A Distributed Cooperative Search Algorithm Using Multiple Contexts and Pruning

Abstract Distributed constraint optimization problems (DCOPs) have been studied as a fundamental framework of cooperative problem solving in multiagent systems. Exact solvers based on tree search and dynamic programming have been proposed for DCOPs. Based on tree searches, the solvers perform iterative processing that depends on message communication. Therefore the overhead of the communication affects the execution time of the search. On the other hand, solvers that are completely based on dynamic programming require no iterative processing among agents. However, its memory and message size is exponential to the induced width of pseudo-trees. Although memory-bounded solvers that employ both methods have been proposed, several solvers search for one solution at a time. Other solvers employ no pruning based on global cost values. Especially, when the communication overhead is relatively large, simultaneously expanding multiple solutions and transferring them by the same message are reasonable. In this study, we present the basic algorithms of memory-bounded solvers that employ multiple search points.