Inductive Learning of Answer Set Programs

In recent years, non-monotonic Inductive Logic Programming (ILP) has received growing interest. Specifically, several new learning frameworks and algorithms have been introduced for learning under the answer set semantics, allowing the learning of common-sense knowledge involving defaults and exceptions, which are essential aspects of human reasoning.

The first part of this seminar will present our recent advances which have extended the theory of ILP and yielded a new collection of algorithms, called ILASP (Inductive Learning of Answer Set Programs), which are able to learn ASP programs consisting of normal rules, choice rules and both hard and weak constraints. Learning such programs allows ILASP to be applied in settings which had previously been outside the scope of ILP. In particular, weak constraints represent preference orderings, and so learning weak constraints allows ILASP to be used for preference learning.

In the second part, I will present a noise-tolerant version of ILASP, which has been successful at learning both from synthetic and from real data sets. In particular, we have shown that on many of the data sets ILASP achieves a higher accuracy than other ILP systems that have previously been applied to those same data sets.