Regret Analysis of the Anytime Optimally Confident UCB Algorithm