14 octobre 2019Johan Thapper (LIGM - UPEM)
The valued constraint satisfaction problem (VCSP) is a framework for describing and studying many classes of discrete optimisation problems. It is an optimisation version of the constraint satisfaction problem (CSP), where the goal is to decide whether one can assign labels to variables under some given set of constraints. The VCSP captures both Max CSP-type problems, where one wants to optimise the number of satisfied constraints, and integer programming-type problems, where one adds an objective function to an ordinary CSP to indicate the desirability of each solution.
I will talk about algorithms and hardness results for VCSPs where we restrict the types of cost functions that are allowed to appear in the description of an instance. The algorithms that I consider are based on linear programming (LP), in particular LP relaxations in the Sherali-Adams hierarchy. It turns out that for large classes of VCSPs, there is a complete dichotomy between problems that can be solved by this type of algorithm and problems that can encode NP-hard problems.