报告题目：Selective conformal inference

报告时间：2026年2月28日下午19:00

会议链接：//meeting.tencent.com/dm/a6Yt66RiMauh

腾讯会议ID：503-121-119

主办单位：红杏视频 /科研处

报告人：邹长亮

报告人简介：邹长亮，南开大学统计与数据科学学院教授。主要从事统计学及其与数据科学领域的交叉研究和实际应用。研究兴趣包括：预测性推断、高维数据统计学习、变点和异常点检测等。近年来在统计学和机器学习领域的权威期刊和会议上发表发表论文五十余篇，入选爱思唯尔“中国高被引学者”。主持基金委优青、杰青、重点项目、重大项目课题和科技部重点研发计划课题等。任教育部科技委委员、全国应用统计专业硕士教学指导委员会委员、中国现场统计研究会副理事长等。

摘要：Conformal inference is a popular tool for conducting distribution-free predictive inference. This talk addresses selective conformal inference: providing valid uncertainty quantification for targets after data-dependent selection. We present a robust framework to maintain statistical rigor in adaptive analysis. Our primary contributions include: 1) Developing theory and algorithms to rigorously control the False Coverage-statement Rate (FCR), ensuring coverage validity only among the selected targets. 2) Introducing practical methods for various settings, including conformalized multiple testing and the efficient algorithm for online selective prediction. This suite of methods enables reliable post-selection uncertainty quantification.