A System for Abductive Learning of Logic Programs

We present the system LAP (Learning Abductive Programs) that is able to learn abductive logic programs from examples and from a background abductive theory. A new type of induction problem has been dened as an extension of the Inductive Logic Programming framework. In the new problem denition, both the background and the target the- ories are abductive logic programs and abductive derivability is used as the coverage relation. LAP is based on the basic top-down ILP algorithm that has been suit- ably extended. In particular, coverage of examples is tested by using the abductive proof procedure dened by Kakas and Mancarella [24]. As- sumptions can be made in order to cover positive examples and to avoid the coverage of negative ones, and these assumptions can be used as new training data. LAP can be applied for learning in the presence of incomplete knowledge and for learning exceptions to classication rules.