报告题目：COMRADE: A High Dimensional Conditional Independence Test by Cauchy Combination of Random Projected Smaller Ones

报告时间：2022-04-08 15:10 - 15:50

报告人：尹建新 副教授 中国人民大学

腾讯会议ID：957-794-776

报告入口：https://meeting.tencent.com/dm/5BKN9b4u1B3v

Abstract：Measuring and testing conditional independence are fundamental problems in statistical sciences. Testing for high-dimensional conditional independence is especially challenging. This paper proposed a procedure COMRADE(Cauchy cOMbination of tests after RAnDom projEction) that combines many random projected dimension-reduction based low-dimensional tests to test the conditional independence of two high-dimensional random vectors X and Y given another vector Z. With rather mild regularity conditions, it can be shown that the tail probability of the proposed statistic asymptotically follows a standard Cauchy distribution under null hypothesis, and the consistency under alternative hypotheses of normal distribution. The superior performance of COMRADE compared with many existed methods is demonstrated via Monte Carlo studies. A real example of conditional independence between different gene pathways around breast cancer is explored with this test and some new insight can be obtained.