Control of Hypothesis Space Using Meta-knowledge in Inductive Learning

摘要: Inductive logic programming (ILP) is effective for classification learning because it constructs hypotheses combining background knowledge. On the other hand makes cost of search hypothesis large. This paper proposes a method to prune using kind semantic When an ILP system uses top-down search, after visits clause (rule) explore another by adding condition. The added condition may be redundant with conditions in or causes body unsatisfied. We study represent and use treat redundancy unsatisfactory as meta-knowledge predicates. In this we give formalism show algorithm. also generate automatically. generates which controls contradiction respect predicates testing properties extensionally.