Multiple Sampling and Cooperative Search Strategy on Sampling-Based Distributed Constraint Optimization Method

Distributed Constraint Optimization Problem (DCOP) has been studied as a fundamental problem in multiagent system. Distributed Gibbs (DGibbs) is a sampling-based solution method that performs a stochastic search on pseudo-trees of DCOPs. Since DGibbs is a synchronous distributed algorithm, where several agents exclusively perform local search, its message communication costs are relatively large. Moreover, due to the redundant sampling, the improvement of solution quality in the first steps is delayed. In this study, we investigate the effects of concurrent sampling in the same messages. In addition, a cooperative search of neighborhood agents improves the solution quality in the first steps of the search processing.