Kernel-based low-rank feature extraction on a budget for big data streams

Fatemeh Sheikholeslami; Dimitris Berberidis; Georgios B. Giannakis

doi:10.1109/GlobalSIP.2015.7418333

Kernel-based low-rank feature extraction on a budget for big data streams

Fatemeh Sheikholeslami, Dimitris Berberidis, Georgios B. Giannakis

Electrical and Computer Engineering

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

8 Scopus citations

Abstract

With nowadays big data torrent, identifying low-dimensional latent structures and extracting features from massive datasets are tasks of paramount importance. To this end, as real data generally lie on (or close to) nonlinear manifolds, kernel-based approaches are well motivated. Being nonparametric, unfortunately kernel-based feature extraction incurs complexity that grows prohibitively with the number of data. In response to this formidable challenge, the present work puts forward a low-rank, kernel-based feature extraction method, where the number of kernel functions is confined to an affordable budget. The resultant algorithm is particularly tailored for online operation, where data streams need not even be stored in memory. Tests on synthetic and real datasets demonstrate and benchmark the efficiency of the proposed method on linear classification applied to the extracted features.

Original language	English (US)
Title of host publication	2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	928-932
Number of pages	5
ISBN (Electronic)	9781479975914
DOIs	https://doi.org/10.1109/GlobalSIP.2015.7418333
State	Published - Feb 23 2016
Event	IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015 - Orlando, United States Duration: Dec 13 2015 → Dec 16 2015

Publication series

Name	2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015

Other

Other	IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015
Country/Territory	United States
City	Orlando
Period	12/13/15 → 12/16/15

Bibliographical note

Funding Information:
Work in this paper was supported by NSF grants, 1343248, 1500713, and NIH grant no. 1R01GM104975-01.

Publisher Copyright:
© 2015 IEEE.

Keywords

budgeted learning
classification
kernel methods
online feature extraction
subspace tracking

Access

10.1109/GlobalSIP.2015.7418333

OpenUrl availability

Full text

Cite this

Sheikholeslami, F., Berberidis, D., & Giannakis, G. B. (2016). Kernel-based low-rank feature extraction on a budget for big data streams. In 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015 (pp. 928-932). Article 7418333 (2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/GlobalSIP.2015.7418333

Kernel-based low-rank feature extraction on a budget for big data streams. / Sheikholeslami, Fatemeh; Berberidis, Dimitris; Giannakis, Georgios B.
2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015. Institute of Electrical and Electronics Engineers Inc., 2016. p. 928-932 7418333 (2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015).

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

Sheikholeslami, F, Berberidis, D & Giannakis, GB 2016, Kernel-based low-rank feature extraction on a budget for big data streams. in 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015., 7418333, 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015, Institute of Electrical and Electronics Engineers Inc., pp. 928-932, IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015, Orlando, United States, 12/13/15. https://doi.org/10.1109/GlobalSIP.2015.7418333

Sheikholeslami F, Berberidis D, Giannakis GB. Kernel-based low-rank feature extraction on a budget for big data streams. In 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015. Institute of Electrical and Electronics Engineers Inc. 2016. p. 928-932. 7418333. (2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015). doi: 10.1109/GlobalSIP.2015.7418333

Sheikholeslami, Fatemeh ; Berberidis, Dimitris ; Giannakis, Georgios B. / Kernel-based low-rank feature extraction on a budget for big data streams. 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015. Institute of Electrical and Electronics Engineers Inc., 2016. pp. 928-932 (2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015).

@inproceedings{6f714b619214474d828ee9a27fd3f5dd,

title = "Kernel-based low-rank feature extraction on a budget for big data streams",

abstract = "With nowadays big data torrent, identifying low-dimensional latent structures and extracting features from massive datasets are tasks of paramount importance. To this end, as real data generally lie on (or close to) nonlinear manifolds, kernel-based approaches are well motivated. Being nonparametric, unfortunately kernel-based feature extraction incurs complexity that grows prohibitively with the number of data. In response to this formidable challenge, the present work puts forward a low-rank, kernel-based feature extraction method, where the number of kernel functions is confined to an affordable budget. The resultant algorithm is particularly tailored for online operation, where data streams need not even be stored in memory. Tests on synthetic and real datasets demonstrate and benchmark the efficiency of the proposed method on linear classification applied to the extracted features.",

keywords = "budgeted learning, classification, kernel methods, online feature extraction, subspace tracking",

author = "Fatemeh Sheikholeslami and Dimitris Berberidis and Giannakis, {Georgios B.}",

note = "Funding Information: Work in this paper was supported by NSF grants, 1343248, 1500713, and NIH grant no. 1R01GM104975-01. Publisher Copyright: {\textcopyright} 2015 IEEE.; IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015 ; Conference date: 13-12-2015 Through 16-12-2015",

year = "2016",

month = feb,

day = "23",

doi = "10.1109/GlobalSIP.2015.7418333",

language = "English (US)",

series = "2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015",

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

pages = "928--932",

booktitle = "2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015",

}

TY - GEN

T1 - Kernel-based low-rank feature extraction on a budget for big data streams

AU - Sheikholeslami, Fatemeh

AU - Berberidis, Dimitris

AU - Giannakis, Georgios B.

PY - 2016/2/23

Y1 - 2016/2/23

N2 - With nowadays big data torrent, identifying low-dimensional latent structures and extracting features from massive datasets are tasks of paramount importance. To this end, as real data generally lie on (or close to) nonlinear manifolds, kernel-based approaches are well motivated. Being nonparametric, unfortunately kernel-based feature extraction incurs complexity that grows prohibitively with the number of data. In response to this formidable challenge, the present work puts forward a low-rank, kernel-based feature extraction method, where the number of kernel functions is confined to an affordable budget. The resultant algorithm is particularly tailored for online operation, where data streams need not even be stored in memory. Tests on synthetic and real datasets demonstrate and benchmark the efficiency of the proposed method on linear classification applied to the extracted features.

AB - With nowadays big data torrent, identifying low-dimensional latent structures and extracting features from massive datasets are tasks of paramount importance. To this end, as real data generally lie on (or close to) nonlinear manifolds, kernel-based approaches are well motivated. Being nonparametric, unfortunately kernel-based feature extraction incurs complexity that grows prohibitively with the number of data. In response to this formidable challenge, the present work puts forward a low-rank, kernel-based feature extraction method, where the number of kernel functions is confined to an affordable budget. The resultant algorithm is particularly tailored for online operation, where data streams need not even be stored in memory. Tests on synthetic and real datasets demonstrate and benchmark the efficiency of the proposed method on linear classification applied to the extracted features.

KW - budgeted learning

KW - classification

KW - kernel methods

KW - online feature extraction

KW - subspace tracking

UR - http://www.scopus.com/inward/record.url?scp=84964699626&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84964699626&partnerID=8YFLogxK

U2 - 10.1109/GlobalSIP.2015.7418333

DO - 10.1109/GlobalSIP.2015.7418333

M3 - Conference contribution

AN - SCOPUS:84964699626

T3 - 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015

SP - 928

EP - 932

BT - 2015 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - IEEE Global Conference on Signal and Information Processing, GlobalSIP 2015

Y2 - 13 December 2015 through 16 December 2015

ER -

Kernel-based low-rank feature extraction on a budget for big data streams

Abstract

Publication series

Other

Bibliographical note

Keywords

Access

OpenUrl availability

Other files and links

Fingerprint

Cite this