IEEE Transactions on Automatic Control, Vol.62, No.4, 2041-2047, 2017
Myopic Allocation Policy With Asymptotically Optimal Sampling Rate
In this note, we consider the statistical ranking and selection problem of finding the best alternative when the performances of each alternative must be estimated by sampling. We provide a myopic allocation policy that asymptotically achieves the sampling ratios given by the optimal computing budget allocation, an approximate solution of the optimal large deviations rate for the decreasing probability of false selection. We analyze the asymptotic sampling ratio for both known variances and unknown variances under a Bayesian framework. Numerical results substantiate the theoretical results.