<?xml version="1.0" encoding="utf-8"?>
<TEI xmlns="http://www.tei-c.org/ns/1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:hal="http://hal.archives-ouvertes.fr/" xmlns:gml="http://www.opengis.net/gml/3.3/" xmlns:gmlce="http://www.opengis.net/gml/3.3/ce" version="1.1" xsi:schemaLocation="http://www.tei-c.org/ns/1.0 http://api.archives-ouvertes.fr/documents/aofr-sword.xsd">
  <teiHeader>
    <fileDesc>
      <titleStmt>
        <title>HAL TEI export of hal-03944846</title>
      </titleStmt>
      <publicationStmt>
        <distributor>CCSD</distributor>
        <availability status="restricted">
          <licence target="https://creativecommons.org/publicdomain/zero/1.0/">CC0 1.0 - Universal</licence>
        </availability>
        <date when="2026-05-24T00:45:37+02:00"/>
      </publicationStmt>
      <sourceDesc>
        <p part="N">HAL API Platform</p>
      </sourceDesc>
    </fileDesc>
  </teiHeader>
  <text>
    <body>
      <listBibl>
        <biblFull>
          <titleStmt>
            <title xml:lang="en">Fishr: Invariant Gradient Variances for Out-of-Distribution Generalization</title>
            <author role="aut">
              <persName>
                <forename type="first">Alexandre</forename>
                <surname>Rame</surname>
              </persName>
              <idno type="halauthorid">2670242-0</idno>
              <affiliation ref="#struct-1085231"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Corentin</forename>
                <surname>Dancette</surname>
              </persName>
              <email type="md5">e41f9197f3a37b0b286704e04862a5b5</email>
              <email type="domain">sorbonne-universite.fr</email>
              <idno type="idhal" notation="string">corentin-dancette</idno>
              <idno type="idhal" notation="numeric">1357070</idno>
              <idno type="halauthorid" notation="string">1433337-1357070</idno>
              <idno type="RESEARCHERID">http://www.researcherid.com/rid/EQX-2468-2022</idno>
              <idno type="IDREF">https://www.idref.fr/270396497</idno>
              <idno type="ORCID">https://orcid.org/0009-0007-9274-4910</idno>
              <idno type="ARXIV">https://arxiv.org/a/dancette_c_1</idno>
              <idno type="GOOGLE SCHOLAR">https://scholar.google.fr/citations?user=2zReQdQAAAAJ</idno>
              <affiliation ref="#struct-1085231"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Matthieu</forename>
                <surname>Cord</surname>
              </persName>
              <idno type="halauthorid">13283-0</idno>
              <affiliation ref="#struct-1085231"/>
              <affiliation ref="#struct-546381"/>
            </author>
            <editor role="depositor">
              <persName>
                <forename>Corentin</forename>
                <surname>Dancette</surname>
              </persName>
              <email type="md5">e41f9197f3a37b0b286704e04862a5b5</email>
              <email type="domain">sorbonne-universite.fr</email>
            </editor>
            <funder ref="#projanr-51639"/>
          </titleStmt>
          <editionStmt>
            <edition n="v1" type="current">
              <date type="whenSubmitted">2023-01-18 11:51:55</date>
              <date type="whenModified">2025-10-10 15:42:21</date>
              <date type="whenReleased">2023-02-07 15:08:43</date>
              <date type="whenProduced">2022-04-25</date>
              <date type="whenEndEmbargoed">2023-01-18</date>
              <ref type="file" target="https://hal.science/hal-03944846v1/document">
                <date notBefore="2023-01-18"/>
              </ref>
              <ref type="file" subtype="author" n="1" target="https://hal.science/hal-03944846v1/file/fishr-final.pdf" id="file-3944846-3446595">
                <date notBefore="2023-01-18"/>
              </ref>
              <ref type="externalLink" target="http://arxiv.org/pdf/2109.02934"/>
            </edition>
            <respStmt>
              <resp>contributor</resp>
              <name key="873818">
                <persName>
                  <forename>Corentin</forename>
                  <surname>Dancette</surname>
                </persName>
                <email type="md5">e41f9197f3a37b0b286704e04862a5b5</email>
                <email type="domain">sorbonne-universite.fr</email>
              </name>
            </respStmt>
          </editionStmt>
          <publicationStmt>
            <distributor>CCSD</distributor>
            <idno type="halId">hal-03944846</idno>
            <idno type="halUri">https://hal.science/hal-03944846</idno>
            <idno type="halBibtex">rame:hal-03944846</idno>
            <idno type="halRefHtml">&lt;i&gt;39th International Conference on Machine Learning (ICML 2022)&lt;/i&gt;, Apr 2022, Baltimore, MD, United States. pp.18347--18377</idno>
            <idno type="halRef">39th International Conference on Machine Learning (ICML 2022), Apr 2022, Baltimore, MD, United States. pp.18347--18377</idno>
            <availability status="restricted">
