报告题目：On dimensionality effects in linear discriminant analysis for large dimensional data

报告时间：2016年12月2日下午16:00-16:45

报告地点：3522vip浦京集团504室

报 告 人：王成 博士

Abstract：We study the asymptotic results of linear discriminant analysis (LDA) in large dimensional data where the observation dimension , is of the same order of magnitude as the sample size. Roughly, we know when, LDA is an“good”classifier which means the empirical misclassification error tends to the theoretical one and if, we should pay some price for estimating the means and covariance matrix. In this work, we study the dimensionality effects of the regularized LDA (RLDA). As a special case, the explicit theoretical results about dimensionality effects in LDA is derived. In details, we get the asymptotic distribution of the misclassification error using recent results in random matrix theory. Based on these results, a scale adjusted classifier will be suggested to handle data with un-equal sample sizes and the theoretical results on the tuning parameter is derived under mild conditions.

报告人简介：

王成现为上海交通大学应用数学科学学院 特别研究员。本科、硕士、博士毕业于中国科学技术大学，先后访问了 新加坡国立大学、新加坡南洋理工大学、悉尼科技大学、香港理工大学，并于香港浸会大学从事博士后研究。主要研究方向为 随机矩阵理论及应用、高维数据假设检验、高维数据线性和二次型判别分析；目前正承担，上海市青年科技英才“扬帆计划” 项目、上海市“曙光计划” 项目各一项；已于Computational Statistics & Data Analysis、Journal of Multivariate Analysis等杂志发表10多篇高水平SCI论文。先后担任过The Annals of Statistics、Bernoulli、IEEE Transactions on Signal Processing,、Journal of Applied Statistics、Journal of Multivariate Analysis、Statistica Sinica、Science China等审稿人