An Upper Confidence Bound Approach to Estimating the Maximum Mean