<?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-03735983</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="2026-05-15T12:53:56+02:00"/>
      </publicationStmt>
      <sourceDesc>
        <p part="N">HAL API Platform</p>
      </sourceDesc>
    </fileDesc>
  </teiHeader>
  <text>
    <body>
      <listBibl>
        <biblFull>
          <titleStmt>
            <title xml:lang="en">Interpolation, extrapolation, and local generalization in common neural networks</title>
            <author role="aut">
              <persName>
                <forename type="first">Laurent</forename>
                <surname>Bonnasse-Gahot</surname>
              </persName>
              <email type="md5">1824ac135d0a66b66b0923a86fa366fc</email>
              <email type="domain">ehess.fr</email>
              <idno type="idhal" notation="string">lbg</idno>
              <idno type="idhal" notation="numeric">179118</idno>
              <idno type="halauthorid" notation="string">12255-179118</idno>
              <idno type="ORCID">https://orcid.org/0000-0002-2936-2780</idno>
              <affiliation ref="#struct-1318"/>
            </author>
            <editor role="depositor">
              <persName>
                <forename>Laurent</forename>
                <surname>Bonnasse-Gahot</surname>
              </persName>
              <email type="md5">1824ac135d0a66b66b0923a86fa366fc</email>
              <email type="domain">ehess.fr</email>
            </editor>
          </titleStmt>
          <editionStmt>
            <edition n="v1" type="current">
              <date type="whenSubmitted">2022-07-21 21:43:04</date>
              <date type="whenModified">2024-03-18 10:24:06</date>
              <date type="whenReleased">2022-07-22 13:31:49</date>
              <date type="whenProduced">2022-07-21</date>
              <date type="whenEndEmbargoed">2022-07-21</date>
              <ref type="file" target="https://hal.science/hal-03735983v1/document">
                <date notBefore="2022-07-21"/>
              </ref>
              <ref type="file" subtype="author" n="1" target="https://hal.science/hal-03735983v1/file/preprint_neural_networks_interpolation.pdf" id="file-3735983-3261799">
                <date notBefore="2022-07-21"/>
              </ref>
              <ref type="annex" n="0" target="https://hal.science/hal-03735983v1/file/preprint_neural_networks_interpolation_si.pdf" id="file-3735983-3261800">
                <date notBefore="2022-07-21"/>
              </ref>
              <ref type="externalLink" target="http://arxiv.org/pdf/2207.08648"/>
            </edition>
            <respStmt>
              <resp>contributor</resp>
              <name key="809244">
                <persName>
                  <forename>Laurent</forename>
                  <surname>Bonnasse-Gahot</surname>
                </persName>
                <email type="md5">1824ac135d0a66b66b0923a86fa366fc</email>
                <email type="domain">ehess.fr</email>
              </name>
            </respStmt>
          </editionStmt>
          <publicationStmt>
            <distributor>CCSD</distributor>
            <idno type="halId">hal-03735983</idno>
            <idno type="halUri">https://hal.science/hal-03735983</idno>
            <idno type="halBibtex">bonnassegahot:hal-03735983</idno>
            <idno type="halRefHtml">2022</idno>
            <idno type="halRef">2022</idno>
            <availability status="restricted">
              <licence target="https://about.hal.science/hal-authorisation-v1/">HAL Authorization<ref corresp="#file-3735983-3261799"/><ref corresp="#file-3735983-3261800"/></licence>
            </availability>
          </publicationStmt>
          <seriesStmt>
            <idno type="stamp" n="CNRS">CNRS - Centre national de la recherche scientifique</idno>
            <idno type="stamp" n="EHESS">Ecole des Hautes Etudes en Sciences Sociales</idno>
            <idno type="stamp" n="CAMS" corresp="EHESS">Collection HAL du Centre d'Analyse et de Mathématique Sociales (CAMS, UMR CNRS-EHESS)</idno>
          </seriesStmt>
          <notesStmt/>
          <sourceDesc>
            <biblStruct>
              <analytic>
                <title xml:lang="en">Interpolation, extrapolation, and local generalization in common neural networks</title>
                <author role="aut">
                  <persName>
                    <forename type="first">Laurent</forename>
                    <surname>Bonnasse-Gahot</surname>
                  </persName>
                  <email type="md5">1824ac135d0a66b66b0923a86fa366fc</email>
                  <email type="domain">ehess.fr</email>
                  <idno type="idhal" notation="string">lbg</idno>
                  <idno type="idhal" notation="numeric">179118</idno>
                  <idno type="halauthorid" notation="string">12255-179118</idno>
                  <idno type="ORCID">https://orcid.org/0000-0002-2936-2780</idno>
                  <affiliation ref="#struct-1318"/>
                </author>
              </analytic>
              <monogr>
                <imprint/>
              </monogr>
              <idno type="arxiv">2207.08648</idno>
            </biblStruct>
          </sourceDesc>
          <profileDesc>
            <langUsage>
              <language ident="en">English</language>
            </langUsage>
            <textClass>
              <classCode scheme="halDomain" n="info.info-lg">Computer Science [cs]/Machine Learning [cs.LG]</classCode>
              <classCode scheme="halDomain" n="info.info-cv">Computer Science [cs]/Computer Vision and Pattern Recognition [cs.CV]</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>There has been a long history of works showing that neural networks have hard time extrapolating beyond the training set. A recent study by Balestriero et al. (2021) challenges this view: defining interpolation as the state of belonging to the convex hull of the training set, they show that the test set, either in input or neural space, cannot lie for the most part in this convex hull, due to the high dimensionality of the data, invoking the well known curse of dimensionality. Neural networks are then assumed to necessarily work in extrapolative mode. We here study the neural activities of the last hidden layer of typical neural networks. Using an autoencoder to uncover the intrinsic space underlying the neural activities, we show that this space is actually low-dimensional, and that the better the model, the lower the dimensionality of this intrinsic space. In this space, most samples of the test set actually lie in the convex hull of the training set: under the convex hull definition, the models thus happen to work in interpolation regime. Moreover, we show that belonging to the convex hull does not seem to be the relevant criteria. Different measures of proximity to the training set are actually better related to performance accuracy. Thus, typical neural networks do seem to operate in interpolation regime. Good generalization performances are linked to the ability of a neural network to operate well in such a regime.</p>
            </abstract>
          </profileDesc>
        </biblFull>
      </listBibl>
    </body>
    <back>
      <listOrg type="structures">
        <org type="laboratory" xml:id="struct-1318" status="VALID">
          <idno type="RNSR">199812885U</idno>
          <idno type="ROR">00vb8fc53</idno>
          <orgName>Centre d'Analyse et de Mathématique sociales</orgName>
          <orgName type="acronym">CAMS</orgName>
          <date type="start">1998-01-01</date>
          <desc>
            <address>
              <addrLine>54 boulevard Raspail 75006 Paris</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://cams.ehess.fr/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-99539" type="direct"/>
            <relation name="UMR8557" active="#struct-441569" type="direct"/>
          </listRelation>
        </org>
        <org type="institution" xml:id="struct-99539" status="VALID">
          <idno type="IdRef">026374889</idno>
          <idno type="ROR">https://ror.org/02d9dg697</idno>
          <orgName>École des hautes études en sciences sociales</orgName>
          <orgName type="acronym">EHESS</orgName>
          <date type="start">1975-01-01</date>
          <desc>
            <address>
              <addrLine>54, boulevard Raspail 75006 Paris</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.ehess.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>
      </listOrg>
    </back>
  </text>
</TEI>