Coupled tensor decomposition: A step towards robust components

Matthieu Genicot, P. A. Absil, Renaud Lambiotte, Saber Sami

Résultats de recherche: Contribution dans un livre/un catalogue/un rapport/dans les actes d'une conférenceArticle dans les actes d'une conférence/un colloque

Résumé

Combining information present in multiple datasets is one of the key challenges to fully benefit from the increasing availability of data in a variety of fields. Coupled tensor factorization aims to address this challenge by performing a simultaneous decomposition of different tensors. However, tensor factorization tends to suffer from a lack of robustness as the number of components affects the results to a large extent. In this work, a general framework for coupled tensor factorization is built to extract reliable components. Results from both individual and coupled decompositions are compared and divergence measures are used to adapt the number of components. It results in a joint decomposition method with (i) a variable number of components, (ii) shared and unshared components among tensors and (iii) robust components. Results on simulated data show a better modelling of the sources composing the datasets and an improved evaluation of the number of shared sources.

langue originaleAnglais
titre2016 24th European Signal Processing Conference, EUSIPCO 2016
EditeurEuropean Signal Processing Conference, EUSIPCO
Pages1308-1312
Nombre de pages5
Volume2016-November
ISBN (Electronique)9780992862657
Les DOIs
étatPublié - 28 nov. 2016
Evénement24th European Signal Processing Conference, EUSIPCO 2016 - Budapest, Hongrie
Durée: 28 août 20162 sept. 2016

Une conférence

Une conférence24th European Signal Processing Conference, EUSIPCO 2016
PaysHongrie
La villeBudapest
période28/08/162/09/16

Empreinte digitale

Tensors
Decomposition
Factorization
Availability

Citer ceci

Genicot, M., Absil, P. A., Lambiotte, R., & Sami, S. (2016). Coupled tensor decomposition: A step towards robust components. Dans 2016 24th European Signal Processing Conference, EUSIPCO 2016 (Vol 2016-November, p. 1308-1312). [7760460] European Signal Processing Conference, EUSIPCO. https://doi.org/10.1109/EUSIPCO.2016.7760460
Genicot, Matthieu ; Absil, P. A. ; Lambiotte, Renaud ; Sami, Saber. / Coupled tensor decomposition : A step towards robust components. 2016 24th European Signal Processing Conference, EUSIPCO 2016. Vol 2016-November European Signal Processing Conference, EUSIPCO, 2016. p. 1308-1312
@inproceedings{718a818d5bb649c4a440b587e63d39af,
title = "Coupled tensor decomposition: A step towards robust components",
abstract = "Combining information present in multiple datasets is one of the key challenges to fully benefit from the increasing availability of data in a variety of fields. Coupled tensor factorization aims to address this challenge by performing a simultaneous decomposition of different tensors. However, tensor factorization tends to suffer from a lack of robustness as the number of components affects the results to a large extent. In this work, a general framework for coupled tensor factorization is built to extract reliable components. Results from both individual and coupled decompositions are compared and divergence measures are used to adapt the number of components. It results in a joint decomposition method with (i) a variable number of components, (ii) shared and unshared components among tensors and (iii) robust components. Results on simulated data show a better modelling of the sources composing the datasets and an improved evaluation of the number of shared sources.",
author = "Matthieu Genicot and Absil, {P. A.} and Renaud Lambiotte and Saber Sami",
year = "2016",
month = "11",
day = "28",
doi = "10.1109/EUSIPCO.2016.7760460",
language = "English",
volume = "2016-November",
pages = "1308--1312",
booktitle = "2016 24th European Signal Processing Conference, EUSIPCO 2016",
publisher = "European Signal Processing Conference, EUSIPCO",

}

Genicot, M, Absil, PA, Lambiotte, R & Sami, S 2016, Coupled tensor decomposition: A step towards robust components. Dans 2016 24th European Signal Processing Conference, EUSIPCO 2016. VOL. 2016-November, 7760460, European Signal Processing Conference, EUSIPCO, p. 1308-1312, 24th European Signal Processing Conference, EUSIPCO 2016, Budapest, Hongrie, 28/08/16. https://doi.org/10.1109/EUSIPCO.2016.7760460

Coupled tensor decomposition : A step towards robust components. / Genicot, Matthieu; Absil, P. A.; Lambiotte, Renaud; Sami, Saber.

2016 24th European Signal Processing Conference, EUSIPCO 2016. Vol 2016-November European Signal Processing Conference, EUSIPCO, 2016. p. 1308-1312 7760460.

Résultats de recherche: Contribution dans un livre/un catalogue/un rapport/dans les actes d'une conférenceArticle dans les actes d'une conférence/un colloque

TY - GEN

T1 - Coupled tensor decomposition

T2 - A step towards robust components

AU - Genicot, Matthieu

AU - Absil, P. A.

AU - Lambiotte, Renaud

AU - Sami, Saber

PY - 2016/11/28

Y1 - 2016/11/28

N2 - Combining information present in multiple datasets is one of the key challenges to fully benefit from the increasing availability of data in a variety of fields. Coupled tensor factorization aims to address this challenge by performing a simultaneous decomposition of different tensors. However, tensor factorization tends to suffer from a lack of robustness as the number of components affects the results to a large extent. In this work, a general framework for coupled tensor factorization is built to extract reliable components. Results from both individual and coupled decompositions are compared and divergence measures are used to adapt the number of components. It results in a joint decomposition method with (i) a variable number of components, (ii) shared and unshared components among tensors and (iii) robust components. Results on simulated data show a better modelling of the sources composing the datasets and an improved evaluation of the number of shared sources.

AB - Combining information present in multiple datasets is one of the key challenges to fully benefit from the increasing availability of data in a variety of fields. Coupled tensor factorization aims to address this challenge by performing a simultaneous decomposition of different tensors. However, tensor factorization tends to suffer from a lack of robustness as the number of components affects the results to a large extent. In this work, a general framework for coupled tensor factorization is built to extract reliable components. Results from both individual and coupled decompositions are compared and divergence measures are used to adapt the number of components. It results in a joint decomposition method with (i) a variable number of components, (ii) shared and unshared components among tensors and (iii) robust components. Results on simulated data show a better modelling of the sources composing the datasets and an improved evaluation of the number of shared sources.

UR - http://www.scopus.com/inward/record.url?scp=85005950526&partnerID=8YFLogxK

U2 - 10.1109/EUSIPCO.2016.7760460

DO - 10.1109/EUSIPCO.2016.7760460

M3 - Conference contribution

VL - 2016-November

SP - 1308

EP - 1312

BT - 2016 24th European Signal Processing Conference, EUSIPCO 2016

PB - European Signal Processing Conference, EUSIPCO

ER -

Genicot M, Absil PA, Lambiotte R, Sami S. Coupled tensor decomposition: A step towards robust components. Dans 2016 24th European Signal Processing Conference, EUSIPCO 2016. Vol 2016-November. European Signal Processing Conference, EUSIPCO. 2016. p. 1308-1312. 7760460 https://doi.org/10.1109/EUSIPCO.2016.7760460