An Efficient Inference Method for Obtaining Optimal Solution in Predicate-Logic Hypothetical Reasoning

Abstract

Knowledge is often incomplete ; that is, it often involves exception or contradiction. Handling incomplete knowledge in knowledge-bases is an important function to expand the capability of current knowledge-bases. Hypothetical reasoning handles such incomplete knowledge as hypothesis. It is abductive reasoning and can be directly applicable to model-based diagnosis problems, design problems, etc. Thus hypothetical reasoning is an important framework as an advanced knowledge-based system both from theoretical and practical view points. The most crucial problem with hypothetical reasoning is its slow speed due to its non-monotonic reasoning nature. We have developed fast hypothetical reasoning systems to overcome this problem. KICK-HOPE (Knowledge-Base Handling Incomplete Knowledge-by Holding Parallel Solutions on Environment Lattice) is one of those systems ; its reasoning mechanism is related to SLD-AL, or QSQR method developed as an efficient deductive data-base technology and works for predicate Hornclause knowledge. KICK-HOPE shows high efficiency for many problems. However, when the constraint for hypotheses is not strong, the number of synthesized hypotheses becomes too large to calculate, because KICK-HOPE calculates all the solutions. In many practical cases, we need one or a few good solutions quickly rather than all the solutions. Therefore, we have developed a fast hypothetical reasoning system, based on KICK-HOPE, for obtaining an optimal solution which is evaluated according to the assigned hypothesis' weight. For finding the optimal solution efficiently, beam search, best-first search and branch-and-bound search strategies are incorporated to hypothesis-synthesis on KICK-HOPE. The effectiveness of this method is shown experimentally by faultdiagnosis problems of logic-circuits.