Using Relevance to Speed Up Inference. Some Empirical Results

Joselyto RianiRenata Wassermann

One of the main problems in using logic for solving problems is the high computational costs involved in inference. In this paper, we propose the use of a notion of relevance in order to cut the search space for a solution. Instead of trying to infer a formula agr directly from a large knowledge base K, we consider first only the most relevant sentences in K for the proof. If those are not enough, the set can be increased until, at the worst case, we consider the whole base K.We show how to define a notion of relevance for first-order logic with equality and analyze the results of implementing the method and testing it over more than 700 problems from the TPTP problem library.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: