Power-of-2-arms for bandit learning with switching costs

Ming Shi, Xiaojun Lin, Lei Jiao