Toward Pattern-Recognizing Visual Prostheses

Abstract

Chip technology and VLSI techniques have advanced to where photo-transistor arrays as retinal models and prostheses may soon become serious possibilities. This paper is concerned with principles of logical design/computer architecture permitting such arrays to recognize patterns by parallel/distributed computation, producing decisions/outputs suitable either for human interpretation, for actuating other parts of the system (e.g., motor responses), or for entering into feedback loops with or without human links in the system. The central concept is the busautomaton (BA), a parallel computer of speed and power far exceeding that of cellular automata. Its main novel feature, not shared with the latter, is a distributed switching/communication network under local control by corresponding individual automata (cells). Widely separated cells can effectively become "nearest neighbors", and cooperate in complex problems (like pattern recognition), solving them in parallel in times comparable to those for simple logical decisions. The wide variety of problems already solved in theory (or close to it) will be reviewed, including many pattern recognitions, complex computations, formal language processing, and a new approach to modeling skill acquisition. Obstacles to implementing rudimentary eye or eye-brain-effector systems will be assessed from BA and physical viewpoints.