MASC: Multiprocessor Architecture for Symbolic Processing.

摘要

The MASC program addresses the support of symbolic computation for Artificial Intelligence (AI), as part of the DARPA Strategic Computing Program. The work has been aimed at the fundamental questions of choosing an appropriate programming model for AI and finding efficient implementation techniques to support it on multiprocessor systems. The Strategic Computing Program is motivated by the need to provide real-time response by AI subsystems in planned or contemplated DoD systems. Our efforts have directly supported this goal, contributing fundamental results in language design, language implementation, and application parallelism. The work has been driven by the needs of real AI applications, drawn from our substantial experience in natural language processing, knowledge representation, and expert systems. The technologies we have developed are broadly applicable, and several specific recommendations for their further development and application are presented. The most profound result is the design and demonstration of a powerful technique for the compilation of logic programs to applicative form. This brings the power of compilation and optimization techniques for functional programming to bear on logic programming languages. We recommend that these technologies be extended and applied to current research efforts in language design for very high level programming and program prototyping. Other results and recommendations are contained in the body of the report. Keywords: Parallel processing. (kr)