
<!DOCTYPE article
  PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.1 20151215//EN" "JATS-journalpublishing1.dtd">
<article article-type="research-article" dtd-version="1.1" xml:lang="en"><front xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:c="http://ns.iop.org/namespaces/content" xmlns:date="http://ns.iop.org/namespaces/functions/date" xmlns:m="http://ns.iop.org/namespaces/meta" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"><journal-meta><journal-id journal-id-type="publisher-id">erc</journal-id><journal-id journal-id-type="coden"/><journal-title-group><journal-title xml:lang="en">Environmental Research Communications</journal-title><abbrev-journal-title abbrev-type="IOP" xml:lang="en">ERC</abbrev-journal-title><abbrev-journal-title abbrev-type="publisher" xml:lang="en">Environ. Res. Commun.</abbrev-journal-title></journal-title-group><issn pub-type="epub">2515-7620</issn><publisher><publisher-name>IOP Publishing</publisher-name></publisher></journal-meta><article-meta><article-id pub-id-type="publisher-id">ercacf81b</article-id><article-id pub-id-type="doi">10.1088/2515-7620/acf81b</article-id><article-id pub-id-type="manuscript">acf81b</article-id><article-id pub-id-type="other">ERC-101442.R1</article-id><article-categories><subj-group subj-group-type="display-article-type"><subject>Paper</subject></subj-group><subj-group subj-group-type="section"><subject>Environmental engineering &amp; technology</subject></subj-group></article-categories><title-group><article-title>How to estimate carbon footprint when training deep learning models? A guide and review</article-title></title-group><contrib-group><contrib contrib-type="author" xlink:type="simple"><name name-style="western"><surname>Bouza</surname><given-names>Lucía</given-names></name><xref ref-type="aff" rid="affiliation01">1</xref></contrib><contrib contrib-type="author" corresp="yes" xlink:type="simple"><contrib-id authenticated="true" contrib-id-type="orcid">0000-0002-4858-4944</contrib-id><name name-style="western"><surname>Bugeau</surname><given-names>Aurélie</given-names></name><xref ref-type="aff" rid="affiliation02">2</xref><xref ref-type="aff" rid="affiliation03">3</xref><email>aurelie.bugeau@u-bordeaux.fr</email></contrib><contrib contrib-type="author" xlink:type="simple"><name name-style="western"><surname>Lannelongue</surname><given-names>Loïc</given-names></name><xref ref-type="aff" rid="affiliation04">4</xref><xref ref-type="aff" rid="affiliation05">5</xref><xref ref-type="aff" rid="affiliation06">6</xref><xref ref-type="aff" rid="affiliation07">7</xref></contrib><aff id="affiliation01">
               <label>1</label>
               <institution xlink:type="simple">Université Paris Cité</institution>, CNRS, MAP5 UMR 8145, 75006, Paris, <country>France</country>
            </aff><aff id="affiliation02">
               <label>2</label>
               <institution xlink:type="simple">Univ. Bordeaux</institution>, Bordeaux INP, CNRS, LaBRI, Talence, <country>France</country>
            </aff><aff id="affiliation03">
               <label>3</label>
IUF, <country>France</country>
            </aff><aff id="affiliation04">
               <label>4</label>
Cambridge Baker Systems Genomics Initiative, Department of Public Health and Primary Care, <institution xlink:type="simple">University of Cambridge</institution>, Cambridge, <country>United Kingdom</country>
            </aff><aff id="affiliation05">
               <label>5</label>
British Heart Foundation Cardiovascular Epidemiology Unit, Department of Public Health and Primary Care, <institution xlink:type="simple">University of Cambridge</institution>, Cambridge, <country>United Kingdom</country>
