Abstract

Fuzzy logic provides a means of working with qualitative data and rules. Like neural networks, fuzzy logic provides a mapping from a space of input variables to a space of output variables. Unlike neural networks, fuzzy logic provides a direct means of incorporating human expertise. For a given problem, the computational requirements for a fuzzy logic solution are often significantly less than those required by a neural network solution. However, fuzzy logic implementations are faced with a need for global communication in the distribution and collection of data. An optical processing architecture has been designed that allows for the global communication paths required by fuzzy logic. Multiple fuzzy rules can be processed in parallel. The fuzzy nature of the data is maintained throughout the system, thus preserving accuracy and fault tolerance. As a final step, a centroid is formed for each fuzzy output variable, providing non-fuzzy (crisp) values for additional processing.

© 1992 Optical Society of America