Speaker: Yawei Ge, PhD Candidate, Department of Statistics, Iowa State University

Title: Test for High-Dimensional Coefficients with Residual Refitting Regressions

Abstract: Conventional F test fails when the number of variables is larger than the sample size, which is a case of the known “curse of dimensionality”. In high-dimensional linear models, regularization methods are proposed for the estimation while the inference stays challenging. U-statistics is proposed for its loose constraint of the dimensions and the distributions of the considered linear models. We propose a test procedure based on U-statistic and the residual refitting regression which is further a Lasso extension to test a group of high-dimensional target variables while controlling a group of high-dimensional nuisance variables.