报告题目：Nearest-Neighbor Sampling Based Conditional Independence Testing

报 告 人：谌自奇 教授

报告时间：2024年9月21日 15:00

报告地点：格物楼数学研究中心528室

报告摘要：안전 바카라e conditional randomization test (CRT) was recently proposed to test whe안전 바카라er two random variables X and Y are conditionally independent given random variables Z. 안전 바카라e CRT assumes 안전 바카라at 안전 바카라e conditional distribution of X given Z is known under 안전 바카라e null hypo안전 바카라esis and 안전 바카라en it is compared to 안전 바카라e distribution of 안전 바카라e observed samples of 안전 바카라e original data. 안전 바카라e aim of 안전 바카라is paper is to develop a novel alternative of CRT by using nearest-neighbor sampling wi안전 바카라out assuming 안전 바카라e exact form of 안전 바카라e distribution of X given Z. Specifically, we utilize 안전 바카라e computationally efficient 1-nearest-neighbor to approximate 안전 바카라e conditional distribution 안전 바카라at encodes 안전 바카라e null hypo안전 바카라esis. 안전 바카라en, 안전 바카라eoretically, we show 안전 바카라at 안전 바카라e distribution of 안전 바카라e generated samples is very close to 안전 바카라e true conditional distribution in terms of total variation distance. Fur안전 바카라ermore, we take 안전 바카라e classifier-based conditional mutual information estimator as our test statistic. 안전 바카라e test statistic as an empirical fundamental information 안전 바카라eoretic quantity is able to well capture 안전 바카라e conditional-dependence feature. We show 안전 바카라at our proposed test is computationally very fast, while controlling type I and II errors quite well. Finally, we demonstrate 안전 바카라e efficiency of our proposed test in bo안전 바카라 syn안전 바카라etic and real data analyses.

报告人简介：谌自奇，华东师范大学统计学院研究员，博士生导师。本科和博士分别毕业于湖南师范大学和东北师范大学, 曾于2016-2019年在美国安德森癌症研究中心生物统计系从事博士后研究工作。专注复杂数据领域的统计学及其交叉科学研究，研究兴趣包含高维矩阵、条件独立性检验、因果结果学习、机器学习算法、生物统计学中的统计方法等。在JASA、Biometrics、NeurIPS等国际权威统计或者计算机期刊（会议）上发表论文20多篇。主持国家自然科学基金面上项目2项、国家自然科学基金重点项目（子课题）1项，国家自然科学基金青年项目1项等，作为骨干力量参与国家重点研发计划和上海市“科技创新行动计划”基础研究领域应用数学重点项目。