学术报告：A Powerful Test for Independence Using Refitted Cross-Validation
发布时间：2021-04-16 浏览次数：287
报告时间：2021年4月19日下午15:30-16:30
报告地点：雁山校区 理2-204
报告题目：A Powerful Test for Independence Using Refitted Cross-Validation
报告人：周亭攸
报告摘要：Measuring the association between random variables is of fundamental
importance in data science and statistical inference. In this work, we propose a distribution free independence test based on projected distance covariance using refitted cross-validation. The proposed test is consistent against all fixed alternatives, insensitive to the dimension and free of tuning parameters. More importantly, under the null hypothesis, the test can control type-I error rate at the nominal level pretty well and the asymptotic null distribution has an explicit form that does not involve any unknown parameters, which facilitates to compute the empirical p-value efficiently in practice. Under the alternative hypothesis, the test statistic enjoys the oracle property. Extensive Monte Carlo simulations reveal that our test is superior to the competitors and as powerful as if both random variables are univariate.
报告人简介：周亭攸，理学博士，浙江财经大学讲师，硕士生导师。现担任青年统计学家协会理事学术兼职。主要研究领域为高维数据分析的统计理论、方法及应用研究工作，涉及半参数模型，充分降维，复杂数据相依性度量等。主持了国家自科基金青年项目一项，在《Journal of the American Statistical Association》、《Statistica Sinica》等期刊发表学术论文。