OSSA procedures with common random numbers

Develop one-shot screening for acceptability (OSSA) procedures that exploit common random numbers by constructing confidence regions that account for dependence across systems, while retaining valid screening guarantees and useful screening power.

Background

The paper notes that common random numbers can be advantageous in ranking-and-selection problems because they induce dependence across systems and may improve the performance of the worst returned system. However, common random numbers typically require all systems to receive a common number of replications, and the resulting dependence is not incorporated into the OSSA/FOSSA confidence-region constructions developed in the paper.

Preliminary numerical experiments suggest that applying the existing FOSSA procedures to common-random-number outputs does not undermine the promised probability of acceptable selection guarantee, but the paper does not provide a formal development for this setting. It therefore identifies the construction of dependence-aware confidence regions and corresponding OSSA procedures as future research.