<?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-02140748v3</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="2026-05-10T08:12:04+02:00"/>
      </publicationStmt>
      <sourceDesc>
        <p part="N">HAL API Platform</p>
      </sourceDesc>
    </fileDesc>
  </teiHeader>
  <text>
    <body>
      <listBibl>
        <biblFull>
          <titleStmt>
            <title xml:lang="en">An Inertial Newton Algorithm for Deep Learning</title>
            <author role="aut">
              <persName>
                <forename type="first">Camille</forename>
                <surname>Castera</surname>
              </persName>
              <email type="md5">792a571ae915b9a55af6e11680fdeb3e</email>
              <email type="domain">protonmail.com</email>
              <idno type="idhal" notation="string">camille-castera</idno>
              <idno type="idhal" notation="numeric">175473</idno>
              <idno type="halauthorid" notation="string">44956-175473</idno>
              <idno type="ORCID">https://orcid.org/0000-0002-7384-6387</idno>
              <idno type="IDREF">https://www.idref.fr/260859648</idno>
              <affiliation ref="#struct-443875"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Jérôme</forename>
                <surname>Bolte</surname>
              </persName>
              <email type="md5">72869e8074753be9e77ef96a3e3b8bd3</email>
              <email type="domain">math.jussieu.fr</email>
              <idno type="idhal" notation="numeric">846638</idno>
              <idno type="halauthorid" notation="string">118938-846638</idno>
              <idno type="IDREF">https://www.idref.fr/078638623</idno>
              <idno type="ORCID">https://orcid.org/0000-0002-1676-8407</idno>
              <affiliation ref="#struct-81148"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Cédric</forename>
                <surname>Févotte</surname>
              </persName>
              <email type="md5">de8ff2bf9b3c16250c18275768758506</email>
              <email type="domain">irit.fr</email>
              <idno type="idhal" notation="string">cedric-fevotte</idno>
              <idno type="idhal" notation="numeric">184864</idno>
              <idno type="halauthorid" notation="string">47288-184864</idno>
              <idno type="ORCID">https://orcid.org/0000-0003-3801-5534</idno>
              <idno type="GOOGLE SCHOLAR">https://scholar.google.com/citations?user=kAWEUq0AAAAJ</idno>
              <idno type="IDREF">https://www.idref.fr/083298460</idno>
              <orgName ref="#struct-441569"/>
              <affiliation ref="#struct-443875"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Edouard</forename>
                <surname>Pauwels</surname>
              </persName>
              <email type="md5">415266e6fc2a5f27a5dc092b9a48a184</email>
              <email type="domain">gmail.com</email>
              <idno type="idhal" notation="string">edouard-pauwels</idno>
              <idno type="idhal" notation="numeric">12830</idno>
              <idno type="halauthorid" notation="string">30358-12830</idno>
              <idno type="ORCID">https://orcid.org/0000-0002-8180-075X</idno>
              <orgName ref="#struct-217752"/>
              <affiliation ref="#struct-217752"/>
            </author>
            <editor role="depositor">
              <persName>
                <forename>Camille</forename>
                <surname>Castera</surname>
              </persName>
              <email type="md5">792a571ae915b9a55af6e11680fdeb3e</email>
              <email type="domain">protonmail.com</email>
            </editor>
            <funder ref="#projanr-50372"/>
            <funder ref="#projeurop-710321"/>
          </titleStmt>
          <editionStmt>
            <edition n="v1">
              <date type="whenSubmitted">2019-05-27 15:17:12</date>
            </edition>
            <edition n="v2">
              <date type="whenSubmitted">2019-06-06 15:11:10</date>
            </edition>
            <edition n="v3" type="current">
              <date type="whenSubmitted">2019-12-12 02:00:08</date>
              <date type="whenModified">2025-10-22 18:04:08</date>
              <date type="whenReleased">2019-12-16 15:15:54</date>
              <date type="whenProduced">2019-12-12</date>
              <date type="whenEndEmbargoed">2019-12-12</date>
              <ref type="file" target="https://hal.science/hal-02140748v3/document">
                <date notBefore="2019-12-12"/>
              </ref>
              <ref type="file" subtype="author" n="1" target="https://hal.science/hal-02140748v3/file/arXiv.pdf" id="file-2406002-2297742">
                <date notBefore="2019-12-12"/>
              </ref>
              <ref type="externalLink" target="http://arxiv.org/pdf/1905.12278"/>
            </edition>
            <edition n="v4">
              <date type="whenSubmitted">2020-10-12 18:49:58</date>
