搜索
报告题目：A Domain Adaptation Framework for Causal Effect Estimation Under Distribution Shift
报 告 人：王文武，曲阜师范大学
报告摘要：Estimating causal effects in real-world observational studies is often hindered by the limited availability of labeled data in the target domain, making it crucial to utilize abundant information from related source domains. However, differences in data distributions can substantially reduce the effectiveness of transfer-based methods, and these distribution shifts introduce distinct challenges for causal inference. To address this, we propose Domain Adaptation with Transfer Causal Learning (DA-TCL), a unified framework integrating domain adaptation into the L1-regularized TCL model. DATCL employs kernel mean matching and balanced distribution adaptation for source-target distribution alignment, and features two adaptation strategies: unified adaptation and grouped adaptation. The adapted data then estimates the Average Treatment Effect using L1-TCL with neural network nuisance models. Extensive experiments on synthetic data (with varying structural models and shift severity) and the IHDP benchmark (with simulated covariate shifts) demonstrate DA-TCL’s superior accuracy under domain shifts. Notably, on the IHDP benchmark, our method achieves a 54.76% error reduction compared to the L1-TCL baseline.
报告人简介：王文武，曲阜师范大学统计与数据科学学院教授、博士生导师；校统计学研究所副所长，院研究生工作负责人；香港大学博士后研究员，香港浸会大学访问学者。研究兴趣包含：非参数统计、稳健统计、机器学习与人工智能、因果效应评估等。目前，主持教育部人文社会科学研究一般项目1项、山东省自然科学基金面上项目1项；主持完成国家自然科学基金面上项目1项、国家统计局重点项目1项。已发表SCI/SSCI/CSSCI论文20余篇，包括：Journal of Machine Learning Research、Knowledge-Based Systems、Statistics in Medicine、TEST、Psychometrika、Structural Equation Modeling、Journal of Hazardous Materials、《统计研究》等学术期刊。
报告时间：2025年11月19日 15:30-16:30
报告地点：文渊楼B536教室
主办单位：数学与统计学院
学院宣传片
