Learning simple recursive concepts by discovering missing examples

Abstract

In this paper we introduce a system, SmartPlus, which learns recursive concepts from a small incomplete training set the members of which all lie on non-intersecting resolution path with respect to the target recursive theory and involve different constants. Unlike recent approaches to this learning problem, our method is based upon discovering the missing examples from the training set. Here missing examples mean the ground facts corresponding to the first recursive call of the given positive examples. After finding the missing examples SmartPlus perform a heuristic (independent of training set size) based top-down search through the hypothesis space in order to learn the recursive clauses. We provide some experimental results which verifies SmartPlus’s capacity to learn recursive concepts from a small number of examples (4 to 5 positive examples and negative examples of the same order) all lying on non-intersecting resolution path involving different constants.