新加坡南洋理工大学陶大成博士将于5月22日下午来我系做学术报告，欢迎感兴趣的师生踊跃参与。

时间：5月22日下午3点

地点：电信系会议室（南一楼三楼东头）

题目：Discriminative Subspace Selection: Problems, Solutions and Applications（判决子空间选择：问题，方法及应用）

报告人：陶大成博士

报告摘要：

Discriminative subspace selection (DSS) is a powerful tool in machine learning. The past decades has witnessed the effectiveness and the efficiency of DSS for subsequent classification and data visualization. In this talk, we start from the most conventional DSS algorithm, Fisher's linear discriminant analysis (FLDA). Afterwards, we enumerate FLDA's problems, i.e., the heteroscedastic problem, the multimodal problem, the class

separation problem, the under sampled problem, the data nonlinearlity, and the sample selection bias. It is worth emphasizing that all DSS algorithms share some of these problems, so it is essential to find general solutions to them. In particular, we recently present five frameworks, i.e., the general averaged divergence analysis, the general tensor discriminant analysis, the patch alignment framework, the ensemble manifold regularization, and the Bregman divergence based regularization, to address or at least reduce the aforementioned problems. Thorough experimental evidence on various real datasets and artificial datasets suggests these frameworks are effective to deal with popular DSS problems.

报告人简介：

陶大成博士在中国科技大学取得学士学位，在香港中文大学取得硕士学位，在英国伦敦大学取得哲学博士学位。现南洋理工大学计算机工程系助理教授。同时，他还是西安电子科技大学和武汉大学客座教授。

陶大成博士在国际顶级期刊（TPAMI, TKDE, TMM,TCSVT等）和顶级会议（CVPR,ICIP, ICDM 等.）上共发表论文80余篇。同时，他还是Neurocomputing (Elsevier)，Computational Statistics & Data Analysis (Elsevier) 等国际杂志的副主编。他还担任了Computer Vision and

Image Understanding (Elsevier)，Pattern Recognition (Elsevier)，Signal Processing (Elsevier)等期刊杂志的客座编委。