Sample-Optimal Low-Rank Approximation of Distance Matrices

被引:0
|
作者
Indyk, Piotr [1 ]
Vakilian, Ali [1 ]
Wagner, Tal [1 ]
Woodruff, David P. [2 ]
机构
[1] MIT, Cambridge, MA 02139 USA
[2] Carnegie Mellon Univ, Pittsburgh, PA 15213 USA
来源
CONFERENCE ON LEARNING THEORY, VOL 99 | 2019年 / 99卷
基金
美国国家科学基金会;
关键词
Low-rank Approximation; Distance Matrix; ALGORITHMS;
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
A distance matrix A is an element of R-nxm represents all pairwise distances, A(ij) = d(x(i), y(j)), between two point sets x(1), ..., x(n) and y(1) ... y(m) in an arbitrary metric space (Z, d). Such matrices arise in various computational contexts such as learning image manifolds, handwriting recognition, and multi-dimensional unfolding. In this work we study algorithms for low-rank approximation of distance matrices. Recent work by Bakshi and Woodruff (NeurIPS 2018) showed it is possible to compute a rank-k approximation of a distance matrix in time O((n + m)(1+gamma)) center dot poly (k, 1/epsilon), where epsilon > 0 is an error parameter and gamma > 0 is an arbitrarily small constant. Notably, their bound is sublinear in the matrix size, which is unachievable for general matrices. We present an algorithm that is both simpler and more efficient. It reads only O((n + m)k/epsilon) entries of the input matrix, and has a running time of O(n + m) center dot poly (k, 1/epsilon). We complement the sample complexity of our algorithm with a matching lower bound on the number of entries that must be read by any algorithm. We provide experimental results to validate the approximation quality and running time of our algorithm.
引用
收藏
页数:29
相关论文
共 50 条
  • [1] Sublinear Time Low-Rank Approximation of Distance Matrices
    Bakshi, Ainesh
    Woodruff, David P.
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 31 (NIPS 2018), 2018, 31
  • [2] On optimal low-rank approximation of non-negative matrices
    Grussler, Christian
    Rantzer, Anders
    2015 54TH IEEE CONFERENCE ON DECISION AND CONTROL (CDC), 2015, : 5278 - 5283
  • [3] Structured low-rank approximation for nonlinear matrices
    Fazzi, Antonio
    NUMERICAL ALGORITHMS, 2023, 93 (04) : 1561 - 1580
  • [4] The inertia of the symmetric approximation for low-rank matrices
    Casanellas, Marta
    Fernandez-Sanchez, Jesus
    Garrote-Lopez, Marina
    LINEAR & MULTILINEAR ALGEBRA, 2018, 66 (11): : 2349 - 2353
  • [5] Randomized algorithms for the low-rank approximation of matrices
    Liberty, Edo
    Woolfe, Franco
    Martinsson, Per-Gunnar
    Rolchlin, Vladimir
    Tyger, Mark
    PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2007, 104 (51) : 20167 - 20172
  • [6] Adaptive Low-Rank Approximation of Collocation Matrices
    M. Bebendorf
    S. Rjasanow
    Computing, 2003, 70 : 1 - 24
  • [7] Structured low-rank approximation for nonlinear matrices
    Antonio Fazzi
    Numerical Algorithms, 2023, 93 : 1561 - 1580
  • [8] On best uniform approximation by low-rank matrices
    Georgieva, I.
    Hofreither, C.
    LINEAR ALGEBRA AND ITS APPLICATIONS, 2017, 518 : 159 - 176
  • [9] ON A PROBLEM OF WEIGHTED LOW-RANK APPROXIMATION OF MATRICES
    Dutta, Aritra
    Li, Xin
    SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS, 2017, 38 (02) : 530 - 553
  • [10] Adaptive low-rank approximation of collocation matrices
    Bebendorf, M
    Rjasanow, S
    COMPUTING, 2003, 70 (01) : 1 - 24