The main distinction between ILP and other fields of Machine Learning is the use of Prolog as the basic language for expressing learned hypothesis. Recent work in ILP tried to extend the approach to variants such as indefinite clauses, or non monotonicity. In the meantime, the Logic Programming community has been exploring the use of constraints as programing primitives, yielding better results on combinatorial and numerical applications. We propose to the ILP community to use Constraint Logic Programming languages as target instead of Prolog. We will introduce the basics of CLP, and give some examples of learning constraints.
Jean-Francois Puget is currently with ILOG.