Abstract

Constraints are a very natural knowledge representation formalism. However, classical constraints (which are either satisfied or not) are not so flexible and cannot describe real-life features like preferences, costs, priorities, and uncertainties. Therefore recently many formalisms for soft constraints (which can be satisfied at a certain level) have been developed. We address the problem of modeling a real-life problem by using soft constraints. In many real-life situations, one may know his/her preferences over some of the solutions, but have no idea on how to code this knowledge into the constraint problem in terms of local preferences, or also one may be able to give only a rough approximation of the desired levels of satisfaction of the constraints. We therefore suggest to treat the solution preferences as examples, and to employ a learning scheme which learns from such examples (either from scratch or from the available rough model) all the local preferences, so tha...

Citations