Hispana. Acceso en línea al Patrimonio Cultural Digital Español

Está en:  › Dades de registre

ν-SVM solutions of constrained lasso and elastic net

Identificadores del recurso

Neurocomputing 275 (2018): 1921 – 1931

0925-2312

http://hdl.handle.net/10486/692651

10.1016/j.neucom.2017.10.029

1921

1931

275

Procedència

(Biblos-e Archivo)

Fitxa

Títol:: ν-SVM solutions of constrained lasso and elastic net
Tema:: Lasso; GLMNet; Nearest Point Problem; SVM; Informática
Descripció:: Many important linear sparse models have at its core the Lasso problem, for which the GLMNet algorithm is often considered as the current state of the art. Recently M. Jaggi has observed that Constrained Lasso (CL) can be reduced to an SVM-like problem, for which the LIBSVM library provides very efficient algorithms. This suggests that it could also be used advantageously to solve CL. In this work we will refine Jaggi’s arguments to reduce CL as well as constrained Elastic Net to a Nearest Point Problem, which in turn can be rewritten as an appropriate ν-SVM problem solvable by LIBSVM. We will also show experimentally that the well-known LIBSVM library results in a faster convergence than GLMNet for small problems and also, if properly adapted, for larger ones. Screening is another ingredient to speed up solving Lasso. Shrinking can be seen as the simpler alternative of SVM to screening and we will discuss how it also may in some cases reduce the cost of an SVM-based CL solution; With partial support from Spanish government grants TIN2013-42351-P, TIN2016-76406-P, TIN2015-70308-REDT and S2013/ICE-2845 CASI-CAM-CM; work also supported by project FACIL–Ayudas Fundación BBVA a Equipos de Investigación Científica 2016 and the UAM–ADIC Chair for Data Science and Machine Learning. The first author is also supported by the FPU–MEC grant AP-2012-5163. We gratefully acknowledge the use of the facilities of Centro de Computación Científica (CCC) at UAM and thank Red Eléctrica de España for kindly supplying wind energy data
Idioma:: English
Relació:: Gobierno de España. TIN2013-42351-P; Gobierno de España. TIN2016-76406-P; Gobierno de España. TIN2015-70308-REDT; Comunidad de Madrid. S2013/ICE-2845
Autor/Productor:: Torres-Barrán, Alberto; Alaiz Gudín, Carlos María; Dorronsoro Ibero, José Ramón
Editor:: Elsevier
Otros colaboradores/productores:: UAM. Departamento de Ingeniería Informática; Aprendizaje Automático (ING EPS-001)
Drets:: open access
Data:: 2018-01-31
Tipo de recurso:: journal article; info:eu-repo/semantics/acceptedVersion
Format:: application/pdf

