| 主题： |
On the ultrahigh dimensional linear discriminant analysis problem with a diverging number of classes |
| 类型： |
学术报告
|
| 主办方： |
金融研究院 |
| 报告人： |
王汉生教授 |
| 日期： |
3月28日（周四）下午3:30-4:30 |
| 地点： |
知新楼B-1248 |
| 内容： |
This paper is concerned with the problem of variable screening for Fisher's linear discriminant analysis with a diverging number of classes and an ultrahigh dimensional predictor. In the presence of a diverging number of classes, the total number of relevant features may go to infinity at a rate faster than usual. This makes the related statistical problem much more challenging than the conventional one with a fixed number of classes. To solve the problem, we propose here a novel pairwise sure independence screening method for the linear discriminant analysis with an ultrahigh dimensional predictor. The proposed procedure is directly applicable for the situation with a finite number of classes and with a diverging number of classes. We further prove that the proposed method enjoys the strong screening consistency property. Simulation studies are conducted to assess the finite sample performance of the proposed procedure. We also demonstrate the proposed methodology via an empirical analysis of a real-life example on hand-written Chinese character recognition.
王汉生教授的个人主页：
http://hansheng.gsm.pku.edu.cn， http://t.sina.com.cn/HanshengWang
|
