Decentralized Stochastic Multi-Player Multi-Armed Walking Bandits

Multi-player multi-armed bandit is an increasingly relevant decision-making problem, motivated by applications to cognitive radio systems. Most research for this problem focuses exclusively on the settings that players have full access all arms and receive no reward when pulling same arm. Hence solve with goal of maximizing their cumulative reward. However, these neglect several important factors in many real-world applications, where limited a dynamic local subset (i.e., arm could sometimes be ``walking'' not accessible player). To end, paper proposes multi-player walking bandits model, aiming address aforementioned modeling issues. The now maximize reward, however, can only pull from collect if other We adopt Upper Confidence Bound (UCB) deal exploration-exploitation tradeoff employ distributed optimization techniques properly handle collisions. By carefully integrating two techniques, we propose decentralized algorithm near-optimal guarantee regret, easily implemented obtain competitive empirical performance.