oai_dc

Descarregar XML

<?xml version="1.0" encoding="UTF-8" ?>

<oai_dc:dc schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
1. <dc:title>ν-SVM solutions of constrained lasso and elastic net</dc:title>
2. <dc:creator>Torres-Barrán, Alberto</dc:creator>
3. <dc:creator>Alaiz Gudín, Carlos María</dc:creator>
4. <dc:creator>Dorronsoro Ibero, José Ramón</dc:creator>
5. <dc:contributor>UAM. Departamento de Ingeniería Informática</dc:contributor>
6. <dc:contributor>Aprendizaje Automático (ING EPS-001)</dc:contributor>
7. <dc:subject>Lasso</dc:subject>
8. <dc:subject>GLMNet</dc:subject>
9. <dc:subject>Nearest Point Problem</dc:subject>
10. <dc:subject>SVM</dc:subject>
11. <dc:subject>Informática</dc:subject>
12. <dc:description>Many important linear sparse models have at its core the Lasso problem, for which the GLMNet algorithm is often considered as the current state of the art. Recently M. Jaggi has observed that Constrained Lasso (CL) can be reduced to an SVM-like problem, for which the LIBSVM library provides very efficient algorithms. This suggests that it could also be used advantageously to solve CL. In this work we will refine Jaggi’s arguments to reduce CL as well as constrained Elastic Net to a Nearest Point Problem, which in turn can be rewritten as an appropriate ν-SVM problem solvable by LIBSVM. We will also show experimentally that the well-known LIBSVM library results in a faster convergence than GLMNet for small problems and also, if properly adapted, for larger ones. Screening is another ingredient to speed up solving Lasso. Shrinking can be seen as the simpler alternative of SVM to screening and we will discuss how it also may in some cases reduce the cost of an SVM-based CL solution</dc:description>
13. <dc:description>With partial support from Spanish government grants TIN2013-42351-P, TIN2016-76406-P, TIN2015-70308-REDT and S2013/ICE-2845 CASI-CAM-CM; work also supported by project FACIL–Ayudas Fundación BBVA a Equipos de Investigación Científica 2016 and the UAM–ADIC Chair for Data Science and Machine Learning. The first author is also supported by the FPU–MEC grant AP-2012-5163. We gratefully acknowledge the use of the facilities of Centro de Computación Científica (CCC) at UAM and thank Red Eléctrica de España for kindly supplying wind energy data</dc:description>
14. <dc:date>2018-01-31</dc:date>
15. <dc:type>journal article</dc:type>
16. <dc:type>info:eu-repo/semantics/acceptedVersion</dc:type>
17. <dc:identifier>Neurocomputing 275 (2018): 1921 – 1931</dc:identifier>
18. <dc:identifier>0925-2312</dc:identifier>
19. <dc:identifier>http://hdl.handle.net/10486/692651</dc:identifier>
20. <dc:identifier>10.1016/j.neucom.2017.10.029</dc:identifier>
21. <dc:identifier>1921</dc:identifier>
22. <dc:identifier>1931</dc:identifier>
23. <dc:identifier>275</dc:identifier>
24. <dc:language>eng</dc:language>
25. <dc:relation>Gobierno de España. TIN2013-42351-P</dc:relation>
26. <dc:relation>Gobierno de España. TIN2016-76406-P</dc:relation>
27. <dc:relation>Gobierno de España. TIN2015-70308-REDT</dc:relation>
28. <dc:relation>Comunidad de Madrid. S2013/ICE-2845</dc:relation>
29. <dc:rights>open access</dc:rights>
30. <dc:format>application/pdf</dc:format>
31. <dc:publisher>Elsevier</dc:publisher>
</oai_dc:dc>

