报告题目：Kernel-based independence tests for high-dimensional response data

报 告 人：李启寨 研究员

所在单位：中科院数学与系统科学研究院

报告时间：2022年7月11日 星期一 上午9:30-10:30

报告地点：腾讯会议：154-726-320 会议密码：2022

报告摘要： Testing independence between high-dimensional response variable and some covariates is frequently encountered in statistical applications nowadays, and the kernelbased methods have been developed recently. However, the traditional kernel-based methods may suffer from substantial power loss under the situations with moderate to high correlations among responses. In this work, we first propose a set of kernel-based independence tests on the basis of angles between two reproducing kernel Hilbert spaces, and obtain their asymptotical null distributions. Then, we construct two tests including maximal kernel-based independence test (MKIT) and maximin efficient robust test (MERT). Under some regular conditions, we prove that MKIT and MERT asymptotically follow extreme-value type I-Gumbel distribution and normal distribution, respectively. The powers of MKIT and MERT are also investigated. Extensive simulation studies show that MKIT and MERT are more powerful and robust than many existing procedures over a wide range of situations. Applications to heterogeneous stock mice and prostate cancer pathway data ulteriorly demonstrate the performances of proposed methods.

报告人简介: 李启寨，中国科学院数学与系统科学研究院研究员，中国科学院大学教授，美国统计学会会士，国际统计学会推选会员，2001年本科毕业于中国科技大学，2006年博士毕业于中国科学院研究生院；研究方向：生物医学统计等；发表及接收发表SCI论文100余篇。现任中国数学会常务理事、中国现场统计研究会常务理事等。