首 页机构概况机构设置科研人才队伍合作交流研究生教育博士后图书馆创新文化党群园地院重点实验室彭桓武中心信息公开
  学术活动
  您现在的位置:首页 > 学术活动 > Lunch Seminar
Spectral detection of global structures in the noisy data
2016-10-18  【 】【打印】【关闭

Institute of Theoretical Physics

Chinese Academy of Sciences

 Key Laboratory of Theoretical Physics

Lunch Seminar

Title

题目

Spectral detection of global structures in the noisy data
 

Speaker

报告人

张潘研究员
 

Affiliation

所在单位

中国科学院理论物理研究所

Date

日期

10月18日12:00

Venue

地点

Conference Hall 322, ITP/理论物理所322报告厅

 

Abstract

摘要

Detecting global structures, such as clustering and low-rank structures, in the high-dimensional data is a very important problem in machine learning. Spectral methods are popular in this problem because they give a natural and scalable way to reduce the dimensionality of data using eigenvectors or singular vectors of matrices related to the data. However when the data is sparse or noisy, spectral methods usually fail to work due to localization of eigenvectors or singular vectors induced by the sparsity or noise.
In this talk I will introduce a general method that is inspired by localization of wave functions and the matrix perturbation analysis, to solve the localization problem. I will also show applications of our method in several machine learning problems such as similarity clustering, rank estimation and matrix completion.

Reference:
Pan Zhang, Robust Spectral Detection of Global Structures in the Data by Learning a Regularization, Advances in Neural Information Processing Systems 2016, arXiv:1609.02906
IE6.0浏览器,1024X768分辨率 版权所有 © 中国科学院理论物理研究所
地址:北京市海淀区中关村东路55号 邮政编码:100190
京ICP备05002865号】 京公网安备1101080094号