报告题目:Numerical Computation for Nonnegative Matrix Factorization

报告人:Chu Delin(NUS, Singapore)

报告时间:2025年6月27日(周五)上午11:00-12:00

报告地点:格物楼 528

报告摘要:

Nonnegative matrix factorization (NF ) is a prominent technique for dat?dimensionality reduction,In this talk,a framework called ARKNLS (AlternatingRank-k Nonnegativity constrained Least Squares)is introduced for computing NMFFirst,a recursive formula for the solution of the rank-k nonnegativity-constrainedleast squares (NS)is established This recursive formula can be used to derive theclosed-form solution for the Rank-k NI S problem for any integer k, As a result, eachsubproblem for an alternating rank-k nonnegative least squares framework can beobtained based on this closed forin solution.This talk is thenfocused on theframework with=3.)which leads to a new algorithm for NMF via the closed-formsolution of the rank-3 NS problem.Furtherorea new strategy that efficientlyovercomes the potential singularity problem in rank-3 NLS within the context ofNMF computation is also presented, Extensive numerical comparisons using real ancsynthetic data sets demonstrate that the proposed algorithm provides state-of-the-arperformance in terms of computational accuracy and cpu time.

报告人简介:

储德林，新加坡国立大学教授，德国“洪堡学者”和日本“JSPS 学者”。先后任职于香港大学，清华大学，德国TUChemnitz(开姆尼斯工业大学)、UniversityoBielefeld(比勒费尔德大学)等国内外知名高校。主要研究领域是科学计算、数值代数及其应用。现为国际顶级期刊SIAM Joummal on Scientific Computing 副主编、SIAM Joural on Matrix Analysis and Applications 副主编、Automatica 副主编、Journal of Computational and Applied Mathematics 编秀、Journal of the FranklinInstitute 客座编秀。已在Mathematics ofComputation、Numerische Mathematik、SIAM Journal on Matrix Analysis and Applications 、SIAM Journal on ScientificComputing、SIAM Joural on Control and Optimization、 SIAM Journal on AppliedDynamical SvstemsJournal of Scientific ComputingTEEE Transactions onPattern Analysis and Machine Intelligence 、 IEEE Transactions on Neural Networksand Learning Systems 等国际知名学术期刊发表论文 100 余篇。