题目：走进随机森林---一个投影寻踪方法

报告人：夏应存 教授

讲座时间：2022年12月28日（星期三），15：00

交流平台：腾讯会议号 440-506-292

报告人简介：

 夏应存, 新加坡国立大学统计与数据科学系教授。研究兴趣包括非参数回归，高维数据分析，疾病传播统计建模等。研究成果发表在AOS， JASA， JRSSB, Biometrika, JOE, PNAS等期刊. Nature News等多个学术媒体对其提出的疾病跨域传播模型做了专题报道。 JRSSB， Statistical Science和Statistica Sinica对其论文进行了公开讨论。  夏应存曾在暨南大学工作多年， 荣获国务院侨办颁发的“优秀教师”称号。   

摘要：

In this talk, I will present some of my joint research work with my collaborators on projection pursuit regression (PPR) and random forests (RF). Based on the recent milestone progress in the theory of RF and the greedy algorithms, we first revisit PPR and propose an ensemble approach via "feature bagging" for general nonparametric regression, hereafter referred to as ePPR. Compared to standard RF, ePPR has two main advantages:  1) its theoretical consistency can be proved for more general regression functions as long as they are continuous, and higher consistency rates can be achieved,  and 2) ePPR does not split the samples and thus each term of PPR is estimated using the whole data, making the minimization more efficiently and guaranteeing the smoothness of the estimator. we also design a new random forest based on the oblique decision tree, called ODRF hereafter, to overcome the theoretical problems and deficiency of the standard RF, and show the advantages of ODRF over standard RF in both theoretical consistency and numerical performance.