Fuzzy associative memories: identification and control of complex systems

Fundamental properties of fuzzy associative memories (FAMs), which form a class of fuzzy neural networks, are examined. Fuzzy neural networks combine notions of conventional neural networks and fuzzy set theory, which has been proved suitable for coping with system uncertainty. Atomic and composite FAMs are defined, and their recall behavior is studied. The recall process is recognized as a reasoning process. An approximate FAM is defined on the basis of the l/sup p/-distance of two fuzzy subsets to provide a quantitative measure of the recall process. Fuzzy neural networks are applied to problems which possess high complexity and uncertainty. The FAM implementation issue is addressed in terms of three functional units: a min-net which performs fuzzy conjunction operations, a max-net which performs fuzzy disjunction operations, and a learning unit which adjusts the connection coefficients whenever necessary.<
>