A Gaussian Algorithmic Stability Framework for Post-Selection Inference
作者:
时间:2025-09-28
阅读量:98次
  • 演讲人: 钟威(厦门大学,教授)
  • 时间:2025年9月30日10:00
  • 地点:浙江大学紫金港校区行政楼1417报告厅
  • 主办单位:浙江大学数据科学研究中心

Abstract:For statistical inference after model selection with the same data, the classical theory of statistical inference may not be valid, known as the double dipping problem. Zrnic & Jordan (2023) proposed an algorithmic stability method that incorporates randomization to enable post-selection corrections without computational burdens. However, it suffers from the conservativeness of the confidence intervals for complex composite algorithms due to the heavy-tailed nature of the Laplace noise. To this end, we propose a novel Gaussian Algorithmic Stability (GAS) framework for post-selection inference by introducing a new f-stability concept to obtain narrower valid confidence intervals. Different from the three tuning parameters required in Zrnic & Jordan (2023), our approach involves only two, thereby simplifying the tuning process. Theoretically, we establish the coverage guarantee of the proposed method and examine the regimes under which the proposed method yields narrower confidence intervals. In addition, we propose an aggregation approach to reduce randomness in parts which are irrelevant to inference without losing the validity of statistical inference. Numerical studies demonstrate that the proposed method has superior empirical performance.

Bio: 钟威,现任厦门大学王亚南经济研究院、经济学院统计学与数据科学系南强特聘教授、系主任、博士生导师。2012年获得美国宾夕法尼亚州立大学统计学博士学位,国家优青(2019),福建省杰青(2019),国家重大人才工程领军人才(教育部,2024)。主要从事高维数据统计分析、统计学习算法等研究。先后担任美国统计协会会刊JASA等6个期刊编委(AE),在Annals of Statistics, Journal of the American Statistical Association, Biometrika, Journal of Machine Learning Research, 中国科学数学等国内外统计学权威期刊发表(含接收)40多篇论文,获得2020年获得厦门大学青年教师技能比赛特等奖, 2022年获得霍英东教育基金会高等院校青年科学奖,2024年获得第九届高等学校科学研究优秀成果奖,2024年获得宝钢优秀教师奖,2025年以负责人获得厦门大学卓越教学团队奖等。