报告题目：POMT: A Private Online Multiple Testing Framework with Adaptive Budgets

 报告时间：2025年10月24日 16:00—18:00

 报告地点：太阳集团tyc5997北区第4教学楼208室

 报告人：许王莉

 报告人单位：中国人民大学

 摘要：Online multiple testing is increasingly prevalent in biomedical research and information technology, where protection of sensitive personal information is essential. To date, the only work on privacy-preserving online multiple testing is PAPRIKA \citep{zhang2021paprika}, which leverages log-transformed $p$-values, Laplace noise, and the sparse vector technique (SVT). While providing rigorous privacy and false discovery rate (FDR) guarantees, PAPRIKA distorts the super-uniformity of null perturbed $p$-values and introduces complex dependence among rejection thresholds, leading to conservative rejection rules at the cost of reduced power. We address these limitations by proposing a general Private Online Multiple Testing (POMT) framework. Our approach (i) develops a novel $p$-value transformation that preserves super-uniformity while achieving privacy guarantee, (ii) generalizes SVT to online multiple testing that effectively reduces privacy costs, and (iii) introduces a unified budget calibration rule that adaptively mitigates the additional Type-I error due to imposed privacy mechanisms. Theoretical analyses demonstrate that POMT guarantees finite-sample control of Type-I errors, including both family-wise error rate (FWER) and FDR, while also achieving higher statistical power than PAPRIKA. Extensive simulations and real-data studies validate our theoretical findings and show that POMT provides strong privacy protection with only a modest loss of power compared to non-private baselines.

 报告人简介: 许王莉，中国人民大学教授，博士生导师，中国人民大学吴玉章讲席教授。先后主持5项国家自然科学基金，北京市自然科学基金重点研究专题，教育部人文社会科学重点研究基地重大项目和教育部人文社科基金等多项科研课题。在顶尖期刊JASA, JRSSB, Biometrika, TPAMI等发表百余篇论文。先后入选“新世纪优秀人才计划”和“北京市科技新星计划”，先后获得中国第十二届北京市统计科研优秀成果奖一等奖(2014)，第一届统计科学技术进步二等奖（2021）。