oai_openaire

Descarregar XML

<?xml version="1.0" encoding="UTF-8" ?>

<oaire:resource schemaLocation="http://namespace.openaire.eu/schema/oaire/ https://www.openaire.eu/schema/repo-lit/4.0/openaire.xsd">
1. <datacite:titles>
  1. <datacite:title>ν-SVM solutions of constrained lasso and elastic net</datacite:title>
  </datacite:titles>
2. <datacite:creators>
  1. <datacite:creator>
    1. <datacite:creatorName>Torres-Barrán, Alberto</datacite:creatorName>
    </datacite:creator>
  2. <datacite:creator>
    1. <datacite:creatorName>Alaiz Gudín, Carlos María</datacite:creatorName>
    2. <datacite:nameIdentifier nameIdentifierScheme="ORCID" schemeURI="https://orcid.org">0000-0001-9410-1192</datacite:nameIdentifier>
    3. <datacite:affiliation>Universidad Autónoma de Madrid</datacite:affiliation>
    </datacite:creator>
  3. <datacite:creator>
    1. <datacite:creatorName>Dorronsoro Ibero, José Ramón</datacite:creatorName>
    2. <datacite:nameIdentifier nameIdentifierScheme="ORCID" schemeURI="https://orcid.org">0000-0002-5271-0616</datacite:nameIdentifier>
    3. <datacite:affiliation>Universidad Autónoma de Madrid</datacite:affiliation>
    </datacite:creator>
  </datacite:creators>
3. <datacite:contributors>
  1. <datacite:contributor contributorType="Other">
    1. <datacite:contributorName>UAM. Departamento de Ingeniería Informática</datacite:contributorName>
    </datacite:contributor>
  2. <datacite:contributor contributorType="Other">
    1. <datacite:contributorName>Aprendizaje Automático (ING EPS-001)</datacite:contributorName>
    </datacite:contributor>
  </datacite:contributors>
4. <datacite:subjects>
  1. <datacite:subject>Lasso</datacite:subject>
  2. <datacite:subject>GLMNet</datacite:subject>
  3. <datacite:subject>Nearest Point Problem</datacite:subject>
  4. <datacite:subject>SVM</datacite:subject>
  5. <datacite:subject>Informática</datacite:subject>
  </datacite:subjects>
5. <dc:description>Many important linear sparse models have at its core the Lasso problem, for which the GLMNet algorithm is often considered as the current state of the art. Recently M. Jaggi has observed that Constrained Lasso (CL) can be reduced to an SVM-like problem, for which the LIBSVM library provides very efficient algorithms. This suggests that it could also be used advantageously to solve CL. In this work we will refine Jaggi’s arguments to reduce CL as well as constrained Elastic Net to a Nearest Point Problem, which in turn can be rewritten as an appropriate ν-SVM problem solvable by LIBSVM. We will also show experimentally that the well-known LIBSVM library results in a faster convergence than GLMNet for small problems and also, if properly adapted, for larger ones. Screening is another ingredient to speed up solving Lasso. Shrinking can be seen as the simpler alternative of SVM to screening and we will discuss how it also may in some cases reduce the cost of an SVM-based CL solution</dc:description>
6. <dc:description>With partial support from Spanish government grants TIN2013-42351-P, TIN2016-76406-P, TIN2015-70308-REDT and S2013/ICE-2845 CASI-CAM-CM; work also supported by project FACIL–Ayudas Fundación BBVA a Equipos de Investigación Científica 2016 and the UAM–ADIC Chair for Data Science and Machine Learning. The first author is also supported by the FPU–MEC grant AP-2012-5163. We gratefully acknowledge the use of the facilities of Centro de Computación Científica (CCC) at UAM and thank Red Eléctrica de España for kindly supplying wind energy data</dc:description>
7. <datacite:dates>
  1. <datacite:date dateType="Issued">2018-01-31</datacite:date>
  </datacite:dates>
8. <oaire:resourceType resourceTypeGeneral="literature" uri="http://purl.org/coar/resource_type/c_6501">journal article</oaire:resourceType>
9. <oaire:version uri="http://purl.org/coar/version/c_be7fb7dd8ff6fe43">NA</oaire:version>
10. <datacite:identifier identifierType="HANDLE">http://hdl.handle.net/10486/692651</datacite:identifier>
11. <datacite:alternateIdentifiers>
  1. <datacite:alternateIdentifier alternateIdentifierType="DOI">10.1016/j.neucom.2017.10.029</datacite:alternateIdentifier>
  </datacite:alternateIdentifiers>
12. <datacite:relatedIdentifiers>
  1. <datacite:relatedIdentifier relatedIdentifierType="ISSN" relationType="IsPartOf">0925-2312</datacite:relatedIdentifier>
  </datacite:relatedIdentifiers>
13. <dc:language>eng</dc:language>
14. <datacite:rights rightsURI="http://purl.org/coar/access_right/c_abf2">open access</datacite:rights>
15. <dc:format>application/pdf</dc:format>
16. <datacite:sizes>
  1. <datacite:size>34 pag.</datacite:size>
  </datacite:sizes>
17. <dc:publisher>Elsevier</dc:publisher>
18. <oaire:file objectType="fulltext">https://repositorio.uam.es/bitstream/10486/692651/1/solutions_torres_neuro_2018_ps.pdf</oaire:file>
</oaire:resource>

