Block coordinate descent algorithms for large-scale sparse multiclass classification

Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multicl...
Ausführliche Beschreibung

Gespeichert in:

Autor*in:	Blondel, Mathieu [verfasserIn] Seki, Kazuhiro Uehara, Kuniaki

Format:	Artikel
Sprache:	Englisch

Erschienen:	2013

Schlagwörter:	Multiclass classification Group sparsity Block coordinate descent

Anmerkung:	© The Author(s) 2013

Übergeordnetes Werk:	Enthalten in: Machine learning - Springer US, 1986, 93(2013), 1 vom: 08. Mai, Seite 31-52
Übergeordnetes Werk:	volume:93 ; year:2013 ; number:1 ; day:08 ; month:05 ; pages:31-52

Links:	Volltext

DOI / URN:	10.1007/s10994-013-5367-2

Katalog-ID:	OLC2026524734

Internformat


LEADER	01000caa a22002652 4500
001	OLC2026524734
003	DE-627
005	20230503172248.0
007	tu
008	200820s2013 xx \|\|\|\|\| 00\| \|\|eng c
024	7		\|a 10.1007/s10994-013-5367-2 \|2 doi
035			\|a (DE-627)OLC2026524734
035			\|a (DE-He213)s10994-013-5367-2-p
040			\|a DE-627 \|b ger \|c DE-627 \|e rakwb
041			\|a eng
082	0	4	\|a 150 \|a 004 \|q VZ
100	1		\|a Blondel, Mathieu \|e verfasserin \|4 aut
245	1	0	\|a Block coordinate descent algorithms for large-scale sparse multiclass classification
264		1	\|c 2013
336			\|a Text \|b txt \|2 rdacontent
337			\|a ohne Hilfsmittel zu benutzen \|b n \|2 rdamedia
338			\|a Band \|b nc \|2 rdacarrier
500			\|a © The Author(s) 2013
520			\|a Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy.
650		4	\|a Multiclass classification
650		4	\|a Group sparsity
650		4	\|a Block coordinate descent
700	1		\|a Seki, Kazuhiro \|4 aut
700	1		\|a Uehara, Kuniaki \|4 aut
773	0	8	\|i Enthalten in \|t Machine learning \|d Springer US, 1986 \|g 93(2013), 1 vom: 08. Mai, Seite 31-52 \|w (DE-627)12920403X \|w (DE-600)54638-0 \|w (DE-576)014457377 \|x 0885-6125 \|7 nnns
773	1	8	\|g volume:93 \|g year:2013 \|g number:1 \|g day:08 \|g month:05 \|g pages:31-52
856	4	1	\|u https://doi.org/10.1007/s10994-013-5367-2 \|z lizenzpflichtig \|3 Volltext
912			\|a GBV_USEFLAG_A
912			\|a SYSFLAG_A
912			\|a GBV_OLC
912			\|a SSG-OLC-MAT
912			\|a GBV_ILN_24
912			\|a GBV_ILN_32
912			\|a GBV_ILN_70
912			\|a GBV_ILN_4012
912			\|a GBV_ILN_4046
912			\|a GBV_ILN_4318
951			\|a AR
952			\|d 93 \|j 2013 \|e 1 \|b 08 \|c 05 \|h 31-52

