Agentes computacionais: sobre o esquecimento de crenças integrado ao processamento de objetivos

Abstract

Oblivion is an important epistemic process for removing information from the agent’s primary attention. Unlike belief revision, whose purpose for contraction is based on the recovery of the factual consistency of the agent’s internal state, minimizing information loss, oblivion is aimed at improving the inferential reasoning computational efficiency. This however is not a trivial problem, since it is not simply the removal of beliefs that were already used and are no longer useful, but the removal of beliefs that might be irrelevant in the future. An agent’s belief base grows due to perceptions it receives from the environment and the deductions it makes, however not all these beliefs remain relevant over time. Keeping them in primary attention can impair computational efficiency; on the other hand, removing them without considering their relationship to the agent’s goals can delay or prevent the execution of such goals. So how can they be related to compose a criterion for forgetting beliefs? To answer this question, in this work, I formalize computational models of storing and forgetting beliefs, and evaluate them in relation to their contributions to memory and processing efficiencies. To my knowledge, according to the results of the literature review, this is the first work to handle oblivion in an integrated way with goal processing. The results of the experiments showed that the models proposed here allow -in specific situations- an improvement in processing and memory efficiencies. Finally, I emphasize that the contributions of this research are independent of the application domain and the agent’s practical reasoning model.