            </edition>
            <edition n="v5">
              <date type="whenSubmitted">2021-07-02 12:00:56</date>
            </edition>
            <edition n="v6">
              <date type="whenSubmitted">2021-08-20 14:28:19</date>
            </edition>
            <respStmt>
              <resp>contributor</resp>
              <name key="741231">
                <persName>
                  <forename>Camille</forename>
                  <surname>Castera</surname>
                </persName>
                <email type="md5">792a571ae915b9a55af6e11680fdeb3e</email>
                <email type="domain">protonmail.com</email>
              </name>
            </respStmt>
          </editionStmt>
          <publicationStmt>
            <distributor>CCSD</distributor>
            <idno type="halId">hal-02140748</idno>
            <idno type="halUri">https://hal.science/hal-02140748</idno>
            <idno type="halBibtex">castera:hal-02140748</idno>
            <idno type="halRefHtml">2019</idno>
            <idno type="halRef">2019</idno>
            <availability status="restricted">
              <licence target="https://about.hal.science/hal-authorisation-v1/">HAL Authorization<ref corresp="#file-2406002-2297742"/></licence>
            </availability>
          </publicationStmt>
          <seriesStmt/>
          <notesStmt/>
          <sourceDesc>
            <biblStruct>
              <analytic>
                <title xml:lang="en">An Inertial Newton Algorithm for Deep Learning</title>
                <author role="aut">
                  <persName>
                    <forename type="first">Camille</forename>
                    <surname>Castera</surname>
                  </persName>
                  <email type="md5">792a571ae915b9a55af6e11680fdeb3e</email>
                  <email type="domain">protonmail.com</email>
                  <idno type="idhal" notation="string">camille-castera</idno>
                  <idno type="idhal" notation="numeric">175473</idno>
                  <idno type="halauthorid" notation="string">44956-175473</idno>
                  <idno type="ORCID">https://orcid.org/0000-0002-7384-6387</idno>
                  <idno type="IDREF">https://www.idref.fr/260859648</idno>
                  <affiliation ref="#struct-443875"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Jérôme</forename>
                    <surname>Bolte</surname>
                  </persName>
                  <email type="md5">72869e8074753be9e77ef96a3e3b8bd3</email>
                  <email type="domain">math.jussieu.fr</email>
                  <idno type="idhal" notation="numeric">846638</idno>
                  <idno type="halauthorid" notation="string">118938-846638</idno>
                  <idno type="IDREF">https://www.idref.fr/078638623</idno>
                  <idno type="ORCID">https://orcid.org/0000-0002-1676-8407</idno>
                  <affiliation ref="#struct-81148"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Cédric</forename>
                    <surname>Févotte</surname>
                  </persName>
                  <email type="md5">de8ff2bf9b3c16250c18275768758506</email>
                  <email type="domain">irit.fr</email>
                  <idno type="idhal" notation="string">cedric-fevotte</idno>
                  <idno type="idhal" notation="numeric">184864</idno>
                  <idno type="halauthorid" notation="string">47288-184864</idno>
                  <idno type="ORCID">https://orcid.org/0000-0003-3801-5534</idno>
                  <idno type="GOOGLE SCHOLAR">https://scholar.google.com/citations?user=kAWEUq0AAAAJ</idno>
                  <idno type="IDREF">https://www.idref.fr/083298460</idno>
                  <orgName ref="#struct-441569"/>
                  <affiliation ref="#struct-443875"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Edouard</forename>
                    <surname>Pauwels</surname>
                  </persName>
                  <email type="md5">415266e6fc2a5f27a5dc092b9a48a184</email>
                  <email type="domain">gmail.com</email>
                  <idno type="idhal" notation="string">edouard-pauwels</idno>
                  <idno type="idhal" notation="numeric">12830</idno>
                  <idno type="halauthorid" notation="string">30358-12830</idno>
                  <idno type="ORCID">https://orcid.org/0000-0002-8180-075X</idno>
                  <orgName ref="#struct-217752"/>
                  <affiliation ref="#struct-217752"/>
                </author>
              </analytic>
              <monogr>
                <imprint/>
              </monogr>
              <idno type="arxiv">1905.12278</idno>
            </biblStruct>
          </sourceDesc>
          <profileDesc>
            <langUsage>
              <language ident="en">English</language>
            </langUsage>