Indexfelder

author_variant	m b mb k s ks k u ku
matchkey_str	article:08856125:2013----::lccodntdsetloihsolreclsasml
hierarchy_sort_str	2013
publishDate	2013
allfields	10.1007/s10994-013-5367-2 doi (DE-627)OLC2026524734 (DE-He213)s10994-013-5367-2-p DE-627 ger DE-627 rakwb eng 150 004 VZ Blondel, Mathieu verfasserin aut Block coordinate descent algorithms for large-scale sparse multiclass classification 2013 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2013 Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. Multiclass classification Group sparsity Block coordinate descent Seki, Kazuhiro aut Uehara, Kuniaki aut Enthalten in Machine learning Springer US, 1986 93(2013), 1 vom: 08. Mai, Seite 31-52 (DE-627)12920403X (DE-600)54638-0 (DE-576)014457377 0885-6125 nnns volume:93 year:2013 number:1 day:08 month:05 pages:31-52 https://doi.org/10.1007/s10994-013-5367-2 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_24 GBV_ILN_32 GBV_ILN_70 GBV_ILN_4012 GBV_ILN_4046 GBV_ILN_4318 AR 93 2013 1 08 05 31-52
spelling	10.1007/s10994-013-5367-2 doi (DE-627)OLC2026524734 (DE-He213)s10994-013-5367-2-p DE-627 ger DE-627 rakwb eng 150 004 VZ Blondel, Mathieu verfasserin aut Block coordinate descent algorithms for large-scale sparse multiclass classification 2013 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2013 Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. Multiclass classification Group sparsity Block coordinate descent Seki, Kazuhiro aut Uehara, Kuniaki aut Enthalten in Machine learning Springer US, 1986 93(2013), 1 vom: 08. Mai, Seite 31-52 (DE-627)12920403X (DE-600)54638-0 (DE-576)014457377 0885-6125 nnns volume:93 year:2013 number:1 day:08 month:05 pages:31-52 https://doi.org/10.1007/s10994-013-5367-2 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_24 GBV_ILN_32 GBV_ILN_70 GBV_ILN_4012 GBV_ILN_4046 GBV_ILN_4318 AR 93 2013 1 08 05 31-52
allfields_unstemmed	10.1007/s10994-013-5367-2 doi (DE-627)OLC2026524734 (DE-He213)s10994-013-5367-2-p DE-627 ger DE-627 rakwb eng 150 004 VZ Blondel, Mathieu verfasserin aut Block coordinate descent algorithms for large-scale sparse multiclass classification 2013 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2013 Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. Multiclass classification Group sparsity Block coordinate descent Seki, Kazuhiro aut Uehara, Kuniaki aut Enthalten in Machine learning Springer US, 1986 93(2013), 1 vom: 08. Mai, Seite 31-52 (DE-627)12920403X (DE-600)54638-0 (DE-576)014457377 0885-6125 nnns volume:93 year:2013 number:1 day:08 month:05 pages:31-52 https://doi.org/10.1007/s10994-013-5367-2 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_24 GBV_ILN_32 GBV_ILN_70 GBV_ILN_4012 GBV_ILN_4046 GBV_ILN_4318 AR 93 2013 1 08 05 31-52
allfieldsGer	10.1007/s10994-013-5367-2 doi (DE-627)OLC2026524734 (DE-He213)s10994-013-5367-2-p DE-627 ger DE-627 rakwb eng 150 004 VZ Blondel, Mathieu verfasserin aut Block coordinate descent algorithms for large-scale sparse multiclass classification 2013 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2013 Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. Multiclass classification Group sparsity Block coordinate descent Seki, Kazuhiro aut Uehara, Kuniaki aut Enthalten in Machine learning Springer US, 1986 93(2013), 1 vom: 08. Mai, Seite 31-52 (DE-627)12920403X (DE-600)54638-0 (DE-576)014457377 0885-6125 nnns volume:93 year:2013 number:1 day:08 month:05 pages:31-52 https://doi.org/10.1007/s10994-013-5367-2 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_24 GBV_ILN_32 GBV_ILN_70 GBV_ILN_4012 GBV_ILN_4046 GBV_ILN_4318 AR 93 2013 1 08 05 31-52
allfieldsSound	10.1007/s10994-013-5367-2 doi (DE-627)OLC2026524734 (DE-He213)s10994-013-5367-2-p DE-627 ger DE-627 rakwb eng 150 004 VZ Blondel, Mathieu verfasserin aut Block coordinate descent algorithms for large-scale sparse multiclass classification 2013 Text txt rdacontent ohne Hilfsmittel zu benutzen n rdamedia Band nc rdacarrier © The Author(s) 2013 Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. Multiclass classification Group sparsity Block coordinate descent Seki, Kazuhiro aut Uehara, Kuniaki aut Enthalten in Machine learning Springer US, 1986 93(2013), 1 vom: 08. Mai, Seite 31-52 (DE-627)12920403X (DE-600)54638-0 (DE-576)014457377 0885-6125 nnns volume:93 year:2013 number:1 day:08 month:05 pages:31-52 https://doi.org/10.1007/s10994-013-5367-2 lizenzpflichtig Volltext GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_24 GBV_ILN_32 GBV_ILN_70 GBV_ILN_4012 GBV_ILN_4046 GBV_ILN_4318 AR 93 2013 1 08 05 31-52
language	English
source	Enthalten in Machine learning 93(2013), 1 vom: 08. Mai, Seite 31-52 volume:93 year:2013 number:1 day:08 month:05 pages:31-52
sourceStr	Enthalten in Machine learning 93(2013), 1 vom: 08. Mai, Seite 31-52 volume:93 year:2013 number:1 day:08 month:05 pages:31-52
format_phy_str_mv	Article
institution	findex.gbv.de
topic_facet	Multiclass classification Group sparsity Block coordinate descent
dewey-raw	150
