Optical Inference Machines
Cardinal Warde · 1988
Abstract : We have been investigating an optical inference machine based on neural network principles. This approach represents a diversion from the mapped- template architecture (C. Warde and J. A. Kottas, Appl. Opt. 25, 940, 1986) to overcome the latter system's limitations. During the second year of this grant, we designed the Infernet, a neural network architecture which we believe can solve symbolic inference problems, and a corresponding optical implementation. This report describes our new system from an architectural viewpoint. We also describe an automated two-dimensional photo-calibration system, developed for our computerized control system, that will provide fast analog intensity measurements on images as well as calibration of image registration from plane-to-plane in the optical processor.