|
|
|
“Neural Networks based on the Incoherent Optical Neuron Model”
by Chein-Hsun Wang and B. Keith Jenkins
April 1991
To fully use the advantages of optics in parallel implementations of neural networks, an incoherent optical neuron (ION) model can be used to optically implement neurons with both excitatory and inhibitory inputs. The main purpose of this model is t provide for the requisite subtraction of signals at each neuron unit without the phase sensitivity of a fully coherent optical system and without the cumbrance of photon-electron conversion and electronic subtraction. The ION model, in conjunction with coherent or incoherent optical weighted interconnections, can be used to implement arbitrarily connected neural networks. This chapter describes techniques for implementation of both analog and binary inner product neuron units as well as mass action law neuron units. Potential use of the ION model in implementing the neocognition model, multilayer networks, selective attention for winner-take-all networks, and simple cells of the visual cortex are discussed. In addition, experimental results on the optical implementation of a 2-D array of IONs and of sequential feature detection operations of the visual cortex, are reviewed.