暗网禁区 学术报告[2026]084号

（高水平大学建设系列报告1343号）

报告题目：Homogeneity Pursuit in Ranking Inference Based on Pairwise Comparison

报告人：陶宇心 助理教授 （南方科技大学）

报告时间：2026年9月12日16:30-17:30

报告地点：校友广场305会议室

报告摘要：The Bradley-Terry-Luce (BTL) model is one of the most celebrated models for ranking inference based on pairwise comparison data, ranking individuals by their latent preference scores. A critical question that arises is the uncertainty quantification for ranks. Intuitively, the relative ranks of two individuals become unreliable when their preference scores differ only subtly. In this paper, we explore the homogeneity of preference scores in the BTL model, which assumes that individuals cluster into groups with the same scores. We propose novel penalized maximum likelihood estimators (MLE) to simultaneously and rigorously perform estimation and clustering. We establish the statistical properties of the proposed methods and develop corresponding inference procedures. By leveraging this group structure, we achieve a faster convergence rate and sharper confidence intervals for the MLE of preference scores, providing new insight into the power of exploiting low-dimensional structures in high-dimensional settings. To address computational challenges, we further develop a majorized ADMM algorithm for efficient optimization with convergence guarantees. Extensive simulations and real data analyses, including NBA team rankings and statistical journal rankings, demonstrate the improved prediction performance and enhanced interpretability of our model.

报告人简介：陶宇心，南方科技大学统计与数据科学系助理教授，博士生导师。博士毕业于清华大学统计系，曾访问美国哈佛大学统计系。研究方向为金融计量学、时间序列分析、网络数据分析、偏好排序等，以及统计学在生态学、流行病学领域的应用。成果发表于美国国家科学院院刊PNAS、Journal of Econometrics、Statistica Sinica等。主持国家自然科学基金青年项目（C类)、广东省自然科学基金面上项目。曾获国际数理统计协会(IMS)新研究员旅行奖、IMS汉南研究生旅行奖、清华大学国家奖学金、北京应用统计学会学术研讨会优秀论文奖等。

邀请人：王江洲

暗网禁区

2026年9月9日