科学研究
打造高水平科技创新平台和一流科研团队！
学术活动
数学与统计学院"21世纪学科前沿"系列学术报告预告
Second-order Least Squares Method for High-dimensional Variable Selection
编辑： 数学学院 董学敏 时间：2015-06-01
报告题目：Second-order Least Squares Method for High-dimensional Variable Selection
报告时间：2015年6月2日下午3:00-4:00
报告地点：良乡1-208
报告人：Professor Liqun Wang, Department of Statistics, University of Manitoba, Canada
摘要：High-dimensional variable selection problems arise in many scientific fields, including genome and health science, economics and finance, astronomy and physics, signal processing and imaging. In statistics, various regularization methods have been studied based on either likelihood or least squares principles. In this talk, I will propose a regularized second order least squares method for variable selection in linear or nonlinear regression models. This method is based the first two conditional moments of the response variable given on the predictor variables. It is asymptotically more efficient than the ordinary least squares method when the regression error has nonzero third moment. Consequently the new method is more robust against asymmetric error distributions. I will demonstrate the effectiveness of this method through Monte Carlo simulation studies. A real data application will be presented to further illustrate the method.
报告时间：2015年6月2日下午3:00-4:00
报告地点：良乡1-208
报告人：Professor Liqun Wang, Department of Statistics, University of Manitoba, Canada
摘要：High-dimensional variable selection problems arise in many scientific fields, including genome and health science, economics and finance, astronomy and physics, signal processing and imaging. In statistics, various regularization methods have been studied based on either likelihood or least squares principles. In this talk, I will propose a regularized second order least squares method for variable selection in linear or nonlinear regression models. This method is based the first two conditional moments of the response variable given on the predictor variables. It is asymptotically more efficient than the ordinary least squares method when the regression error has nonzero third moment. Consequently the new method is more robust against asymmetric error distributions. I will demonstrate the effectiveness of this method through Monte Carlo simulation studies. A real data application will be presented to further illustrate the method.