Abstract

Simplified models for the way information is stored in neural networks consist of two major components: neurons and interconnections among the neurons. Information is stored in the interconnections and the operation of the neurons themselves is often modeled as thresholding. An optical architecture is described and experimentally demonstrated that utilizes a microchannel spatial light modulator (MSLM) to optically simulate a 2-D array of neurons that can represent 2-D patterns and holograms to implement the interconnection among the pixels in the array. Two separate holograms are used and each is fabricated as a multiplexed Fourier transform hologram of a set of images. The optical system is configured so that when an input image matches one of the patterns stored in the first hologram, the associated image from the second hologram is reconstructed and imaged onto the MSLM which thresholds it. For the case of autoassociations the input and output images are the same and feedback can be used by imaging the output of the MSLM back through the optical system.

© 1986 Optical Society of America