Coalition structure formation using anytime dynamic programming

Abstract

The optimal coalition structure generation is an important problem in multi-agent systems that remains difficult to solve. This paper presents a novel anytime dynamic programming algorithm to compute the optimal coalition structure. The proposed algorithm can be interrupted, and upon interruption, uses heuristic to select the largest valued coalition from each subproblem of size x and picks the rest of the unassigned agent from other subproblem of size n − x, where n is the total number of agents. We compared the performance of our algorithm against the only existing proposal in the literature for the optimal coalition structure problem that uses anytime dynamic programming using 9 distinct datasets (each corresponding to a different distribution). The empirical evaluation shows that our algorithm always generates better or, at least, as good a solution as the previous anytime dynamic programming algorithm.