題目：A Fast Algorithm to Parallelly Construct a Large System of Structural Equations
主講人：張大保 副教授 （Purdue University）
時間：2016年12月29日上午11:00
地點：主樓418
主講人介紹：
Dr. Dabao Zhang is an associate professor in the Department of Statistics, Purdue University. He received his PhD from Cornell University. Before that, he got his Bachelor and Master degrees from Nankai University and Peking University, respectively. His current researches mainly focus on (1) developing supervised dimension reduction methods which help exploring and visualizing high-dimensional data which has been funded by NSF CAREER Award; (2) building directed graphical models based on structural equations; (3) defining R2 for models beyond (homoscedastic) linear regression models.
內容介紹：
We propose a two-stage penalized least squares method to build large systems of structural equations based on a new view of the classical two-stage least squares method. We show that, with large numbers of endogenous and exogenous variables, the system can be constructed via consistent prediction of a set of surrogate variables at the first stage, and consistent selection of regulatory effects at the second stage. While the consistent prediction at the first stage can be obtained via the ridge regression, the adaptive lasso is employed at the second stage to achieve the consistent selection. The resultant estimates of regulatory effects enjoy the oracle properties. This method is computationally fast and allows for parallel implementation. We demonstrate its effectiveness via simulation studies and real data analysis.
（承辦：管理科學與物流系，科研與學術交流中心）
