Output details
11 - Computer Science and Informatics
Imperial College London
Meta-Interpretive Learning of Higher-Order Dyadic Datalog: Predicate Invention Revisited
<24>We introduce an efficient new form of Inductive Logic Programming (ILP) which overcomes limitations of state-of-the-art systems by supporting predicate invention and the learning of recursion. The approach uses a Prolog meta-interpreter driven by a set of higher-order Datalog definite clauses. The resulting Meta-Interpretive Learning approach has been shown to be significantly more efficient than standard ILP. The paper shows that the approach can be extended to a Turing-equivalent representation.This form of ILP has the potential to be applied to tasks such as the learning of proof tactics and complex recursive robot planning strategies. IJCAI'13 Acceptance: 28%/1473