Expert-Rete: um motor de inferência para agentes cognitivos

Abstract

This dissertation has focus on inference techniques used in Artificial Intelligence, demonstrating the relevance of the inference engine in Multiagent Systems. A general-purpose inference engine is implemented based on the Rete Algorithm, the Expert-Rete, includig as contributions the optimization of the one-input nodes; the optimization of the twoinput nodes; 6 strategies for conflict resolution; adoption of 3 formalisms to knowledge representation; other configuration parameters, like maximum number of interaction between rules and facts, maximum number of inference cycles. The Expert-Rete inference engine was incorporated in the library Expert-Coop++, used to build Multiagent Systems. After this implementation, the Expert-Rete was analyzed among some aspects, comparing with the inference engine based on the forward chaining, already developed in the library, proving that Expert-Rete is more efficient than the former method for complex base rules.