报告题目：SIMEX-Assisted Auxiliary Learning under Covariate and Model Shifts

报告人： 林路教授 （山东大学）

报告时间：2026年5月27日下午 15:30-16:30

报告地点：学院会议室109

报告摘要: Motivated by the dual challenges of covariate and model shifts in auxiliary learning, we investigate a setting where auxiliary covariates exhibit mean shifts and variance inflation relative to the primary data, and the estimating equations differ across tasks under unknown parametric relationships. Such heterogeneity severely undermines conventional methods that assume homogeneous structures across auxiliary and primary tasks. We address these challenges by integrating two key innovations: (1) applying simulation extrapolation (SIMEX) on the complex plane to correct for bias induced by covariate shift, and (2) incorporating transfer learning to dynamically accommodate model shift arising from latent parameter discrepancies in heterogeneous estimating equations. This approach significantly extends beyond conventional auxiliary learning paradigms. The proposed framework achieves robustness to varying degrees of covariate shift, as well as adaptability to model shift and parameter discrepancy across tasks. Importantly, our approach avoids complicated procedures such as density ratio estimation and importance weighting, remaining highly tractable. We establish rigorous consistency and asymptotic normality of the resulting estimator, and demonstrate its superior empirical performance through extensive simulation studies and real data analysis.

报告人简介：林路是山东大学中泰证券金融研究院教授、博士生导师，第一和第二届教育部应用统计专业硕士教育指导委员会成员，山东省政府参事。从事大数据、统计学习、高维统计、非参数和半参数统计以及金融统计等方的研究，在国内外统计学、机器学习和相关应用学科顶级期刊和重要期刊(包括Ann. Statist., JMLR, Stat. Comput.和中国科学) 发表研究论文130余篇；多个金融策略资政报告得到省长的正面批示；主持过多项国家自然科学基金课题、全国统计科学研究重大项目、教育部新文科课题、山东省自然科学基金重点项目等；获得国家统计局颁发的全国统计优秀研究成果一等和二等奖，山东省优秀教学成果一等奖(均排名第一)。