Optimal Regret Bounds for Collaborative Learning in Bandits

Amitis Shidani, Sattar Vakili

We consider regret minimization in a general collaborative multi-agent multi-armed bandit model, in which each agent faces a finite set of arms and may communicate with other agent…