 1987 Optical Society of America