Abstract

We have suggested elsewhere that the combination of page oriented holographic memories with optically addressed spatial light modulators led to the ability to assess many symptoms and many derived tentative diagnoses in parallel. Our previous work showed a general approach to avoid the necessity of an exhaustive search of memory to identify a new input. We discuss the implementation of that approach using a neural network as the controller of the total system. The result is a pattern recognition system which exhibits many useful features of the human system. In particular, it forms hypotheses, determines what measurements to make to increase the certainty of those hypotheses, makes the needed observations, and either increases the certainty of diagnosis or reduces the certainty of the initial diagnosis and oscillates to a new one.

© 1987 Optical Society of America