Abstract

We are investigating associative processing architectures for tackling the massive parallel symbolic-processing demands of problems from image understanding, robotic manipulation and locomotion, expert-system problem solving, and other difficult artificial intelligence domains. A modular approach is taken, where a number of smaller adaptive, associative "modules" are nonlinearly interconnected and cascaded under the guidance of a variety of "organizational principles" to construct larger architectures for solving specific problems.¹,² Each module is a complete associative memory which adapts as it is exposed to associated information patterns (u,v), (e.g., feature vectors, encoded symbols, images,…), so that subsequent presentation of one pattern u results in recall of its paired pattern v.

© 1987 Optical Society of America