Streaming First-Order Optimization Methods With Momentum for Dominant Eigenspace Estimation

被引:0
作者
Zhou, Siyun [1 ]
机构
[1] Univ Elect Sci & Technol China, Sch Math Sci, Chengdu, Peoples R China
关键词
first-order methods; momentum; streaming principal eigenspace recovery; ONLINE PCA; CONVERGENCE; ALGORITHMS;
D O I
10.1002/nla.2595
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
In this article, we consider the dominant eigenspace recovery for PCA in the streaming scenario, and design several new algorithms for this problem by incorporating various momentum schemes into two classical gradient based streaming eigensolvers, which emerges as a frequently-used trick in improving gradient type methods. Theoretically, we establish the convergence guarantees of the proposed first-order momentum methods, and prove their decreasing variance under favorable conditions. Extensive numerical experiments are conducted to corroborate the result that momentum does yield variance reduction, and also highlight the advantages of lowering the stepsize sensitivity. In addition, our methods with appropriately chosen momentum yield a dramatically enhanced performance in terms of both the convergence speed and accuracy.
引用
收藏
页数:21
相关论文
共 50 条
[21]   PEP<sc>it</sc>: computer-assisted worst-case analyses of first-order optimization methods in Python']Python [J].
Goujaud, Baptiste ;
Moucer, Celine ;
Glineur, Francois ;
Hendrickx, Julien M. ;
Taylor, Adrien B. ;
Dieuleveut, Aymeric .
MATHEMATICAL PROGRAMMING COMPUTATION, 2024, 16 (03) :337-367
[22]   FOM - a MATLAB toolbox of first-order methods for solving convex optimization problems [J].
Beck, Amir ;
Guttmann-Beck, Nili .
OPTIMIZATION METHODS & SOFTWARE, 2019, 34 (01) :172-193
[23]   ON THE EFFECT OF PERTURBATIONS IN FIRST-ORDER OPTIMIZATION METHODS WITH INERTIA AND HESSIAN DRIVEN DAMPING [J].
Attouch, Hedy ;
Fadili, Jalal ;
Kungurtsev, Vyacheslav .
EVOLUTION EQUATIONS AND CONTROL THEORY, 2023, 12 (01) :71-117
[24]   A Unified Analysis of First-Order Methods for Smooth Games via Integral Quadratic Constraints [J].
Zhan, Guodong ;
Bao, Xuchan ;
Lessard, Laurent ;
Grosse, Roger .
JOURNAL OF MACHINE LEARNING RESEARCH, 2021, 22
[25]   A Study of Condition Numbers for First-Order Optimization [J].
Guille-Escuret, Charles ;
Goujaud, Baptiste ;
Girotti, Manuela ;
Mitliagkas, Ioannis .
24TH INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND STATISTICS (AISTATS), 2021, 130
[26]   First-Order Methods for Nonconvex Quadratic Minimization [J].
Carmon, Yair ;
Duchi, John C. .
SIAM REVIEW, 2020, 62 (02) :395-436
[27]   Perturbed Fenchel duality and first-order methods [J].
Gutman, David H. ;
Pena, Javier F. .
MATHEMATICAL PROGRAMMING, 2023, 198 (01) :443-469
[28]   Transient growth of accelerated first-order methods [J].
Samuelson, Samantha ;
Mohammadi, Hesameddin ;
Jovanovic, Mihailo R. .
2020 AMERICAN CONTROL CONFERENCE (ACC), 2020, :2858-2863
[29]   An acceleration procedure for optimal first-order methods [J].
Baes, Michel ;
Buergisser, Michael .
OPTIMIZATION METHODS & SOFTWARE, 2014, 29 (03) :610-628
[30]   Distributed Learning Systems with First-Order Methods [J].
Liu, Ji ;
Zhang, Ce .
FOUNDATIONS AND TRENDS IN DATABASES, 2020, 9 (01) :1-100