Abstract

The Infernet is an architecture for an inference machine that is composed of three sections: (1) an input processor that performs specific preprocessing on the inputs; (2) an output processor for generating outputs; and (3) a hierarchical inference processor which functions as the inference engine. Each section is comprised of several neural network layers. Time-domain pulse neurons are being considered as the basic cells within the layers. Temporal waves of neural activity which are supported by the interconnection topology are being considered for encoding and accessing high-level knowledge. An optical implementation for a neural layer has been designed using two optically addressed spatial light modulators (SLMs), an electrically addressed SLM, and an electronic controller.

© 1988 Optical Society of America