报告题目：Energy-based adaptive CUR matrix decomposition

会议时间：2022/11/26 14:30-17:00 (GMT+08:00)中国标准时间-北京

腾讯会议：391-415-264

会议密码：654321

Abstract：CUR decompositions are interpretable dataanalysis tools that express a data matrix interms of a small number of actual columnsand/or actual rows of the data matrix. Onebottleneck of existing relative-error CUR algorithms lies on high computational complexityfor computing important sampling probabilities. In this paper, we provide a simple yet effective framework that considers energy-basedsampling algorithm. On one hand, we providea intuitive and fast relative-error sampling algorithm for column selection problem. On theother hand, by combining the relative-errorsampling algorithm with adaptive samplingalgorithm we provide a novel CUR matrixapproximation algorithms which is referred toas energy-based adaptive sampling algorithm.The sampling algorithm is the first adaptiverelative-error CUR decomposition in the coherent sense. Specially, in each stage of ouralgorithm, we sample columns or rows fromdata matrix using sampling probabilities thatare directly proportional to Euclidean normsof the columns or rows of the original data andresidual matrix, respectively. Our empiricalresults exactly indicate that the new adaptivesampling algorithm typically achieves a goodbalance between computational complexityand approximate accuracy.

简介：

徐礼文，男，博士，北方工业大学统计学科责任教授。中国现场统计研究会理事, 全国工业统计学教学研究会理事, 全国工业统计学教学研究会青年统计学家协会常务理事，北京市博士后杰出英才，北京市属高校青年拔尖人才，美国佐治亚大学统计学系/大数据实验室访问学者。现主要从事基于深度学习的大数据人工智能、图神经网络、强化学习等研究。主持国家自然科学基金2项、国家社科基金1项、中国博士后科学基金2项和北京市自然科学基金1项；在国内外发表论文50余篇，出版著作3部。