First Efficient Convergence for Streaming k-PCA: a Global, Gap-Free, and Near-Optimal Rate

被引:50
作者
Allen-Zhu, Zeyuan [1 ]
Li, Yuanzhi [2 ]
机构
[1] Microsoft Res, Cambridge, MA 02142 USA
[2] Princeton Univ, Princeton, NJ 08544 USA
来源
2017 IEEE 58TH ANNUAL SYMPOSIUM ON FOUNDATIONS OF COMPUTER SCIENCE (FOCS) | 2017年
关键词
principal component analysis; streaming algorithm; online algorithm; global convergence; stochastic optimization; convergence; optimal algorithm; nonconvex optimization;
D O I
10.1109/FOCS.2017.51
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
We study streaming principal component analysis (PCA), that is to find, in Omicron(dk) space, the top k eigenvectors of a d x d hidden matrix Sigma with online vectors drawn from covariance matrix Sigma. We provide global convergence for Oja's algorithm which is popularly used in practice but lacks theoretical understanding for k > 1. We also provide a modified variant Oja(++) that runs even faster than Oja's. Our results match the information theoretic lower bound in terms of dependency on error, on eigengap, on rank k, and on dimension d, up to poly-log factors. In addition, our convergence rate can be made gap-free, that is proportional to the approximation error and independent of the eigengap. In contrast, for general rank k, before our work (1) it was open to design any algorithm with efficient global convergence rate; and (2) it was open to design any algorithm with (even local) gap-free convergence rate in Omicron(dk) space.
引用
收藏
页码:487 / 492
页数:6
相关论文
共 21 条
  • [1] [Anonymous], 2016, ICML
  • [2] [Anonymous], 2016, NIPS
  • [3] [Anonymous], ICML 17
  • [4] [Anonymous], 2016, ARXIV E PRINTS
  • [5] [Anonymous], 2013, ADV NEURAL INFORM PR
  • [6] [Anonymous], 2016, NIPS
  • [7] Balcan M. F., 2016, C LEARN THEOR, P284
  • [8] Direct 3D ultrasound fusion for transesophageal echocardiography
    Mao, Zhehua
    Zhao, Liang
    Huang, Shoudong
    Fan, Yiting
    Lee, Alex Pui-Wai
    [J]. COMPUTERS IN BIOLOGY AND MEDICINE, 2021, 134
  • [9] De Sa C, 2015, PR MACH LEARN RES, V37, P2332
  • [10] Concentration Inequalities and Martingale Inequalities: A Survey
    Fan Chung
    Lu, Linyuan
    [J]. INTERNET MATHEMATICS, 2006, 3 (01) : 79 - 127