schliessen

Filtern

 

Bibliotheken

Fixed point and Bregman iterative methods for matrix rank minimization

The linearly constrained matrix rank minimization problem is widely applicable in many fields such as control, signal processing and system identification. The tightest convex relaxation of this problem is the linearly constrained nuclear norm minimization. Although the latter can be cast as a semid... Full description

Journal Title: Mathematical programming 2009-09-23, Vol.128 (1-2), p.321-353
Main Author: Ma, Shiqian
Other Authors: Goldfarb, Donald , Chen, Lifeng
Format: Electronic Article Electronic Article
Language: English
Subjects:
Publisher: Berlin/Heidelberg: Springer-Verlag
ID: ISSN: 0025-5610
Link: http://pascal-francis.inist.fr/vibad/index.php?action=getRecordDetail&idt=24208792
Zum Text:
SendSend as email Add to Book BagAdd to Book Bag
Staff View
recordid: cdi_proquest_miscellaneous_889387590
title: Fixed point and Bregman iterative methods for matrix rank minimization
format: Article
creator:
  • Ma, Shiqian
  • Goldfarb, Donald
  • Chen, Lifeng
subjects:
  • Algorithms
  • Analysis
  • Applied sciences
  • Approximation
  • Calculus of Variations and Optimal Control
  • Optimization
  • Combinatorics
  • Constraints
  • DNA microarrays
  • Exact sciences and technology
  • Full Length Paper
  • Images
  • Iterative methods
  • Management science
  • Mathematical and Computational Physics
  • Mathematical Methods in Physics
  • Mathematical programming
  • Mathematics
  • Mathematics and Statistics
  • Mathematics of Computing
  • Matrices
  • Matrix
  • Methods
  • Minimization
  • Norms
  • Numerical Analysis
  • Operational research and scientific management
  • Operational research. Management science
  • Optimization
  • Programming
  • Semidefinite programming
  • Signal processing
  • Studies
  • Theoretical
ispartof: Mathematical programming, 2009-09-23, Vol.128 (1-2), p.321-353
description: The linearly constrained matrix rank minimization problem is widely applicable in many fields such as control, signal processing and system identification. The tightest convex relaxation of this problem is the linearly constrained nuclear norm minimization. Although the latter can be cast as a semidefinite programming problem, such an approach is computationally expensive to solve when the matrices are large. In this paper, we propose fixed point and Bregman iterative algorithms for solving the nuclear norm minimization problem and prove convergence of the first of these algorithms. By using a homotopy approach together with an approximate singular value decomposition procedure, we get a very fast, robust and powerful algorithm, which we call FPCA (Fixed Point Continuation with Approximate SVD), that can solve very large matrix rank minimization problems (the code can be downloaded from http://www.columbia.edu/~sm2756/FPCA.htm for non-commercial use). Our numerical results on randomly generated and real matrix completion problems demonstrate that this algorithm is much faster and provides much better recoverability than semidefinite programming solvers such as SDPT3. For example, our algorithm can recover 1000 × 1000 matrices of rank 50 with a relative error of 10 −5 in about 3 min by sampling only 20% of the elements. We know of no other method that achieves as good recoverability. Numerical experiments on online recommendation, DNA microarray data set and image inpainting problems demonstrate the effectiveness of our algorithms.
language: eng
source:
identifier: ISSN: 0025-5610
fulltext: no_fulltext
issn:
  • 0025-5610
  • 1436-4646
url: Link


