• 统计研究中心
当前位置: 首页> 系列讲座> 正文

密西根大学宋学坤教授: Regularized estimation with L-zero penalty in the causal mediation analysis

光华讲坛——社会名流与企业家论坛第 5754 期

(线上讲座)

主题Regularized estimation with L-zero penalty in the causal mediation analysis

主讲人密西根大学宋学坤教授

主持人统计学院 林华珍教授

时间2020年7月13日(周一)上午10:00-11:00

直播平台及会议IDZOOM997 7137 1776

主办单位:统计研究中心、数据科学与商业智能联合实验室和统计学院 科研处

主讲人简介:

Dr. Song is Professor of Biostatistics, and Associate Chair of Research, at the Department of Biostatistics, School of Public Health in the University of Michigan, Ann Arbor, since January, 2008. He received his PhD in Statistics from the University of British Columbia, Vancouver, Canada in 1996. He has published over 170 peer-reviewed papers. Dr. Song's research interests include distributed inference, high-dimensional data analysis, longitudinal data analysis, missing data problems, spatiotemporal modeling, and statistical methods in precision health. He is ASA Fellow and Elected Member of the International Statistical Institute. Dr. Song now serves as Associate Editor of the Journal of American Statistical Association, the Canadian Journal of Statistics, and the Journal of Multivariate Analysis.

宋学坤,美国密西根大学安娜堡分校公共卫生学院生物统计系教授和主管科研副系主任。1996年从加拿大温哥华英属哥伦比亚大学取得统计学博士学位。已发表论文170余篇。主要研究方向包括分布式推理、高维数据分析、纵向数据分析、缺失数据问题、时空建模和精确健康统计方法。他是ASA FellowISIElected Member。现任统计学国际期刊JASA the Canadian Journal of Statistics the Journal of Multivariate AnalysisAssociate Editor。详情请见其个人主页:https://sph.umich.edu/faculty-profiles/song-peter.html


内容提要:

In this talk I will introduce an L0-regularized estimation in the causal mediation analysis. This new approach provides a hopeful solution to one of the current open problems concerning hypothesis testing for a causal mediation pathway. My discussions will be based on the framework of structural equation models, a class of the simplest DAGs that present the causal relationships. In particular, I will demonstrate the superb computing power of our new mixed integer programming solver with a novel envelop search for an exact solution through computationally fast upper and lower bounds. The resulting L0 estimator has been rigorously justified to have key theoretical guarantees.

在本报告中,主讲人将介绍因果中介分析中的L0-regularized估计。这种新方法为因果中介途径的假设检验问题提供了一个有希望的解决方案。 本报告的讨论将基于结构方程模型的框架。结构方程模型是呈现因果关系的一类最简单的有向无环图(DAG)模型。特别地,该论文将通过新颖的压缩挤压搜索算法来展现新的混合整数优化器的计算能力,该优化搜索是通过快速估算有约束的非凸目标函数的上下边界来寻找精确解。 得出的L0估计量已被严格证明具有关键的理论保证。


上一条:香港科技大学杨灿副教授: MR-APSS: a unified approach to Mendelian Randomization accounting for pleiotropy, sample overlap and selection bias using the genome-wide summary data

下一条:加州大学河滨分校李业华教授: Spatially Dependent Functional Data: Covariance Estimation, Principal Component Analysis, and Kriging