Multi-Player Bandits: A Trekking Approach

We study stochastic multi-armed bandits with many players. The players do not know the number of players, cannot communicate each other and if multiple select a common arm they collide none them receive any reward. consider static scenario, where remains fixed, dynamic enter leave at time. provide algorithms based on novel ‘trekking approach’ that guarantees constant regret for case sub-linear high probability. trekking approach eliminates need to estimate resulting in fewer collisions improved performance compared state-of-the-art algorithms. also develop an epoch-less algorithm requirement time synchronization across provided player can detect presence arm. validate our theoretical using simulation real test-bed experiments.