A SINGULAR VALUE THRESHOLDING ALGORITHM FOR MATRIX COMPLETION

被引:4336
作者
Cai, Jian-Feng [1 ]
Candes, Emmanuel J. [2 ]
Shen, Zuowei [3 ]
机构
[1] Univ Calif Los Angeles, Dept Math, Los Angeles, CA 90095 USA
[2] CALTECH, Pasadena, CA 91125 USA
[3] Natl Univ Singapore, Dept Math, Singapore 117543, Singapore
基金
美国国家科学基金会;
关键词
nuclear norm minimization; matrix completion; singular value thresholding; Lagrange dual function; Uzawa's algorithm; linearized Bregman iteration; LINEARIZED BREGMAN ITERATIONS; INVERSE PROBLEMS; SIGNAL RECOVERY; IMAGE; CONVERGENCE; APPROXIMATION; L(1)-MINIMIZATION; DECOMPOSITION; RESTORATION; SPARSITY;
D O I
10.1137/080738970
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
This paper introduces a novel algorithm to approximate the matrix with minimum nuclear norm among all matrices obeying a set of convex constraints. This problem may be understood as the convex relaxation of a rank minimization problem and arises in many important applications as in the task of recovering a large matrix from a small subset of its entries (the famous Netflix problem). Off-the-shelf algorithms such as interior point methods are not directly amenable to large problems of this kind with over a million unknown entries. This paper develops a simple first-order and easy-to-implement algorithm that is extremely efficient at addressing problems in which the optimal solution has low rank. The algorithm is iterative, produces a sequence of matrices {X-k, Y-k}, and at each step mainly performs a soft-thresholding operation on the singular values of the matrix Y-k. There are two remarkable features making this attractive for low-rank matrix completion problems. The first is that the soft-thresholding operation is applied to a sparse matrix; the second is that the rank of the iterates {X-k} is empirically nondecreasing. Both these facts allow the algorithm to make use of very minimal storage space and keep the computational cost of each iteration low. On the theoretical side, we provide a convergence analysis showing that the sequence of iterates converges. On the practical side, we provide numerical examples in which 1, 000 x 1, 000 matrices are recovered in less than a minute on a modest desktop computer. We also demonstrate that our approach is amenable to very large scale problems by recovering matrices of rank about 10 with nearly a billion unknowns from just about 0.4% of their sampled entries. Our methods are connected with the recent literature on linearized Bregman iterations for l(1) minimization, and we develop a framework in which one can understand these algorithms in terms of well-known Lagrange multiplier algorithms.
引用
收藏
页码:1956 / 1982
页数:27
相关论文
共 68 条
  • [41] CONVEX PROGRAMMING IN HILBERT SPACE
    GOLDSTEIN, AA
    [J]. BULLETIN OF THE AMERICAN MATHEMATICAL SOCIETY, 1964, 70 (05) : 709 - &
  • [42] The Split Bregman Method for L1-Regularized Problems
    Goldstein, Tom
    Osher, Stanley
    [J]. SIAM JOURNAL ON IMAGING SCIENCES, 2009, 2 (02): : 323 - 343
  • [43] FIXED-POINT CONTINUATION FOR l1-MINIMIZATION: METHODOLOGY AND CONVERGENCE
    Hale, Elaine T.
    Yin, Wotao
    Zhang, Yin
    [J]. SIAM JOURNAL ON OPTIMIZATION, 2008, 19 (03) : 1107 - 1130
  • [44] Hiriart-Urruty J. -B, 1993, GRUNDLEHREN MATH WIS, V305
  • [45] Iusem AN, 2003, COMPUT APPL MATH, V22, P37
  • [46] Larsen R., PROPACK - Software for large and sparse SVD calculations
  • [47] Levitin E. S., 1966, USSR Comput. Math. Math. Phys., V6, P1, DOI DOI 10.1016/0041-5553(66)90114-5
  • [48] The mathematics of eigenvalue optimization
    Lewis, AS
    [J]. MATHEMATICAL PROGRAMMING, 2003, 97 (1-2) : 155 - 176
  • [49] Randomized algorithms for the low-rank approximation of matrices
    Liberty, Edo
    Woolfe, Franco
    Martinsson, Per-Gunnar
    Rolchlin, Vladimir
    Tyger, Mark
    [J]. PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2007, 104 (51) : 20167 - 20172
  • [50] Solving a variational image restoration model which involves L∞ constraints
    Lintner, S
    Malgouyres, F
    [J]. INVERSE PROBLEMS, 2004, 20 (03) : 815 - 831