Heuristic Inverse Subsumption in Full-Clausal Theories

摘要

Inverse entailment (IE) is a fundamental approach for hypothesis finding in explanatory induction. Some IE systems can find any hypothesis in full-clausal theories, but need several non-deterministic operators that treat the inverse relation of entailment. In contrast, inverse subsumption (IS) is an alternative approach for finding hypotheses with the inverse relation of subsumption. Recently, it has been shown that IE can be logically reduced into a new form of IS, provided that it ensures the completeness of finding hypotheses in full-clausal theories. On the other hand, it is still open to clarify how the complete IS works well in the practical point of view. For the analysis, we have implemented it with heuristic lattice search techniques used in the state-of-the-art ILP systems. This paper first sketches our IS system, and then shows its experimental result suggesting that the complete IS can practically find better hypotheses with high predictive accuracies.