In this paper, we introduce a multi-armed bandit problem termed max-min
grouped bandits, in which the arms are arranged in possibly-overlapping groups,
and the goal is to find the group whose worst arm has the highest mean reward.
This problem is of interest in applications such as rec