<?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-02437752</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="2026-05-23T10:17:07+02:00"/>
      </publicationStmt>
      <sourceDesc>
        <p part="N">HAL API Platform</p>
      </sourceDesc>
    </fileDesc>
  </teiHeader>
  <text>
    <body>
      <listBibl>
        <biblFull>
          <titleStmt>
            <title xml:lang="en">Adversarial Robustness via Label-Smoothing</title>
            <title xml:lang="fr">Robustesse Adversaire via Label-Smoothing</title>
            <author role="aut">
              <persName>
                <forename type="first">Morgane</forename>
                <surname>Goibert</surname>
              </persName>
              <email type="md5">68407ddf758f12b12dec5ac15f1cc5f6</email>
              <email type="domain">criteo.com</email>
              <idno type="idhal" notation="string">morgane-goibert</idno>
              <idno type="idhal" notation="numeric">1220537</idno>
              <idno type="halauthorid" notation="string">1767388-1220537</idno>
              <idno type="ARXIV">https://arxiv.org/a/goibert_m_1</idno>
              <idno type="GOOGLE SCHOLAR">2gyHK_MAAAAJ</idno>
              <affiliation ref="#struct-458139"/>
              <affiliation ref="#struct-300362"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Elvis</forename>
                <surname>Dohmatob</surname>
              </persName>
              <email type="md5">087e363e667b9dbec4ffc86ab0950a4d</email>
              <email type="domain">inria.fr</email>
              <idno type="idhal" notation="numeric">956319</idno>
              <idno type="halauthorid" notation="string">728609-956319</idno>
              <affiliation ref="#struct-458139"/>
            </author>
            <editor role="depositor">
              <persName>
                <forename>Morgane</forename>
                <surname>GOIBERT</surname>
              </persName>
              <email type="md5">0e1095055249792b7022b873e0673586</email>
              <email type="domain">gmail.com</email>
            </editor>
          </titleStmt>
          <editionStmt>
            <edition n="v1" type="current">
              <date type="whenSubmitted">2020-01-13 21:40:47</date>
              <date type="whenModified">2024-09-17 15:49:28</date>
              <date type="whenReleased">2020-01-17 12:27:29</date>
              <date type="whenProduced">2020-01-13</date>
              <date type="whenEndEmbargoed">2020-01-13</date>
              <ref type="file" target="https://hal.science/hal-02437752v1/document">
                <date notBefore="2020-01-13"/>
              </ref>
              <ref type="file" subtype="author" n="1" target="https://hal.science/hal-02437752v1/file/Article_adversarial_label_smoothing.pdf" id="file-2437752-2324119">
                <date notBefore="2020-01-13"/>
              </ref>
            </edition>
            <respStmt>
              <resp>contributor</resp>
              <name key="449652">
                <persName>
                  <forename>Morgane</forename>
                  <surname>GOIBERT</surname>
                </persName>
                <email type="md5">0e1095055249792b7022b873e0673586</email>
                <email type="domain">gmail.com</email>
              </name>
            </respStmt>
          </editionStmt>
          <publicationStmt>
            <distributor>CCSD</distributor>
            <idno type="halId">hal-02437752</idno>
            <idno type="halUri">https://hal.science/hal-02437752</idno>
            <idno type="halBibtex">goibert:hal-02437752</idno>
            <idno type="halRefHtml">2020</idno>
            <idno type="halRef">2020</idno>
            <availability status="restricted">
              <licence target="https://about.hal.science/hal-authorisation-v1/">HAL Authorization<ref corresp="#file-2437752-2324119"/></licence>
            </availability>
          </publicationStmt>
          <seriesStmt>
            <idno type="stamp" n="ENST">Ecole Nationale Supérieure des Télécommunications</idno>
            <idno type="stamp" n="TELECOM-PARISTECH" corresp="PARISTECH">Télécom Paris</idno>
            <idno type="stamp" n="IP_PARIS">Institut Polytechnique de Paris</idno>
            <idno type="stamp" n="INSTITUT-MINES-TELECOM">Institut Mines Telecom</idno>
          </seriesStmt>
          <notesStmt/>
          <sourceDesc>
            <biblStruct>
              <analytic>
                <title xml:lang="en">Adversarial Robustness via Label-Smoothing</title>
                <title xml:lang="fr">Robustesse Adversaire via Label-Smoothing</title>
                <author role="aut">
                  <persName>
                    <forename type="first">Morgane</forename>
                    <surname>Goibert</surname>
                  </persName>
                  <email type="md5">68407ddf758f12b12dec5ac15f1cc5f6</email>
                  <email type="domain">criteo.com</email>
                  <idno type="idhal" notation="string">morgane-goibert</idno>
                  <idno type="idhal" notation="numeric">1220537</idno>
                  <idno type="halauthorid" notation="string">1767388-1220537</idno>
                  <idno type="ARXIV">https://arxiv.org/a/goibert_m_1</idno>
                  <idno type="GOOGLE SCHOLAR">2gyHK_MAAAAJ</idno>
                  <affiliation ref="#struct-458139"/>
                  <affiliation ref="#struct-300362"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Elvis</forename>
                    <surname>Dohmatob</surname>
                  </persName>
                  <email type="md5">087e363e667b9dbec4ffc86ab0950a4d</email>
                  <email type="domain">inria.fr</email>
                  <idno type="idhal" notation="numeric">956319</idno>
                  <idno type="halauthorid" notation="string">728609-956319</idno>
                  <affiliation ref="#struct-458139"/>
                </author>
              </analytic>
              <monogr>
                <imprint/>
              </monogr>
            </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="UNDEFINED">Preprints, Working Papers, ...</classCode>
              <classCode scheme="halOldTypology" n="UNDEFINED">Preprints, Working Papers, ...</classCode>
              <classCode scheme="halTreeTypology" n="UNDEFINED">Preprints, Working Papers, ...</classCode>
            </textClass>
            <abstract xml:lang="en">
              <p>We study Label-Smoothing as a means for improving adversarial robustness of supervised deep-learning models. After establishing a thorough and unified framework, we propose several variations to this general method: adversarial, Boltzmann and second-best Label-Smoothing methods, and we explain how to construct your own one. On various datasets (MNIST, CIFAR10, SVHN) and models (linear models, MLPs, LeNet, ResNet), we show that Label-Smoothing in general improves adversarial robustness against a variety of attacks (FGSM, BIM, DeepFool, Carlini-Wagner) by better taking account of the dataset geometry. The proposed Label-Smoothing methods have two main advantages: they can be implemented as a modified cross-entropy loss, thus do not require any modifications of the network architecture nor do they lead to increased training times, and they improve both standard and adversarial accuracy.</p>
            </abstract>
          </profileDesc>
        </biblFull>
      </listBibl>
    </body>
    <back>
      <listOrg type="structures">
        <org type="institution" xml:id="struct-458139" status="VALID">
          <orgName>Criteo [Paris]</orgName>
          <desc>
            <address>
              <addrLine>32 rue Blanche 75 009 Paris</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.criteo.com/fr/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-300362" status="VALID">
          <idno type="ROR">https://ror.org/01naq7912</idno>
          <orgName>Télécom ParisTech</orgName>
          <date type="start">2008-01-01</date>
          <date type="end">2019-06-11</date>
          <desc>
            <address>
              <addrLine>46 rue Barrault 75634 Paris Cedex 13</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.telecom-paris.fr</ref>
          </desc>
        </org>
      </listOrg>
    </back>
  </text>
</TEI>