A Distributed Approach for Inferring Production Systems

Abstract

Tools for building production systems encounter the problem of low performance, and many researchers are working on the improvements of performance of these tools. This paper proposes a distributed approach for inferring production systems. The resulting distributed production systems are expected to be built over distributed systems with broadcast capability, and production rules on different sites work in a cooperative way with only a few communications between them. Working memory on a local site is made visible to rules on remote sites. A tool for building distributed production systems, called DPS, has been implemented. DPS not only supports elegant constructs for expressing the capacity of distributed inference but also provides the facilities for building clusters of rules. With these facilities, DPS allows users to make the inference engine focus on a particular set of rules. This paper also describes the knowledge representation and other features about DPS.