题目：Prediction and variable selection for high-dimensional misspecified regression models under covariate shift

报告人：银庆刚

会议时间：2025年7月15日(周二) 16: 00

地点：综合楼644会议室

报告人简介：

Ching-Kang Ing is a Chair Professor at National Tsing Hua University, Taiwan. His research focuses on mathematical statistics, model selection for time series, and high-dimensional data analysis. He has earned international recognition for his contributions and has received numerous prestigious honors, including election as an IMS Fellow, the Academic Award from the Ministry of Education, the Outstanding Research Award from the Ministry of Science and Technology, the Outstanding Scholar Award from the Foundation for the Advancement of Outstanding Scholarship, and the Sun Yat-Sen Academic Award from the Sun Yat-Sen Academic and Cultural Foundation.

报告摘要：

This study investigates the Orthogonal Greedy Algorithm (OGA) for variable selection in high-dimensional regression under covariate shift, where the training and test inputs follow different distributions. Such distributional differences–especially in the presence of model misspecification–can substantially degrade the predictive performance of standard OGA. To address this issue, we propose Importance-Weighted OGA

(IWOGA), a modified procedure that reweights training samples using trimmed density ratios between the test and training covariates. A key contribution of this work is to establish that IWOGA achieves a fast convergence rate in test prediction error, even under covariate shift and model misspecification, provided general sparsity conditions hold.