In a multi-armed bandit, an agent must repeatedly choose one of K arms and only observes the reward of the arm it pulls.
