报告人： 钟齐先 副教授

报告题目： Conditional Independence Testing with Survival Data

时间： 2026年10月28日（周三）10:00—12:00

地点： 西安交通大学数学与统计学院2-3会议室

邀请人： 朱学虎 教授

报告摘要：

Conditional independence testing is a fundamental challenge in statistics and machine learning, with direct applications in causal discovery, feature selection, and fairness assessment. In survival analysis, where outcomes are frequently subject to right-censoring, classical conditional independence tests designed for complete data are often invalid or inefficient due to the necessary shift from mean-based to hazard-based modeling.

This paper proposes a novel, assumption-lean methodology for testing conditional independence with right-censored survival data. Our approach integrates the classical score test principle with double machine learning techniques to introduce an orthogonalized score statistic. This statistic is robust to estimation errors in nuisance functions—specifically the conditional hazard and a data-driven direction function, the latter of which is learned to maximize power against alternative hypotheses.

We further establish uniform asymptotic validity and provide power guarantees under mild conditions on the nuisance estimators. Theoretical results demonstrate that the test retains power against local alternatives converging at the $n^{-1/2}$ rate, even when nuisance estimators converge at rates slightly faster than $n^{-1/4}$. This ensures the effectiveness of the proposed method in high-dimensional and nonparametric settings.

Numerical studies demonstrate that the proposed method effectively controls Type I error and achieves superior power compared to existing competitive methods.

报告人简介：

钟齐先，厦门大学统计学与数据科学系副教授、博士生导师。主要研究领域包括生存数据分析、深度学习、函数型数据分析和因果推断等。

相关学术成果发表于 Annals of Statistics（AOS）、Journal of the American Statistical Association（JASA）、Biometrika、Journal of Econometrics（JOE）、Journal of Business & Economic Statistics（JBES）、NeurIPS 等学术期刊或会议。

主持国家自然科学基金青年项目和面上项目，作为主要成员参与国家重点研发计划青年科学家项目，并担任 Journal of Nonparametric Statistics 编委。