xoai

Descarregar XML

<?xml version="1.0" encoding="UTF-8" ?>

<metadata schemaLocation="http://www.lyncode.com/xoai http://www.lyncode.com/xsd/xoai.xsd">
1. <element name="dc">
  1. <element name="contributor">
    1. <element name="author">
      1. <element name="es_ES">
        <field name="value">Torres-Barrán, Alberto</field>
        <field name="authority">9d2cdd8a-8a24-4ee3-92d0-c102ab77273d</field>
        <field name="confidence">500</field>
        <field name="orcid_id" />
        <field name="value">Alaiz Gudín, Carlos María</field>
        <field name="authority">264350</field>
        <field name="confidence">500</field>
        <field name="orcid_id">0000-0001-9410-1192</field>
        <field name="value">Dorronsoro Ibero, José Ramón</field>
        <field name="authority">259712</field>
        <field name="confidence">500</field>
        <field name="orcid_id">0000-0002-5271-0616</field>
        </element>
      </element>
    2. <element name="other">
      1. <element name="es_ES">
        <field name="value">UAM. Departamento de Ingeniería Informática</field>
        </element>
      </element>
    3. <element name="group">
      1. <element name="es_ES">
        <field name="value">Aprendizaje Automático (ING EPS-001)</field>
        </element>
      </element>
    </element>
  2. <element name="date">
    1. <element name="issued">
      1. <element name="es_ES">
        <field name="value">2018-01-31</field>
        </element>
      </element>
    </element>
  3. <element name="identifier">
    1. <element name="citation">
      1. <element name="en_US">
        <field name="value">Neurocomputing 275 (2018): 1921 – 1931</field>
        </element>
      </element>
    2. <element name="issn">
      1. <element name="es_ES">
        <field name="value">0925-2312</field>
        </element>
      </element>
    3. <element name="uri">
      1. <element name="en_US">
        <field name="value">http://hdl.handle.net/10486/692651</field>
        </element>
      </element>
    4. <element name="doi">
      1. <element name="en_US">
        <field name="value">10.1016/j.neucom.2017.10.029</field>
        </element>
      </element>
    5. <element name="publicationfirstpage">
      1. <element name="es_ES">
        <field name="value">1921</field>
        </element>
      </element>
    6. <element name="publicationlastpage">
      1. <element name="es_ES">
        <field name="value">1931</field>
        </element>
      </element>
    7. <element name="publicationvolume">
      1. <element name="es_ES">
        <field name="value">275</field>
        </element>
      </element>
    </element>
  4. <element name="description">
    1. <element name="abstract">
      1. <element name="en_US">
        <field name="value">Many important linear sparse models have at its core the Lasso problem, for which the GLMNet algorithm is often considered as the current state of the art. Recently M. Jaggi has observed that Constrained Lasso (CL) can be reduced to an SVM-like problem, for which the LIBSVM library provides very efficient algorithms. This suggests that it could also be used advantageously to solve CL. In this work we will refine Jaggi’s arguments to reduce CL as well as constrained Elastic Net to a Nearest Point Problem, which in turn can be rewritten as an appropriate ν-SVM problem solvable by LIBSVM. We will also show experimentally that the well-known LIBSVM library results in a faster convergence than GLMNet for small problems and also, if properly adapted, for larger ones. Screening is another ingredient to speed up solving Lasso. Shrinking can be seen as the simpler alternative of SVM to screening and we will discuss how it also may in some cases reduce the cost of an SVM-based CL solution</field>
        </element>
      </element>
    2. <element name="sponsorship">
      1. <element name="en_US">
        <field name="value">With partial support from Spanish government grants TIN2013-42351-P, TIN2016-76406-P, TIN2015-70308-REDT and S2013/ICE-2845 CASI-CAM-CM; work also supported by project FACIL–Ayudas Fundación BBVA a Equipos de Investigación Científica 2016 and the UAM–ADIC Chair for Data Science and Machine Learning. The first author is also supported by the FPU–MEC grant AP-2012-5163. We gratefully acknowledge the use of the facilities of Centro de Computación Científica (CCC) at UAM and thank Red Eléctrica de España for kindly supplying wind energy data</field>
        </element>
      </element>
    </element>
  5. <element name="format">
    1. <element name="extent">
      1. <element name="es_ES">
        <field name="value">34 pag.</field>
        </element>
      </element>
    2. <element name="mimetype">
      1. <element name="en_US">
        <field name="value">application/pdf</field>
        </element>
      </element>
    </element>
  6. <element name="language">
    1. <element name="iso">
      1. <element name="en_US">
        <field name="value">eng</field>
        </element>
      </element>
    </element>
  7. <element name="publisher">
    1. <element name="en_US">
      1. <field name="value">Elsevier</field>
      </element>
    </element>
  8. <element name="relation">
    1. <element name="projectID">
      1. <element name="es_ES">
        <field name="value">Gobierno de España. TIN2013-42351-P</field>
        <field name="value">Gobierno de España. TIN2016-76406-P</field>