            <textClass>
              <keywords scheme="author">
                <term xml:lang="en">Stochastic Optimization</term>
                <term xml:lang="en">Deep Learning Algorithms</term>
                <term xml:lang="en">Nonconvex optimization</term>
                <term xml:lang="fr">Optimisation non convexe</term>
                <term xml:lang="fr">Optimisation Stochastique</term>
                <term xml:lang="fr">Deep Learning</term>
                <term xml:lang="fr">Algorithmes pour le deep learning</term>
              </keywords>
              <classCode scheme="halDomain" n="info.info-lg">Computer Science [cs]/Machine Learning [cs.LG]</classCode>
              <classCode scheme="halDomain" n="math.math-oc">Mathematics [math]/Optimization and Control [math.OC]</classCode>
              <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 introduce a new second-order inertial method for machine learning called INDIAN, exploiting the geometry of the loss function while only requiring stochastic approximations of the function values and the generalized gradients. This makes the method fully implementable and adapted to large-scale optimization problems such as the training of a deep neural network. The algorithm combines both gradient-descent and Newton-like behaviors as well as inertia. We prove the convergence of INDIAN to critical points for most deep learning problems. To do so, we provide a well-suited framework to analyze deep learning losses involving tame optimization in which we study the continuous dynamical system together with the discrete stochastic approximations. On the theoretical side, we also prove a sublinear convergence rate for the continuous time differential inclusion which underlies the algorithm. From an empirical point of view the algorithm shows promising results on popular DNN training benchmark problems.</p>
            </abstract>
            <particDesc>
              <org type="consortium">PNRIA</org>
            </particDesc>
          </profileDesc>
        </biblFull>
      </listBibl>
    </body>
    <back>
      <listOrg type="structures">
        <org type="regroupinstitution" xml:id="struct-443875" status="VALID">
          <idno type="ROR">https://ror.org/017tgbk05</idno>
          <orgName>Communauté d'universités et établissements de Toulouse</orgName>
          <orgName type="acronym">Comue de Toulouse</orgName>
          <desc>
            <address>
              <addrLine>41 Allée Jules Guesde, 31000 Toulouse</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.univ-toulouse.fr/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-81148" status="VALID">
          <idno type="IdRef">026404354</idno>
          <idno type="ISNI">0000000121902837</idno>
          <idno type="ROR">https://ror.org/0443n9e75</idno>
          <orgName>Université Toulouse Capitole</orgName>
          <orgName type="acronym">UT Capitole</orgName>
          <date type="start">1970-01-01</date>
          <desc>
            <address>
              <addrLine>2 rue du Doyen-Gabriel-Marty - 31042 Toulouse Cedex 9</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.ut-capitole.fr/</ref>
          </desc>
          <listRelation>
            <relation name="443875" active="#struct-443875" type="direct"/>
          </listRelation>
        </org>
        <org type="institution" xml:id="struct-217752" status="OLD">
          <idno type="IdRef">026404672</idno>
          <idno type="ISNI">0000000121617331</idno>
          <idno type="ROR">https://ror.org/02v6kpv12</idno>
          <idno type="Wikidata">Q1273188</idno>
          <orgName>Université Toulouse III - Paul Sabatier</orgName>
          <orgName type="acronym">UT3</orgName>
          <date type="end">2025-01-01</date>
          <desc>
            <address>
              <addrLine>118 route de Narbonne - 31062 Toulouse</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.univ-tlse3.fr/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-443875" type="direct"/>
          </listRelation>
        </org>
      </listOrg>
      <listOrg type="projects">
        <org type="anrProject" xml:id="projanr-50372" status="VALID">
          <idno type="anr">ANR-19-P3IA-0004</idno>
          <orgName>ANITI</orgName>
          <desc>Artificial and Natural Intelligence Toulouse Institute</desc>
          <date type="start">2019</date>
        </org>
        <org type="europeanProject" xml:id="projeurop-710321" status="VALID">
          <idno type="number">681839</idno>
          <idno type="program">ERC-2015-CoG</idno>
          <idno type="call">ERC-2015-CoG</idno>
          <orgName>FACTORY</orgName>
          <desc>New paradigms for latent factor estimation</desc>
          <date type="start">2016-09-01</date>
          <date type="end">2022-08-31</date>
        </org>
      </listOrg>
    </back>
  </text>
</TEI>