              <licence target="https://about.hal.science/hal-authorisation-v1/">HAL Authorization<ref corresp="#file-3944846-3446595"/></licence>
            </availability>
          </publicationStmt>
          <seriesStmt>
            <idno type="stamp" n="CNRS">CNRS - Centre national de la recherche scientifique</idno>
            <idno type="stamp" n="ISIR" corresp="SORBONNE-UNIVERSITE">Institut des Systèmes Intelligents et de Robotique</idno>
            <idno type="stamp" n="GENCI">Publications ayant eu recours aux supercalculateurs du GENCI</idno>
            <idno type="stamp" n="SORBONNE-UNIVERSITE">Sorbonne Université</idno>
            <idno type="stamp" n="SORBONNE-UNIV" corresp="SORBONNE-UNIVERSITE">Sorbonne Université 01/01/2018</idno>
            <idno type="stamp" n="SU-SCIENCES" corresp="SORBONNE-UNIVERSITE">Faculté des Sciences de Sorbonne Université</idno>
            <idno type="stamp" n="SU-TI">Sorbonne Université - Texte Intégral</idno>
            <idno type="stamp" n="ANR">ANR</idno>
            <idno type="stamp" n="ALLIANCE-SU"> Alliance Sorbonne Université</idno>
            <idno type="stamp" n="ISIR_MLIA" corresp="ISIR">Machine Learning and Information Access</idno>
            <idno type="stamp" n="SUPRA_INGENIERIE">Ingénierie</idno>
            <idno type="stamp" n="ANR-IA-20" corresp="ANR-IA">ANR-IA-20</idno>
            <idno type="stamp" n="ANR-IA">ANR-IA</idno>
          </seriesStmt>
          <notesStmt>
            <note type="commentary">31 pages, 14 tables, 7 figures</note>
            <note type="audience" n="2">International</note>
            <note type="invited" n="0">No</note>
            <note type="popular" n="0">No</note>
            <note type="peer" n="1">Yes</note>
            <note type="proceedings" n="1">Yes</note>
          </notesStmt>
          <sourceDesc>
            <biblStruct>
              <analytic>
                <title xml:lang="en">Fishr: Invariant Gradient Variances for Out-of-Distribution Generalization</title>
                <author role="aut">
                  <persName>
                    <forename type="first">Alexandre</forename>
                    <surname>Rame</surname>
                  </persName>
                  <idno type="halauthorid">2670242-0</idno>
                  <affiliation ref="#struct-1085231"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Corentin</forename>
                    <surname>Dancette</surname>
                  </persName>
                  <email type="md5">e41f9197f3a37b0b286704e04862a5b5</email>
                  <email type="domain">sorbonne-universite.fr</email>
                  <idno type="idhal" notation="string">corentin-dancette</idno>
                  <idno type="idhal" notation="numeric">1357070</idno>
                  <idno type="halauthorid" notation="string">1433337-1357070</idno>
                  <idno type="RESEARCHERID">http://www.researcherid.com/rid/EQX-2468-2022</idno>
                  <idno type="IDREF">https://www.idref.fr/270396497</idno>
                  <idno type="ORCID">https://orcid.org/0009-0007-9274-4910</idno>
                  <idno type="ARXIV">https://arxiv.org/a/dancette_c_1</idno>
                  <idno type="GOOGLE SCHOLAR">https://scholar.google.fr/citations?user=2zReQdQAAAAJ</idno>
                  <affiliation ref="#struct-1085231"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Matthieu</forename>
                    <surname>Cord</surname>
                  </persName>
                  <idno type="halauthorid">13283-0</idno>
                  <affiliation ref="#struct-1085231"/>
                  <affiliation ref="#struct-546381"/>
                </author>
              </analytic>
              <monogr>
                <title level="m">Proceedings of Machine Learning Research</title>
                <meeting>
                  <title>39th International Conference on Machine Learning (ICML 2022)</title>
                  <date type="start">2022-04-25</date>
                  <date type="end">2022-04-29</date>
                  <settlement>Baltimore, MD</settlement>
                  <country key="US">United States</country>
                </meeting>
                <imprint>
                  <biblScope unit="volume">162</biblScope>
                  <biblScope unit="pp">18347--18377</biblScope>
                  <date type="datePub">2022</date>
                </imprint>
              </monogr>
              <idno type="arxiv">2109.02934</idno>
            </biblStruct>
          </sourceDesc>
          <profileDesc>
            <langUsage>
              <language ident="en">English</language>
            </langUsage>
            <textClass>
              <classCode scheme="halDomain" n="stat.ml">Statistics [stat]/Machine Learning [stat.ML]</classCode>
              <classCode scheme="halTypology" n="COMM">Conference papers</classCode>
              <classCode scheme="halOldTypology" n="COMM">Conference papers</classCode>
              <classCode scheme="halTreeTypology" n="COMM">Conference papers</classCode>
            </textClass>
