English

The Secretary Problem with Predictions

Data Structures and Algorithms 2023-08-21 v1

Abstract

The value maximization version of the secretary problem is the problem of hiring a candidate with the largest value from a randomly ordered sequence of candidates. In this work, we consider a setting where predictions of candidate values are provided in advance. We propose an algorithm that achieves a nearly optimal value if the predictions are accurate and results in a constant-factor competitive ratio otherwise. We also show that the worst-case competitive ratio of an algorithm cannot be higher than some constant <1/e< 1/\mathrm{e}, which is the best possible competitive ratio when we ignore predictions, if the algorithm performs nearly optimally when the predictions are accurate. Additionally, for the multiple-choice secretary problem, we propose an algorithm with a similar theoretical guarantee. We empirically illustrate that if the predictions are accurate, the proposed algorithms perform well; meanwhile, if the predictions are inaccurate, performance is comparable to existing algorithms that do not use predictions.

Keywords

Cite

@article{arxiv.2306.08340,
  title  = {The Secretary Problem with Predictions},
  author = {Kaito Fujii and Yuichi Yoshida},
  journal= {arXiv preprint arXiv:2306.08340},
  year   = {2023}
}
R2 v1 2026-06-28T11:04:46.679Z