@attributes
NO1
SEARCH_ENGINEprimo_central_multiple_fe
SEARCH_ENGINE_TYPEPrimo Central Search Engine
RANK2.752566
LOCALfalse
PrimoNMBib
record
control
sourceidgale_proqu
recordidTN_cdi_proquest_miscellaneous_889387590
sourceformatXML
sourcesystemPC
galeidA256364398
sourcerecordidA256364398
originalsourceidFETCH-LOGICAL-1463t-2a366b86bcf55584d49303f70d7e360f51b4eb8ebb58ad851474254cd89b956b0
addsrcrecordideNp9kU9rFTEUxYMo-Kx-AHdBEFdTb_5OsqzFp0LBja5DJpM8U2eSZ5In1U9v6hSFgpJFIPzOybn3IPScwDkBGF9XAgTGAUAPwEAO4gHaEc7kwCWXD9EOgIpBSAKP0ZNarwGAMKV2aL-PN37GxxxTwzbN-E3xh9UmHJsvtsXvHq--fclzxSEXvNpW4g0uNn3Fa0xxjT87lNNT9CjYpfpnd_cZ-rx_--ny_XD18d2Hy4urgXDJ2kAtk3JScnJBCKH4zDUDFkaYR88kBEEm7iflp0koOytB-Mip4G5WetJCTnCGXm2-x5K_nXxtZo3V-WWxyedTNUpppkahoZMv7pHX-VRSD2eUlLyPT3WHzjfoYBdvYgq5Fev6mf0aXU4-xP5-QYVkkjOtuoBsAldyrcUHcyxxteWHIWBuizBbEaYXYW6LMKJrXt4lsdXZJfTluVj_CCmnoEZNOzfe83ax_d5uDxWX__5AN2Xtpungy99Z_y36Bc8Np7s
sourcetypeAggregation Database
isCDItrue
recordtypearticle
pqid866401329
display
typearticle
titleFixed point and Bregman iterative methods for matrix rank minimization
creatorMa, Shiqian ; Goldfarb, Donald ; Chen, Lifeng
creatorcontribMa, Shiqian ; Goldfarb, Donald ; Chen, Lifeng
descriptionThe linearly constrained matrix rank minimization problem is widely applicable in many fields such as control, signal processing and system identification. The tightest convex relaxation of this problem is the linearly constrained nuclear norm minimization. Although the latter can be cast as a semidefinite programming problem, such an approach is computationally expensive to solve when the matrices are large. In this paper, we propose fixed point and Bregman iterative algorithms for solving the nuclear norm minimization problem and prove convergence of the first of these algorithms. By using a homotopy approach together with an approximate singular value decomposition procedure, we get a very fast, robust and powerful algorithm, which we call FPCA (Fixed Point Continuation with Approximate SVD), that can solve very large matrix rank minimization problems (the code can be downloaded from http://www.columbia.edu/~sm2756/FPCA.htm for non-commercial use). Our numerical results on randomly generated and real matrix completion problems demonstrate that this algorithm is much faster and provides much better recoverability than semidefinite programming solvers such as SDPT3. For example, our algorithm can recover 1000 × 1000 matrices of rank 50 with a relative error of 10 −5 in about 3 min by sampling only 20% of the elements. We know of no other method that achieves as good recoverability. Numerical experiments on online recommendation, DNA microarray data set and image inpainting problems demonstrate the effectiveness of our algorithms.
identifier
0ISSN: 0025-5610
1EISSN: 1436-4646
2DOI: 10.1007/s10107-009-0306-5
3CODEN: MHPGA4
languageeng
publisherBerlin/Heidelberg: Springer-Verlag
subjectAlgorithms ; Analysis ; Applied sciences ; Approximation ; Calculus of Variations and Optimal Control; Optimization ; Combinatorics ; Constraints ; DNA microarrays ; Exact sciences and technology ; Full Length Paper ; Images ; Iterative methods ; Management science ; Mathematical and Computational Physics ; Mathematical Methods in Physics ; Mathematical programming ; Mathematics ; Mathematics and Statistics ; Mathematics of Computing ; Matrices ; Matrix ; Methods ; Minimization ; Norms ; Numerical Analysis ; Operational research and scientific management ; Operational research. Management science ; Optimization ; Programming ; Semidefinite programming ; Signal processing ; Studies ; Theoretical
ispartofMathematical programming, 2009-09-23, Vol.128 (1-2), p.321-353
rights
0Springer and Mathematical Programming Society 2009
12015 INIST-CNRS
2COPYRIGHT 2011 Springer
3Springer and Mathematical Optimization Society 2011
lds50peer_reviewed
citedbyFETCH-LOGICAL-1463t-2a366b86bcf55584d49303f70d7e360f51b4eb8ebb58ad851474254cd89b956b0
citesFETCH-LOGICAL-1463t-2a366b86bcf55584d49303f70d7e360f51b4eb8ebb58ad851474254cd89b956b0
links
openurl$$Topenurl_article
thumbnail$$Usyndetics_thumb_exl
backlink$$Uhttp://pascal-francis.inist.fr/vibad/index.php?action=getRecordDetail&idt=24208792$$DView record in Pascal Francis
search
creatorcontrib
