Speedup of Hypothetical Reasoning by Experience-based Learning Mechanism

Abstract

The crucial problem of hypothetical reasoning system is its slow inference speed, while it is a useful framework for advanced knowledge-base systems. We present a hypothetical reasoning system with experience-based learning mechanism, which enables the speedup of inference in solving problems similar to the past ones by utilizing learned knowledge from prior problem-solving experience. This system acquires knowledge from the experience of trial and error behavior, which takes place in the hypothetical reasoning process. This learning method is similar to an explanation-based learning. However, unlike the explanation-based learning, this system has a learning capability even at intermediate sub-goals appeared in the inference process. Therefore, the learned knowledge is useful even in the case that a newly given goal to be proved shares the same sub-goals as those learned in the previous inference.