Hybrid Regret Bounds for Combinatorial Semi-Bandits and Adversarial Linear Bandits

We first propose an algorithm for combinatorial semi-bandits with a hybrid regret bound that includes two main features: a best-of-three-worlds guarantee and multiple data-dependent regret bounds.