0Ma, Shiqian
1Goldfarb, Donald
2Chen, Lifeng
title
0Fixed point and Bregman iterative methods for matrix rank minimization
1Mathematical programming
addtitleMath. Program
descriptionThe linearly constrained matrix rank minimization problem is widely applicable in many fields such as control, signal processing and system identification. The tightest convex relaxation of this problem is the linearly constrained nuclear norm minimization. Although the latter can be cast as a semidefinite programming problem, such an approach is computationally expensive to solve when the matrices are large. In this paper, we propose fixed point and Bregman iterative algorithms for solving the nuclear norm minimization problem and prove convergence of the first of these algorithms. By using a homotopy approach together with an approximate singular value decomposition procedure, we get a very fast, robust and powerful algorithm, which we call FPCA (Fixed Point Continuation with Approximate SVD), that can solve very large matrix rank minimization problems (the code can be downloaded from http://www.columbia.edu/~sm2756/FPCA.htm for non-commercial use). Our numerical results on randomly generated and real matrix completion problems demonstrate that this algorithm is much faster and provides much better recoverability than semidefinite programming solvers such as SDPT3. For example, our algorithm can recover 1000 × 1000 matrices of rank 50 with a relative error of 10 −5 in about 3 min by sampling only 20% of the elements. We know of no other method that achieves as good recoverability. Numerical experiments on online recommendation, DNA microarray data set and image inpainting problems demonstrate the effectiveness of our algorithms.
subject
0Algorithms
1Analysis
2Applied sciences
3Approximation
4Calculus of Variations and Optimal Control; Optimization
5Combinatorics
6Constraints
7DNA microarrays
8Exact sciences and technology
9Full Length Paper
10Images
11Iterative methods
12Management science
13Mathematical and Computational Physics
14Mathematical Methods in Physics
15Mathematical programming
16Mathematics
17Mathematics and Statistics
18Mathematics of Computing
19Matrices
20Matrix
21Methods
22Minimization
23Norms
24Numerical Analysis
25Operational research and scientific management
26Operational research. Management science
27Optimization
28Programming
29Semidefinite programming
30Signal processing
31Studies
32Theoretical
issn
00025-5610
11436-4646
fulltextfalse
rsrctypearticle
creationdate2009
recordtypearticle
recordideNp9kU9rFTEUxYMo-Kx-AHdBEFdTb_5OsqzFp0LBja5DJpM8U2eSZ5In1U9v6hSFgpJFIPzOybn3IPScwDkBGF9XAgTGAUAPwEAO4gHaEc7kwCWXD9EOgIpBSAKP0ZNarwGAMKV2aL-PN37GxxxTwzbN-E3xh9UmHJsvtsXvHq--fclzxSEXvNpW4g0uNn3Fa0xxjT87lNNT9CjYpfpnd_cZ-rx_--ny_XD18d2Hy4urgXDJ2kAtk3JScnJBCKH4zDUDFkaYR88kBEEm7iflp0koOytB-Mip4G5WetJCTnCGXm2-x5K_nXxtZo3V-WWxyedTNUpppkahoZMv7pHX-VRSD2eUlLyPT3WHzjfoYBdvYgq5Fev6mf0aXU4-xP5-QYVkkjOtuoBsAldyrcUHcyxxteWHIWBuizBbEaYXYW6LMKJrXt4lsdXZJfTluVj_CCmnoEZNOzfe83ax_d5uDxWX__5AN2Xtpungy99Z_y36Bc8Np7s
startdate20090923
enddate20090923
creator
0Ma, Shiqian
1Goldfarb, Donald
2Chen, Lifeng
general
0Springer-Verlag
1Springer
2Springer Nature B.V
scope
0IQODW
1AAYXX
2CITATION
3BSHEE
43V.
57SC
67WY
77WZ
87XB
987Z
1088I
118AL
128AO
138FD
148FE
158FG
168FK
178FL
18ABJCF
19ABUWG
20ARAPS
21AZQEC
22BENPR
23BEZIV
24BGLVJ
25DWQXO
26FRNLG
27F~G
28GNUQQ
29HCIFZ
30JQ2
31K60
32K6~
33K7-
34L.-
35L6V
36L7M
37L~C
38L~D
39M0C
40M0N
41M2P
42M7S
43P5Z
44P62
45PQBIZ
46PQBZA
47PQEST
48PQQKQ
49PQUKI
50PRINS
51PTHSS
52PYYUZ
53Q9U
sort
creationdate20090923
titleFixed point and Bregman iterative methods for matrix rank minimization
authorMa, Shiqian ; Goldfarb, Donald ; Chen, Lifeng
facets
frbrtype5
frbrgroupidcdi_FETCH-LOGICAL-1463t-2a366b86bcf55584d49303f70d7e360f51b4eb8ebb58ad851474254cd89b956b0
rsrctypearticles
prefilterarticles
languageeng
creationdate2009
topic
0Algorithms
1Analysis
2Applied sciences
3Approximation
4Calculus of Variations and Optimal Control; Optimization
5Combinatorics
6Constraints
7DNA microarrays
8Exact sciences and technology
9Full Length Paper
10Images
11Iterative methods
12Management science
13Mathematical and Computational Physics
14Mathematical Methods in Physics
15Mathematical programming
