Explore no more: Improved high-probability regret bounds for non-stochastic bandits | Read Paper on Bytez