        <field name="value">Gobierno de España. TIN2015-70308-REDT</field>
        <field name="value">Comunidad de Madrid. S2013/ICE-2845</field>
        </element>
      </element>
    </element>
  9. <element name="rights">
    1. <element name="accessRights">
      1. <element name="en_US">
        <field name="value">open access</field>
        </element>
      </element>
    </element>
  10. <element name="subject">
    1. <element name="other">
      1. <element name="en_US">
        <field name="value">Lasso</field>
        <field name="value">GLMNet</field>
        <field name="value">Nearest Point Problem</field>
        <field name="value">SVM</field>
        </element>
      </element>
    2. <element name="eciencia">
      1. <element name="es_ES">
        <field name="value">Informática</field>
        </element>
      </element>
    </element>
  11. <element name="title">
    1. <element name="en_US">
      1. <field name="value">ν-SVM solutions of constrained lasso and elastic net</field>
      </element>
    </element>
  12. <element name="type">
    1. <element name="en_US">
      1. <field name="value">journal article</field>
      </element>
    2. <element name="version">
      1. <element name="en_US">
        <field name="value">info:eu-repo/semantics/acceptedVersion</field>
        </element>
      </element>
    </element>
  13. <element name="authorUAM">
    1. <element name="es_ES">
      1. <field name="value">Torres Barran, Alberto (264836)</field>
      2. <field name="value">Alaiz Gudín, Carlos María (264350)</field>
      3. <field name="value">Dorronsoro Ibero, José Ramón (259712)</field>
      </element>
    </element>
  14. <element name="facultadUAM">
    1. <element name="none">
      1. <field name="value">Escuela Politécnica Superior</field>
      </element>
    </element>
  </element>
2. <element name="bundles">
  1. <element name="bundle">
    1. <field name="name">ORIGINAL</field>
    2. <element name="bitstreams">
      1. <element name="bitstream">
        <field name="name">solutions_torres_neuro_2018_ps.pdf</field>
        <field name="originalName">solutions_torres_neuro_2018_ps.pdf</field>
        <field name="format">application/pdf</field>
        <field name="size">877237</field>
        <field name="url">https://repositorio.uam.es/bitstream/10486/692651/1/solutions_torres_neuro_2018_ps.pdf</field>
        <field name="checksum">4d102ff5032f925a277a5662cbca7fc9</field>
        <field name="checksumAlgorithm">MD5</field>
        <field name="sid">1</field>
        </element>
      </element>
    </element>
  2. <element name="bundle">
    1. <field name="name">LICENSE</field>
    2. <element name="bitstreams">
      1. <element name="bitstream">
        <field name="name">license.txt</field>
        <field name="originalName">license.txt</field>
        <field name="format">text/plain; charset=utf-8</field>
        <field name="size">4105</field>
        <field name="url">https://repositorio.uam.es/bitstream/10486/692651/2/license.txt</field>
        <field name="checksum">c9c6ba57a9a0757a94b73bd12b549d9f</field>
        <field name="checksumAlgorithm">MD5</field>
        <field name="sid">2</field>
        </element>
      </element>
    </element>
  3. <element name="bundle">
    1. <field name="name">TEXT</field>
    2. <element name="bitstreams">
      1. <element name="bitstream">
        <field name="name">solutions_torres_neuro_2018_ps.pdf.txt</field>
        <field name="originalName">solutions_torres_neuro_2018_ps.pdf.txt</field>
        <field name="description">Extracted text</field>
        <field name="format">text/plain</field>
        <field name="size">60891</field>
        <field name="url">https://repositorio.uam.es/bitstream/10486/692651/3/solutions_torres_neuro_2018_ps.pdf.txt</field>
        <field name="checksum">82e80bab516a60a65fe4172ef0f55e78</field>
        <field name="checksumAlgorithm">MD5</field>
        <field name="sid">3</field>
        </element>
      </element>
    </element>
  4. <element name="bundle">
    1. <field name="name">THUMBNAIL</field>
    2. <element name="bitstreams">
      1. <element name="bitstream">
        <field name="name">solutions_torres_neuro_2018_ps.pdf.jpg</field>
        <field name="originalName">solutions_torres_neuro_2018_ps.pdf.jpg</field>
        <field name="description">IM Thumbnail</field>
        <field name="format">image/jpeg</field>
        <field name="size">1283</field>
        <field name="url">https://repositorio.uam.es/bitstream/10486/692651/4/solutions_torres_neuro_2018_ps.pdf.jpg</field>
        <field name="checksum">b1afe8ac8f766ea4c9c6b094c20f051d</field>
        <field name="checksumAlgorithm">MD5</field>
        <field name="sid">4</field>
        </element>
      </element>
    </element>
  </element>
3. <element name="others">
  1. <field name="handle">10486/692651</field>
  2. <field name="identifier">oai:repositorio.uam.es:10486/692651</field>
  3. <field name="lastModifyDate">2022-05-05 10:24:41.675</field>
  </element>
4. <element name="repository">
  1. <field name="name">Universidad Autónoma de Madrid</field>
  2. <field name="mail">biblosearchivo.biblioteca@uam.es</field>
  </element>
5. <element name="license">
  1. <field name="bin">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</field>
  </element>
</metadata>