Distributed cooperative decision making in multi-agent multi-armed bandits

We study a distributed decision-making problem in which multiple agents face the same multi-armed bandit (MAB), and each agent makes sequential choices among arms to maximize its own individual reward. The cooperate by sharing their estimates over fixed communication graph. consider an unconstrained reward model two or more can choose arm collect independent rewards. And we constrained that at time receive no design dynamic, consensus-based, estimation algorithm for cooperative of mean rewards arm. leverage from this develop algorithms: coop-UCB2 coop-UCB2-selective-learning, models, respectively. show both algorithms achieve group performance close centralized fusion center. Further, investigate influence graph structure on performance. propose novel explore–exploit index predicts relative groups terms graph, nodal centrality locations