Dr. Ziqi Chen 谌自奇 博士
Department of Statistics East China Normal University
This talk introduces CRT*, a new framework for conditional independence testing that incorporates heterogeneous external and unlabeled data. It uses smooth residual-bootstrap and transfer learning for distribution estimation, and its advantages are validated by simulations and breast cancer genomic data.
Abstract 摘要
The conditional randomization test (CRT) provides a principled approach to conditional independence (CI) testing, guaranteeing exact type-I error control when the true conditional distribution is known. In practice, however, this distribution must be estimated, and estimation errors can inflate type-I errors, while high dimensionality and limited sample sizes can reduce power. Although external and unlabeled data offer the potential to improve CI testing, naive integration that ignores distributional heterogeneity can compromise type-I error control and fail to enhance power. We propose CRT*, a novel framework that robustly integrates external and unlabeled datasets to enhance CI testing in heterogeneous scenarios. CRT* employs smooth residual-bootstrap (SRB) with transfer learning for conditional distribution estimation, combined with adaptive data fusion via an optimal convex combination of test statistics. We theoretically establish that the SRB-based estimator converges to the true conditional distribution in expected total variation distance. Furthermore, even in high-dimensional regimes, CRT* maintains valid type-I error control and achieves strictly higher power than standard CRT without external data. Simulations and RNA-seq breast cancer data analyses demonstrate that CRT* substantially improves power while maintaining type-I error control in heterogeneous settings.
Speaker Biography 讲者简介
Ziqi Chen is currently a research professor and doctoral supervisor at the School of Statistics, East China Normal University. His long-term research interests include causal structure learning, generative models, high-dimensional statistics, and data fusion. He has served as principal investigator on three General Programs of the National Natural Science Foundation of China and one Young Scientists Fund project of the National Natural Science Foundation of China, as a sub-project leader on one Key Program of the National Natural Science Foundation of China, and as a key researcher on the National Key R&D Program and the Applied Mathematics Key Project of Shanghai's "Science and Technology Innovation Action Plan." His research findings have been published in authoritative statistics journals such as JASA, Biometrics, Bernoulli, Statistica Sinica, and the Scandinavian Journal of Statistics, as well as at top artificial intelligence and data mining conferences including NeurIPS, ICLR, KDD, and AAAI.
Topics 涉及主题
Smooth residual-bootstrap for conditional distribution estimation Adaptive data fusion for heterogeneous datasets Power improvement under valid type-I error control