报告题目：Large-scale multiple testing without p-values

报告时间：5月17日，星期二，上午10：00—12：00

报告方式：线上

腾讯会议号：184-163-410

报告摘要：

In this talk, I will introduce our recent works about large-scale multiple testing without p-values. Large-scale multiple testing is a common and important problem, with extensive and profound applications in many disciplines. False discovery rate (FDR) is a key concept to maintain the ability to reliably detect true alternatives without excessive false positive results. Currently most FDR control methods are based on p-values. However, in many situations, p-value is not readily available or accurate. For these problems, we propose a very simple and effective solution. The core idea is to use sample splitting to construct a series of symmetric statistics, and utilize the symmetry of statistics to approximate the number of false discoveries. I will illustrate the basic idea through several important examples. For different problems, I will discuss the construction of symmetric statistics and establish corresponding theoretical results. Numerical studies show that our proposed methods are promising alternatives when p-values are not easy to obtain.

报告人简介：

      郭旭博士，现为北京师范大学统计学院副教授，博士生导师。他于2014年获得博士学位。郭旭一直从事回归分析中的复杂统计推断包括模型设定检验和大规模显著性检验等方面的理论和应用研究，在包括统计学和计量经济学国际顶尖期刊JRSSB, JASA，Biometrika和Journal of Econometrics等SCI和SSCI期刊发表高水平论文30篇左右，为包括Econometrica，JASA，Journal of Econometrics，Statistica Sinica等统计学和计量经济学期刊审稿。先后主持国家自然科学基金青年基金和国家自然科学基金面上项目。曾荣获北师大第十一届“最受本科生欢迎的十佳教师”。