Models and tools for cooperating rule-based systems
Clifford Grossner · eScholarship@McGill (McGill) · 1994
A Cooperative Distributed Problem Solving system (CDPS) is composed of a set of agents designed to solve a single problem by working together in a cooperative fashion. In this thesis, we present models and tools which aid the CDPS designer in determining how data items are to be shared between the agents (data distribution) in a CDPS, when the agents are implemented as rule-based systems. Our models consider CDPS from two different perspectives: the inter-agent perspective, and the intra-agent perspective. The inter-agent perspective is concerned with issues relating to the manner in which the agents in a CDPS achieve cooperation. The intra-agent perspective is concerned with the internal structure of an agent, and how an agent is affected when it cooperates with other agents. The inter-agent perspective is modeled using a notion of an "organization" that is imposed on a set of agents sharing data items via a "blackboard"; this model permits the CDPS designer to indicate the mechanism used to achieve both forms of cooperation between the rule-based systems in a CDPS: cooperation by sharing data, and cooperation for control. The intra-agent perspective is modeled using a notion of a "path", which captures the sequence of rules fired by a rule-based system. Based upon the path model, we have developed two tools: Path Hunter and Path Tracer. The models and tools described in this thesis permit the CDPS designer to study the impact of the data distribution chosen for a CDPS on the performance of a rule-based system that will be a member of that CDPS; this helps the CDPS designer in setting the data distribution within a CDPS.