Data-Adaptive Inference with FDR Control for Structural Sparsity Problems: Applications to Brain MRI Analysis
Holder： Gaorong Li(The School of Statistics, Beijing Normal University)
Time：2026-05-11 14:00-15:00
Location：Room 818 of the New Public Health Building of Peking University Medical School
Abstract:
Controlling the false discovery rate (FDR) is essential for the reproducibility of scientific findings. However, most existing FDR control methods focus solely on traditional sparsity assumptions and face challenges under structural sparsity. In this paper, we address a fundamental challenge in modern large-scale experiments: the statistical inference in structural sparsity problems under arbitrary transformations. Through the lens of multiple testing over graphs, we propose a unified and theoretically guaranteed solution to handle structural sparsity from moderate to ultra-high dimensions based on noise injection and constraint relaxation ideas. Within this framework, we simultaneously address two data-adaptive inference tasks: structural sparsity recovery and finite-sample FDR control. The effectiveness of the proposed methods is supported by both theories and simulations. Applications to brain MRI analysis in Alzheimer's patients reveal abnormal structures of brain regions associated with memory function.
About the Speaker:
李高荣，北京师范大学统计学院教授，博士生导师，北京师范大学第十二届“最受本科生欢迎的十佳教师”。主要研究方向是非参数统计、高维统计、统计学习、纵向数据、测量误差数据和因果推断等。迄今为止，在Annals of Statistics, Journal of the American Statistical Association, Journal of Business & Economic Statistics, Statistics and Computing, 《中国科学：数学》和《统计研究》等学术期刊上发表学术论文130余篇。出版5部著作：《纵向数据半参数模型》、《现代测量误差模型》(入选“现代数学基础丛书”系列)、《多元统计分析》(荣获北京高校优质本科教材奖和“十四五”普通高等教育本科国家级规划教材）、《统计学习（R语言版）》 (荣获北京高校优质本科教材奖）、《高维统计学》。2024年荣获北京市普通高校优秀本科毕业论文优秀指导教师，2025年荣获北京师范大学高等教育教学成果一等奖。主持国家自然科学基金、北京市自然科学基金和北京市教委科技计划面上项目等国家和省部级科研项目10多项。
Your participation is warmly welcomed!
欢迎扫码关注北大统计科学中心公众号，了解更多讲座信息!