16Mathematics
17Mathematics and Statistics
18Mathematics of Computing
19Matrices
20Matrix
21Methods
22Minimization
23Norms
24Numerical Analysis
25Operational research and scientific management
26Operational research. Management science
27Optimization
28Programming
29Semidefinite programming
30Signal processing
31Studies
32Theoretical
toplevelpeer_reviewed
creatorcontrib
0Ma, Shiqian
1Goldfarb, Donald
2Chen, Lifeng
collection
0Pascal-Francis
1CrossRef
2Academic OneFile (A&I only)
3ProQuest Central (Corporate)
4Computer and Information Systems Abstracts
5ABI/INFORM Collection
6ABI/INFORM Global (PDF only)
7ProQuest Central (purchase pre-March 2016)
8ABI/INFORM Global (Alumni Edition)
9Science Database (Alumni Edition)
10Computing Database (Alumni Edition)
11ProQuest Pharma Collection
12Technology Research Database
13ProQuest SciTech Collection
14ProQuest Technology Collection
15ProQuest Central (Alumni) (purchase pre-March 2016)
16ABI/INFORM Collection (Alumni Edition)
17Materials Science & Engineering Collection
18ProQuest Central (Alumni Edition)
19Advanced Technologies & Aerospace Collection
20ProQuest Central Essentials
21ProQuest Central
22Business Premium Collection
23Technology Collection
24ProQuest Central Korea
25Business Premium Collection (Alumni)
26ABI/INFORM Global (Corporate)
27ProQuest Central Student
28SciTech Premium Collection
29ProQuest Computer Science Collection
30ProQuest Business Collection (Alumni Edition)
31ProQuest Business Collection
32Computer Science Database
33ABI/INFORM Professional Advanced
34ProQuest Engineering Collection
35Advanced Technologies Database with Aerospace
36Computer and Information Systems Abstracts – Academic
37Computer and Information Systems Abstracts Professional
38ABI/INFORM Global
39Computing Database
40Science Database
41Engineering Database
42Advanced Technologies & Aerospace Database
43ProQuest Advanced Technologies & Aerospace Collection
44ProQuest One Business
45ProQuest One Business (Alumni)
46ProQuest One Academic Eastern Edition
47ProQuest One Academic
48ProQuest One Academic UKI Edition
49ProQuest Central China
50Engineering Collection
51ABI/INFORM Collection China
52ProQuest Central Basic
jtitleMathematical programming
delivery
delcategoryRemote Search Resource
fulltextno_fulltext
addata
au
0Ma, Shiqian
1Goldfarb, Donald
2Chen, Lifeng
formatjournal
genrearticle
ristypeJOUR
atitleFixed point and Bregman iterative methods for matrix rank minimization
jtitleMathematical programming
stitleMath. Program
date2009-09-23
risdate2009
volume128
issue1-2
spage321
epage353
pages321-353
issn0025-5610
eissn1436-4646
codenMHPGA4
abstractThe linearly constrained matrix rank minimization problem is widely applicable in many fields such as control, signal processing and system identification. The tightest convex relaxation of this problem is the linearly constrained nuclear norm minimization. Although the latter can be cast as a semidefinite programming problem, such an approach is computationally expensive to solve when the matrices are large. In this paper, we propose fixed point and Bregman iterative algorithms for solving the nuclear norm minimization problem and prove convergence of the first of these algorithms. By using a homotopy approach together with an approximate singular value decomposition procedure, we get a very fast, robust and powerful algorithm, which we call FPCA (Fixed Point Continuation with Approximate SVD), that can solve very large matrix rank minimization problems (the code can be downloaded from http://www.columbia.edu/~sm2756/FPCA.htm for non-commercial use). Our numerical results on randomly generated and real matrix completion problems demonstrate that this algorithm is much faster and provides much better recoverability than semidefinite programming solvers such as SDPT3. For example, our algorithm can recover 1000 × 1000 matrices of rank 50 with a relative error of 10 −5 in about 3 min by sampling only 20% of the elements. We know of no other method that achieves as good recoverability. Numerical experiments on online recommendation, DNA microarray data set and image inpainting problems demonstrate the effectiveness of our algorithms.
copBerlin/Heidelberg
pubSpringer-Verlag
doi10.1007/s10107-009-0306-5