PARKA: A System for Massively Parallel Knowledge Representation
نویسنده
چکیده
PARKA is a frame-based knowledge representation system implemented on massively parallel hardware--the Connection Machine (CM-2). PARKA provides a representation language consisting of concept descriptions (frames) and binary relations on those descriptions (slots). The system is designed explicitly to provide extremely fast property inheritance inference capabilities. In particular, PARKA can perform fast recognition queries of the form find all frames satisfying p property constraints in O(d+p) time--proportional only to the depth, d of the knowledge base (KB), and independent of its size. This performance compares very favorably with serial systems. For conjunctive queries of this type, PARKA's performance is measured in tenths of a second, even for KBs with 100,000+ frames. A PARKA implementation of the Cyc commonsense KB yields similar timing results. In addition, for queries in a case-based planning domain of over 30,000 frames, PARKA has demonstrated speed-ups of more than 10,000 over a highlyoptimized serial representation system. Because PARKA's run-time performance is independent of KB size, it promises to scale up to arbitrarily larger domains. Thus, PARKA is computationally effective even for realistically sized KBs. Such run-time performance makes PARKA one the fastest knowledge representation systems in the world.
منابع مشابه
Massively Parallel Support for Computationally Effective Recognition Queries
PARKA is a frame-based knowledge representation system implemented on the Connection Machine. PARKA provides a representation language consisting of concept descriptions (frames) and binary relations on those descriptions (slots). The system is designed explicitly to provide extremely fast property inheritance inference capabilities. In particular, PARKA can perform fast "recognition" queries o...
متن کاملParka on Mimd-supercomputers
We have ported the SIMD Parka knowledge representation system to generic MIMD machines. The system has been recoded in C and supported using runtime optimization packages developed in the High Performance Systems Software Laboratory at the University of Maryland. New \scanning" algorithms have been developed for inheritance and recognition inferences. These algorithms have been tested with both...
متن کاملMassively Parallel Simulation of Structured Connectionist Networks: An Interim Report
We map structured connectionist models of knowledge representation and reasoning onto existing general purpose massively parallel architectures with the objective of developing and implementing practical, realtime knowledge base systems. Shruti, a connectionist knowledge representation and reasoning system which attempts to model reflexive reasoning, will serve as our representative connectioni...
متن کاملMassively Parallel Knowledge Representation
We present a massively parallel representation of transitive relations, emphasizing the subclass relation, which extends our previous linear tree representation of class hierarchies. This representation makes use of a grid of processors and distributes information about one class over the processors of one column of the grid. As such, it can deal with a subset of directed acyclic graphs. We pro...
متن کامل