报告题目： Generalized likelihood ratio tests based on regularization
报告摘要: In this talk we propose a dimension-reduced generalized likelihood ratio test method based on regularized estimation. The idea of test is to first reduce the dimension of the alternative space in a hypothesis testing problem and then to test the null hypothesis against the dimension-reduced model. We establish the limiting distributions of the proposed test under the null and alternative. It is shown that the proposed test have the limiting power bigger or equal to that of the
traditional likelihood ratio tests and possess oracle properties in hypothesis testing. We apply the proposed test to linear and generalized linear models.
Simulations demonstrate more favorable finite sample performance of our tests over some existing competitive procedures for high dimensional hypothesis testing.
报告人简介：Prof. Jiancheng Jiang （蒋建成）is from
University of North Carolina at Charlotte.
He has published over 50 academic research papers in refereed journals in (bio) statistics and econometrics, including Annals of Statistics, Biometrika, Journal of American Statistical Association, Journal of the Royal Statistical Society, etc.
He is Associate Editor of Statistica Sinica, Artificial Intelligence - AI in Finance, etc.