Dynamic Sampling Policy For Subset Selection

We consider a problem of selecting a subset of a finite number of competing alternatives via Monte Carlo simulation, where the subset contains the top-m alternatives. Under the Bayesian framework, we develop a dynamic sampling policy to efficiently learn and select the top-m alternatives. The proposed sampling policy is proved to be consistent, i.e., the selected alternatives will be the true top-m alternatives as the number of simulation budget goes to infinity. Numerical results show that the proposed sampling policy outperforms the existing ones.