A hybrid approach to solving coarse-grained DisCSPs

A coarse-grained Distributed Constraint Satisfaction Problem (DisCSP) consists of several loosely connected constraint satisfaction subproblems, each assigned to an individual agent. We present Multi-Hyb, a two-phase concurrent hybrid approach for solving DisCSPs. In the first phase, each agent's subproblem is solved using systematic search which generates the key partial solutions to the global problem. Concurrently, a penalty-based local search algorithm attempts to find a global solution from these partial solutions. If phase 1 fails to find a solution, a phase 2 systematic search algorithm solves the problem using the knowledge gained from phase 1. We show that our approach is highly competitive in comparison with other coarse-grained DisCSP algorithms.