报告人
Jiechen Zhang
École Polytechnique Fédérale de Lausanne
时间
2026年3月6日 星期五
下午 14:00-15:00
地点
602会议室
Abstract
The Prophet Inequality is a cornerstone of optimal stopping. While the classic Prophet Inequality relies on full distributional knowledge, recent research has successfully extended it to limited information settings, showing that access to even a single sample is enough to recover the optimal guarantee. In this talk, I investigate a different natural model: what if the decision-maker knows only the moments of the distributions?
To start, I will show a simple algorithm that achieves an $\Omega(1/\log n)$ competitive ratio using only the first moment. Then, I will present a strong negative result: unlike the sample-based case, even knowledge of all moments is insufficient to achieve a competitive ratio better than $O(1/\log n)$. This implies that higher-order moments provide no additional utility over knowing just the mean in the worst case.
Finally, I will discuss how structural assumptions can break this logarithmic barrier. If time permits, I will also share some additional unpublished results and insights derived that go beyond the scope of the paper.
Biography
Jiechen Zhang is a Ph.D. candidate at EPFL, advised by Prof. Andrés Cristi. He received his bachelor's degree with honours in mathematics and computer science from McGill University. His research focuses on stochastic optimization, mechanism design, and online decision-making in large-scale systems.
