Efficient algorithms for 'universally' constrained matrix and tensor factorization

Kejun Huang; Nicholas D. Sidiropoulos; Athanasios P. Liavas

doi:10.1109/EUSIPCO.2015.7362839

Efficient algorithms for 'universally' constrained matrix and tensor factorization

Kejun Huang, Nicholas D. Sidiropoulos, Athanasios P. Liavas

Electrical and Computer Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

4 Scopus citations

Abstract

We propose a general algorithmic framework for constrained matrix and tensor factorization, which is widely used in unsupervised learning. The new framework is a hybrid between alternating optimization (AO) and the alternating direction method of multipliers (ADMM): each matrix factor is updated in turn, using ADMM. This combination can naturally accommodate a great variety of constraints on the factor matrices, hence the term 'universal'. Computation caching and warm start strategies are used to ensure that each update is evaluated efficiently, while the outer AO framework guarantees that the algorithm converges monotonically. Simulations on synthetic data show significantly improved performance relative to state-of-the-art algorithms.

Original language	English (US)
Title of host publication	2015 23rd European Signal Processing Conference, EUSIPCO 2015
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	2521-2525
Number of pages	5
ISBN (Electronic)	9780992862633
DOIs	https://doi.org/10.1109/EUSIPCO.2015.7362839
State	Published - Dec 22 2015
Event	23rd European Signal Processing Conference, EUSIPCO 2015 - Nice, France Duration: Aug 31 2015 → Sep 4 2015

Publication series

Name	2015 23rd European Signal Processing Conference, EUSIPCO 2015

Other

Other	23rd European Signal Processing Conference, EUSIPCO 2015
Country/Territory	France
City	Nice
Period	8/31/15 → 9/4/15

Bibliographical note

Publisher Copyright:
© 2015 EURASIP.

Publisher link

10.1109/EUSIPCO.2015.7362839

Find It

Find It at U of M Libraries

Cite this

Huang, K., Sidiropoulos, N. D., & Liavas, A. P. (2015). Efficient algorithms for 'universally' constrained matrix and tensor factorization. In 2015 23rd European Signal Processing Conference, EUSIPCO 2015 (pp. 2521-2525). Article 7362839 (2015 23rd European Signal Processing Conference, EUSIPCO 2015). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/EUSIPCO.2015.7362839

Efficient algorithms for 'universally' constrained matrix and tensor factorization. / Huang, Kejun; Sidiropoulos, Nicholas D.; Liavas, Athanasios P.
2015 23rd European Signal Processing Conference, EUSIPCO 2015. Institute of Electrical and Electronics Engineers Inc., 2015. p. 2521-2525 7362839 (2015 23rd European Signal Processing Conference, EUSIPCO 2015).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Huang, K, Sidiropoulos, ND & Liavas, AP 2015, Efficient algorithms for 'universally' constrained matrix and tensor factorization. in 2015 23rd European Signal Processing Conference, EUSIPCO 2015., 7362839, 2015 23rd European Signal Processing Conference, EUSIPCO 2015, Institute of Electrical and Electronics Engineers Inc., pp. 2521-2525, 23rd European Signal Processing Conference, EUSIPCO 2015, Nice, France, 8/31/15. https://doi.org/10.1109/EUSIPCO.2015.7362839

@inproceedings{363c88f4012143ba9e2c2239c45a1fbc,

title = "Efficient algorithms for 'universally' constrained matrix and tensor factorization",

abstract = "We propose a general algorithmic framework for constrained matrix and tensor factorization, which is widely used in unsupervised learning. The new framework is a hybrid between alternating optimization (AO) and the alternating direction method of multipliers (ADMM): each matrix factor is updated in turn, using ADMM. This combination can naturally accommodate a great variety of constraints on the factor matrices, hence the term 'universal'. Computation caching and warm start strategies are used to ensure that each update is evaluated efficiently, while the outer AO framework guarantees that the algorithm converges monotonically. Simulations on synthetic data show significantly improved performance relative to state-of-the-art algorithms.",

author = "Kejun Huang and Sidiropoulos, {Nicholas D.} and Liavas, {Athanasios P.}",

note = "Publisher Copyright: {\textcopyright} 2015 EURASIP.; 23rd European Signal Processing Conference, EUSIPCO 2015 ; Conference date: 31-08-2015 Through 04-09-2015",

year = "2015",

month = dec,

day = "22",

doi = "10.1109/EUSIPCO.2015.7362839",

language = "English (US)",

series = "2015 23rd European Signal Processing Conference, EUSIPCO 2015",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "2521--2525",

booktitle = "2015 23rd European Signal Processing Conference, EUSIPCO 2015",

}

TY - GEN

T1 - Efficient algorithms for 'universally' constrained matrix and tensor factorization

AU - Huang, Kejun

AU - Sidiropoulos, Nicholas D.

AU - Liavas, Athanasios P.

PY - 2015/12/22

Y1 - 2015/12/22

N2 - We propose a general algorithmic framework for constrained matrix and tensor factorization, which is widely used in unsupervised learning. The new framework is a hybrid between alternating optimization (AO) and the alternating direction method of multipliers (ADMM): each matrix factor is updated in turn, using ADMM. This combination can naturally accommodate a great variety of constraints on the factor matrices, hence the term 'universal'. Computation caching and warm start strategies are used to ensure that each update is evaluated efficiently, while the outer AO framework guarantees that the algorithm converges monotonically. Simulations on synthetic data show significantly improved performance relative to state-of-the-art algorithms.

AB - We propose a general algorithmic framework for constrained matrix and tensor factorization, which is widely used in unsupervised learning. The new framework is a hybrid between alternating optimization (AO) and the alternating direction method of multipliers (ADMM): each matrix factor is updated in turn, using ADMM. This combination can naturally accommodate a great variety of constraints on the factor matrices, hence the term 'universal'. Computation caching and warm start strategies are used to ensure that each update is evaluated efficiently, while the outer AO framework guarantees that the algorithm converges monotonically. Simulations on synthetic data show significantly improved performance relative to state-of-the-art algorithms.

UR - http://www.scopus.com/inward/record.url?scp=84963938396&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84963938396&partnerID=8YFLogxK

U2 - 10.1109/EUSIPCO.2015.7362839

DO - 10.1109/EUSIPCO.2015.7362839

M3 - Conference contribution

AN - SCOPUS:84963938396

T3 - 2015 23rd European Signal Processing Conference, EUSIPCO 2015

SP - 2521

EP - 2525

BT - 2015 23rd European Signal Processing Conference, EUSIPCO 2015

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 23rd European Signal Processing Conference, EUSIPCO 2015

Y2 - 31 August 2015 through 4 September 2015

ER -

Efficient algorithms for 'universally' constrained matrix and tensor factorization

Abstract

Publication series

Other

Bibliographical note

Publisher link

Find It

Other files and links

Fingerprint

Cite this