Facility locations revisited: An efficient belief propagation approach

This paper studies the fixed-charge facility location problem---an important problem in logistics and operations research that has wide application to many areas of commerce and industry. The problem is to locate a small number of facilities among nodes in a network to provide good service to the client nodes while confining the total construction cost. The problem is NP-hard, but of sufficient importance to warrant developing practical heuristics. To handle large instances, we propose an algorithm based on recent advances in belief propagation and graphical models. In particular, we adapt a form of affinity propagation to approximate the problem at hand. Our experimental results demonstrate significant improvements over other popular heuristics for large-scale facility location problems.