An algorithm for Multi-Criteria Optimization in CSPs

Constraint Satisfaction and Optimization are important areas of Artificial Intelligence. However, in many real-life applications, more functions should be optimized at the same time; the user needs to be provided a set of solutions and a posteriori choose the most preferable. In this paper, we propose an algorithm for solving Multi-Criteria Optimization problems in this setting. The algorithm is complete, i.e., it finds all the non-dominated solutions, and does not make any assumption on the structure of the constraints nor on the type of the objective functions. It exploits Point Quad-Trees for the representation of the non-dominated frontier, in order to efficiently access the data. We describe the implementation and give experimental results showing that our algorithm outperforms widely used methods.