Thompson Sampling for the Duelling Bandits Problem

In surprisingly many situations, absolute rewards are not available (or nonstationary) while relative preferences are easy to collect (or stable). This variation of the bandit problem is known at th