在线威尼斯赌场澳门176:Matrix Co-completion for Multi-Label Classification with Missing Features and Labels

时间:2018-09-06浏览:87设置

在线威尼斯赌场澳门 第176


报告时间:2018913日(周四)15:30 - 16:30

报告地点:计算机学院4001

报告题目:Matrix Co-completion for Multi-Label Classification with Missing Features and Labels



报告人:徐淼 博士

邀请人: 宫辰 教授

摘要:

We consider a challenging multi-label classification problem where both feature matrix  and label matrix  have missing entries. An existing method concatenated  and  together as   where  is a sigmoid function. Then under the assumption of low-rankness, a matrix completion (MC) method is applied to fill the missing entries. However, there is no theoretical guarantee for the recovery result of this method. In this talk, I will present an upper bound on the recovery error of the method. In deriving the bound, we found that adding another trace norm constraint on recovering  will lead to a guaranteed recovery of the whole matrix. Such phenomenon coincides with Elastic Net where both L norm and L are used for regularization. The practical usefulness of the proposed method is demonstrated through experiments on both synthetic and benchmark data.



讲者介绍:

Dr. Miao Xu is currently a Postdoctoral Researcher at RIKEN Center for Advanced Intelligence Project (RIKEN AIP). Before joining RIKEN AIP, she obtained her Ph.D. degree in computer science from Nanjing University in 2017, under the supervision of Prof. Zhi-Hua Zhou. During her Ph.D. study, she got IBM Ph.D. Fellowship, the Doctoral Enhancement Program of Nanjing University, the Presidential Scholarship of Nanjing University and the Excellent Graduate Student Scholarship. Her research interests include weakly-supervised multi-label learning and matrix completion/recovery. She has published several papers on top conferences including NIPS, ICML, IJCAI, AAAI and KDD.




返回原图
/