Stochastic dominance in stochastic DCOPs for risk-sensitive applications

摘要: Distributed constraint optimization problems (DCOPs) are well-suited for modeling multi-agent coordination where the primary interactions between local subsets of agents. However, one limitation DCOPs is assumption that rewards without uncertainty. Researchers have thus extended to Stochastic (SDCOPs), sampled from known probability distribution reward functions, and introduced algorithms find solutions with largest expected reward. Unfortunately, such a solution might be very risky, is, likely result in poor Thus, this paper, we make three contributions: (1) propose stricter objective SDCOPs, namely most stochastically dominating function; (2) introduce an algorithm solutions; (3) show can indeed less risky than maximizing solutions.