Framework for Parallelism in Frame-Based Reasoning

摘要

This paper discusses issues of parallelism within frame-based reasoning systems. This effort was motivated by our recent work on intelligent signal processing and visual pattern recognition. Although rule-based systems prevail as knowledge engineering tools, we chose frame-based systems because they provide a sound knowledge representation scheme (including declarative and procedural knowledge) and a powerful (distributed) inference engine. We look at a frame-based system developed at Rice University and NASA/JSC called WenLy which is composed Wen, representing declarative knowledge, and Ly, representing procedural knowledge and taking charge of inference. Then we examine two approaches to achieve parallelism within that system. One is called the programming approach in which we implement the frames on a parallel Lisp language. The order is called the architectural approach which supports MIMD computers with uniformed memory space. It is obvious that the 'list' structure is a bottleneck to parallelism; on the other hand, it is crucial to frame-based reasoning due to its flexibility and uniformity. We believe that we should sacrifice the degree of parallelism in exchange for a reasoning mechanism and knowledge structure that are closer to those of the human being. (Author)