清华主页 EN
导航菜单

Covariate-shift Robust Adaptive Transfer Learning for High-Dimensional Regression

来源: 06-17

时间:16:00 - 17:30, 2024-06-18

地点:A6-1 ZOOM: 637 734 0280 PW: BIMSA

组织者:Rongling Wu

主讲人:Jingyuan Liu (刘婧媛,厦门大学)

Abstract

The main challenge that sets transfer learning apart from traditional supervised learning is the distribution shift, reflected as the shift between the source and target models and that between the marginal covariate distributions. High-dimensional data introduces unique challenges, such as covariate shifts in the covariate correlation structure and model shifts across individual features in the model. In this work, we tackle model shifts in the presence of covariate shifts in the high-dimensional regression setting. Furthermore, to learn transferable information which may vary across features, we propose an adaptive transfer learning method that can detect and aggregate the feature-wise transferable structures. Non-asymptotic bound is provided for the estimation error of the target model, showing the robustness of the proposed method to high-dimensional covariate shifts.


Speaker Intro

刘婧媛,厦门大学经济学院统计学与数据科学系教授、博士生导师,教育部青年长江学者,厦门大学南强卓越教学名师,厦门大学南强青年拔尖人才A类。美国宾夕法尼亚州立大学统计学博士。科研方面主要从事高维及复杂数据的统计方法、稳定性建模、统计基因学等领域的工作,在JASA,JOE, JBES等国际权威学术期刊发表论文30余篇,担任AOAS等期刊编委,入选福建省杰出青年科研人才计划。

返回顶部
相关文章
  • High-Dimensional Statistical Learning Theory

    Description:High-dimensional statistical learning has become an increasingly important research area. In this course, we will provide theoretical foundations of high-dimensional learning for several widely studied problems with many applications. More specifically, we will review concentration inequalities, VC dimension, metric entropy and statistical implications, consider high-dimensional lin...

  • High-dimensional IV regression for genetical genomics data incorporating network structures

    AbstractGenetical genomics data present promising opportunities for integrating gene expression and genotype information. Lin et al. (2015) proposed an instrumental variables (IV) regression framework to select important genes with high-dimensional genetical genomics data. The IV regression addresses the issue of endogeneity caused by potential correlations between gene expressions and error te...