Parallel associative processing for knowledge bases

Abstract

Data structures such as sets, relations and graphs occur frequently in Artificial Intelligence applications. When these structures are large, as in deductive databases and knowledge bases, the performance of conventional computers is poor. The generic operations on these structures, such as pattern-directed search, set intersection and transitive closure, all involve associative matching of variable-length lexical strings. In this paper, which is an updated version of a recent UNICOM presentation, we describe novel add-on hardware which employs SIMD parallel techniques to achieve the required associative processing. The hardware is embedded in an architecture which allows knowledge bases to be stored and processed in situ. Performance figures are given.