            <abstract xml:lang="en">
              <p>Learning robust models that generalize well under changes in the data distribution is critical for real-world applications. To this end, there has been a growing surge of interest to learn simultaneously from multiple training domains - while enforcing different types of invariance across those domains. Yet, all existing approaches fail to show systematic benefits under controlled evaluation protocols. In this paper, we introduce a new regularization - named Fishr - that enforces domain invariance in the space of the gradients of the loss: specifically, the domain-level variances of gradients are matched across training domains. Our approach is based on the close relations between the gradient covariance, the Fisher Information and the Hessian of the loss: in particular, we show that Fishr eventually aligns the domain-level loss landscapes locally around the final weights. Extensive experiments demonstrate the effectiveness of Fishr for out-of-distribution generalization. Notably, Fishr improves the state of the art on the DomainBed benchmark and performs consistently better than Empirical Risk Minimization. Our code is available at https://github.com/alexrame/fishr.</p>
            </abstract>
          </profileDesc>
        </biblFull>
      </listBibl>
    </body>
    <back>
      <listOrg type="structures">
        <org type="researchteam" xml:id="struct-1085231" status="VALID">
          <orgName>Machine Learning and Information Access</orgName>
          <orgName type="acronym">MLIA</orgName>
          <date type="start">2022-01-01</date>
          <desc>
            <address>
              <addrLine>Tour 65-66 3ème étage4, place Jussieu75005 Paris – France</addrLine>
              <country key="FR"/>
            </address>
          </desc>
          <listRelation>
            <relation active="#struct-541937" type="direct"/>
            <relation active="#struct-413221" type="indirect"/>
            <relation name="UMR7222" active="#struct-441569" type="indirect"/>
          </listRelation>
        </org>
        <org type="laboratory" xml:id="struct-546381" status="VALID">
          <orgName>Valeo.ai</orgName>
          <desc>
            <address>
              <addrLine>Paris</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.valeo.com/fr/valeo-ai/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-552452" type="direct"/>
          </listRelation>
        </org>
        <org type="laboratory" xml:id="struct-541937" status="VALID">
          <idno type="IdRef">241179122</idno>
          <idno type="RNSR">200918463J</idno>
          <idno type="ROR">05neq8668</idno>
          <orgName>Institut des Systèmes Intelligents et de Robotique</orgName>
          <orgName type="acronym">ISIR</orgName>
          <date type="start">2018-01-01</date>
          <desc>
            <address>
              <addrLine>Sorbonne-Université, Boite courrier 173 4 Place JUSSIEU 75252 Paris cedex 05</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.isir.upmc.fr</ref>
          </desc>
          <listRelation>
            <relation active="#struct-413221" type="direct"/>
            <relation name="UMR7222" active="#struct-441569" type="direct"/>
          </listRelation>
        </org>
        <org type="regroupinstitution" xml:id="struct-413221" status="VALID">
          <idno type="IdRef">221333754</idno>
          <idno type="ROR">https://ror.org/02en5vm52</idno>
          <orgName>Sorbonne Université</orgName>
          <orgName type="acronym">SU</orgName>
          <date type="start">2018-01-01</date>
          <desc>
            <address>
              <addrLine>21 rue de l’École de médecine - 75006 Paris</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.sorbonne-universite.fr/</ref>
          </desc>
        </org>
        <org type="regroupinstitution" xml:id="struct-441569" status="VALID">
          <idno type="IdRef">02636817X</idno>
          <idno type="ISNI">0000000122597504</idno>
          <idno type="ROR">https://ror.org/02feahw73</idno>
          <orgName>Centre National de la Recherche Scientifique</orgName>
          <orgName type="acronym">CNRS</orgName>
          <date type="start">1939-10-19</date>
          <desc>
            <address>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.cnrs.fr/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-552452" status="VALID">
          <orgName>VALEO</orgName>
          <desc>
            <address>
              <addrLine>43, Rue Bayen, 75017 Paris</addrLine>
              <country key="FR"/>
            </address>
          </desc>
        </org>
      </listOrg>
      <listOrg type="projects">
        <org type="anrProject" xml:id="projanr-51639" status="VALID">
          <idno type="anr">ANR-20-CHIA-0022</idno>
          <orgName>VISA DEEP</orgName>
          <desc>Raisonnement Visuel dans les architectures de réseaux de neurones profonds</desc>
          <date type="start">2020</date>
        </org>
      </listOrg>
    </back>
  </text>
</TEI>