isfreeaccess_bool	false
container_title	Machine learning
authorswithroles_txt_mv	Blondel, Mathieu @@aut@@ Seki, Kazuhiro @@aut@@ Uehara, Kuniaki @@aut@@
publishDateDaySort_date	2013-05-08T00:00:00Z
hierarchy_top_id	12920403X
dewey-sort	3150
id	OLC2026524734
language_de	englisch
fullrecord	<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">OLC2026524734</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230503172248.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">200820s2013 xx \|\|\|\|\| 00\| \|\|eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s10994-013-5367-2</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC2026524734</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s10994-013-5367-2-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">150</subfield><subfield code="a">004</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Blondel, Mathieu</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Block coordinate descent algorithms for large-scale sparse multiclass classification</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2013</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© The Author(s) 2013</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Multiclass classification</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Group sparsity</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Block coordinate descent</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Seki, Kazuhiro</subfield><subfield code="4">aut</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Uehara, Kuniaki</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">Machine learning</subfield><subfield code="d">Springer US, 1986</subfield><subfield code="g">93(2013), 1 vom: 08. Mai, Seite 31-52</subfield><subfield code="w">(DE-627)12920403X</subfield><subfield code="w">(DE-600)54638-0</subfield><subfield code="w">(DE-576)014457377</subfield><subfield code="x">0885-6125</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:93</subfield><subfield code="g">year:2013</subfield><subfield code="g">number:1</subfield><subfield code="g">day:08</subfield><subfield code="g">month:05</subfield><subfield code="g">pages:31-52</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s10994-013-5367-2</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_24</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_32</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_70</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4012</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4046</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4318</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">93</subfield><subfield code="j">2013</subfield><subfield code="e">1</subfield><subfield code="b">08</subfield><subfield code="c">05</subfield><subfield code="h">31-52</subfield></datafield></record></collection>
author	Blondel, Mathieu
spellingShingle	Blondel, Mathieu ddc 150 misc Multiclass classification misc Group sparsity misc Block coordinate descent Block coordinate descent algorithms for large-scale sparse multiclass classification
authorStr	Blondel, Mathieu
ppnlink_with_tag_str_mv	@@773@@(DE-627)12920403X
format	Article
dewey-ones	150 - Psychology 004 - Data processing & computer science
delete_txt_mv	keep
author_role	aut aut aut
collection	OLC
remote_str	false
illustrated	Not Illustrated
issn	0885-6125
topic_title	150 004 VZ Block coordinate descent algorithms for large-scale sparse multiclass classification Multiclass classification Group sparsity Block coordinate descent
topic	ddc 150 misc Multiclass classification misc Group sparsity misc Block coordinate descent
topic_unstemmed	ddc 150 misc Multiclass classification misc Group sparsity misc Block coordinate descent
topic_browse	ddc 150 misc Multiclass classification misc Group sparsity misc Block coordinate descent
format_facet	Aufsätze Gedruckte Aufsätze
format_main_str_mv	Text Zeitschrift/Artikel
carriertype_str_mv	nc
hierarchy_parent_title	Machine learning
hierarchy_parent_id	12920403X
dewey-tens	150 - Psychology 000 - Computer science, knowledge & systems
hierarchy_top_title	Machine learning
isfreeaccess_txt	false
familylinks_str_mv	(DE-627)12920403X (DE-600)54638-0 (DE-576)014457377
title	Block coordinate descent algorithms for large-scale sparse multiclass classification
ctrlnum	(DE-627)OLC2026524734 (DE-He213)s10994-013-5367-2-p
title_full	Block coordinate descent algorithms for large-scale sparse multiclass classification
author_sort	Blondel, Mathieu
journal	Machine learning
journalStr	Machine learning
lang_code	eng
isOA_bool	false
dewey-hundreds	100 - Philosophy & psychology 000 - Computer science, information & general works
recordtype	marc
publishDateSort	2013
contenttype_str_mv	txt
container_start_page	31
author_browse	Blondel, Mathieu Seki, Kazuhiro Uehara, Kuniaki
container_volume	93
class	150 004 VZ
format_se	Aufsätze
author-letter	Blondel, Mathieu
doi_str_mv	10.1007/s10994-013-5367-2
dewey-full	150 004
