Inductive logic programming and learnability

Abstract

The paper gives an overview of theoretical results in the rapidly growing field of inductive logic programming (ILP). The ILP learning situation (generality model, background knowledge, examples, hypotheses) is formally characterized and various restrictions of it are discussed in the light of their impact on learnability. The two dominant models of learnability, PAC-learning and identification in the limit, are extended to take into account the ILP learning situation. Several learnability results for logic programs are then presented, both positive and negative.