Abstract

A major problem with current neural networks is the creation of spurious memory states in the associative retrieval process. Recently it has been shown that associative retrieval can be more effectively accomplished using a direct storage nearest neighbor (DSNN) algorithm.¹ The DSNN algorithm first computes the Hamming distance between the input and each of the stored memory images. A digital/electronic module is then used to detect the pair having the shortest Hamming distance. The nearest neighbor of the input among the sorted images can be found in one step with no iterations and, furthermore, no spurious states. We present a new real-time optical shortest Hamming distance associative retrieval technique based on the DSNN algorithm. The optical system consists of a 2-D optical exclusive or (XOR) gate, an optoelectronic detector module, and an output display unit. Hamming distances of the input and the memory Images can be obtained In parallel using the optical XOR gate. The shortest Hamming distance is determined by the optoelectronic detector. The output display unit can instantly display the nearest neighbor images. This technique is suitable for both autoassociative and heteroassociatlve retrievals.

© 1988 Optical Society of America