title_sort	block coordinate descent algorithms for large-scale sparse multiclass classification
title_auth	Block coordinate descent algorithms for large-scale sparse multiclass classification
abstract	Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. © The Author(s) 2013
abstractGer	Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. © The Author(s) 2013
abstract_unstemmed	Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy. © The Author(s) 2013
collection_details	GBV_USEFLAG_A SYSFLAG_A GBV_OLC SSG-OLC-MAT GBV_ILN_24 GBV_ILN_32 GBV_ILN_70 GBV_ILN_4012 GBV_ILN_4046 GBV_ILN_4318
container_issue	1
title_short	Block coordinate descent algorithms for large-scale sparse multiclass classification
url	https://doi.org/10.1007/s10994-013-5367-2
remote_bool	false
author2	Seki, Kazuhiro Uehara, Kuniaki
author2Str	Seki, Kazuhiro Uehara, Kuniaki
ppnlink	12920403X
mediatype_str_mv	n
isOA_txt	false
hochschulschrift_bool	false
doi_str	10.1007/s10994-013-5367-2
up_date	2024-07-04T04:09:49.351Z
_version_	1803620124568059904
fullrecord_marcxml	<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">OLC2026524734</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230503172248.0</controlfield><controlfield tag="007">tu</controlfield><controlfield tag="008">200820s2013 xx \|\|\|\|\| 00\| \|\|eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1007/s10994-013-5367-2</subfield><subfield code="2">doi</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)OLC2026524734</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-He213)s10994-013-5367-2-p</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">150</subfield><subfield code="a">004</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Blondel, Mathieu</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Block coordinate descent algorithms for large-scale sparse multiclass classification</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2013</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">Text</subfield><subfield code="b">txt</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">ohne Hilfsmittel zu benutzen</subfield><subfield code="b">n</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">Band</subfield><subfield code="b">nc</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="500" ind1=" " ind2=" "><subfield code="a">© The Author(s) 2013</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Abstract Over the past decade, ℓ1 regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ℓ1/ℓ2) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ℓ1/ℓ2 regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Takáč in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ℓ1/ℓ2-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy.</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Multiclass classification</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Group sparsity</subfield></datafield><datafield tag="650" ind1=" " ind2="4"><subfield code="a">Block coordinate descent</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Seki, Kazuhiro</subfield><subfield code="4">aut</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Uehara, Kuniaki</subfield><subfield code="4">aut</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="t">Machine learning</subfield><subfield code="d">Springer US, 1986</subfield><subfield code="g">93(2013), 1 vom: 08. Mai, Seite 31-52</subfield><subfield code="w">(DE-627)12920403X</subfield><subfield code="w">(DE-600)54638-0</subfield><subfield code="w">(DE-576)014457377</subfield><subfield code="x">0885-6125</subfield><subfield code="7">nnns</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:93</subfield><subfield code="g">year:2013</subfield><subfield code="g">number:1</subfield><subfield code="g">day:08</subfield><subfield code="g">month:05</subfield><subfield code="g">pages:31-52</subfield></datafield><datafield tag="856" ind1="4" ind2="1"><subfield code="u">https://doi.org/10.1007/s10994-013-5367-2</subfield><subfield code="z">lizenzpflichtig</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_A</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_OLC</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-MAT</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_24</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_32</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_70</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4012</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4046</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ILN_4318</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">93</subfield><subfield code="j">2013</subfield><subfield code="e">1</subfield><subfield code="b">08</subfield><subfield code="c">05</subfield><subfield code="h">31-52</subfield></datafield></record></collection>
score	7.4004326

Nicht das Richtige dabei?

Schreiben Sie uns!

Block coordinate descent algorithms for large-scale sparse multiclass classification

Nicht das Richtige dabei?

Zugang & Verfügbarkeit

Vorhandene Bände

Nicht das Richtige dabei?