            </aff><aff id="affiliation06">
               <label>6</label>
Victor Phillip Dahdaleh Heart and Lung Research Institute, <institution xlink:type="simple">University of Cambridge</institution>, Cambridge, <country>United Kingdom</country>
            </aff><aff id="affiliation07">
               <label>7</label>
Health Data Research UK Cambridge, <institution xlink:type="simple">Wellcome Genome Campus and University of Cambridge</institution>, Cambridge, <country>United Kingdom</country>
            </aff></contrib-group><pub-date pub-type="ppub"><day>1</day><month>11</month><year>2023</year></pub-date><pub-date pub-type="epub"><day>21</day><month>11</month><year>2023</year></pub-date><pub-date pub-type="open-access"><day>21</day><month>11</month><year>2023</year></pub-date><volume>5</volume><issue>11</issue><elocation-id content-type="artnum">115014</elocation-id><history><date date-type="received"><day>27</day><month>6</month><year>2023</year></date><date date-type="rev-recd"><day>23</day><month>8</month><year>2023</year></date><date date-type="accepted"><day>8</day><month>9</month><year>2023</year></date><date date-type="oa-requested"><day>25</day><month>08</month><year>2023</year></date></history><permissions><copyright-statement>© 2023 The Author(s). Published by IOP Publishing Ltd</copyright-statement><copyright-year>2023</copyright-year><license license-type="cc-by" xlink:href="http://creativecommons.org/licenses/by/4.0/" xlink:type="simple"><license-p>
                  <graphic content-type="print" orientation="portrait" position="float" xlink:href="https://content.cld.iop.org/journals/2515-7620/5/11/115014/revision2/ercacf81blicense.eps?Expires=1701173532&amp;Signature=LIAkX8wnXd-8JJLEilJk9TFsGdn4CeEk79cLMJDH4LqcCfZVGj6cyIqVKlyvNPnJIIvXVqmdRVMGFbJXpmA4uPRV9iMbJJTgiJHeu3VgwjCeEPaRFaGWdE-i7mNsXQdMy69l431ST2cXrMb5D0DQm0ruEP81BAA15Xla86s1SNkYWZUAGzcNxmIsTrhQZFM8p52AtbAO6-dSmAvelxQa8VAOxPdDasTDL2-fSuHs30dDF-7OSOEPV62kDt9vRb-AR6GuV2diGePzY2TEzHU3oaJEiKxaBdW8qaE5UG6uO0Ov6pMy~J8Ia5FTE8dEiuOFwXKinIeF3KKp0MklDu5Viw__&amp;Key-Pair-Id=KL1D8TIY3N7T8" xlink:type="simple"/>
                  <graphic content-type="online" orientation="portrait" position="float" xlink:href="https://content.cld.iop.org/journals/2515-7620/5/11/115014/revision2/ercacf81blicense.gif?Expires=1701173532&amp;Signature=n3AAfsGtneUkDVtz34wgOYFUHQbZ1ArBoex27W6eE608QP7h8VOTRT1scoReqPvVhJREnBzJl7noNsenb0yXp4-8StBdwky~3Gn7hCHETtA93yBfkCsil4zWzOy4oZsh83HnExXmSXqZC0c85gIG4MkLMexLyvZidJSwSuMBVgVrerybhSohntwplQ1OfAQAz7QONQCIdT~VoXqRtxEwE2bXTd2mFs6dn7Fk7Xzv1INUPsu-XMtU7KaUZbg-Ts12gEgpLBMSLIvtnDfyM3b9b6~sVVcoWxgua5vNmQ8stCHq~pmwNmL6e8~xaEG~2SczGwbrjOJlX31ZCXWQC-0~rw__&amp;Key-Pair-Id=KL1D8TIY3N7T8" xlink:type="simple"/>Original content from this work may be used under the terms of the <ext-link ext-link-type="uri" xlink:href="http://creativecommons.org/licenses/by/4.0" xlink:type="simple">Creative Commons Attribution 4.0 licence</ext-link>. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.</license-p></license></permissions><self-uri content-type="pdf" xlink:href="https://content.cld.iop.org/journals/2515-7620/5/11/115014/revision2/erc_5_11_115014.pdf?Expires=1701173532&amp;Signature=P~SbG~-p1ClGpc9v8Rpxd2qSGOR4Y24e8UpB1iLal1Uhbv9aOxJO4MyDI7GK8Tz-NtRTWQo18HxpB~BC2T-MBksUVFPL1qYmBWlVUtqsh5eX-Of3TnJlpBJ1Esb1~XODr7KyW467IgpnyWIBxUlMjxteyE-k6fgYIVh12~cVZQSnuLNQnalFncrbX3JKP0MwB~aTm-MsWRh4Nj5fhHlE~j4eQHzUqARh3LEXvKOvsgRNNstzjA7f~pAkhvCVIJoLSBp4t4xeeGw95B4GgdC4aPOk9tSL4HICou4Ub36mGeCfTOMqY-i~94POAYYKpmKngkUh4EJwoj5UfcYUIItKmA__&amp;Key-Pair-Id=KL1D8TIY3N7T8" xlink:type="simple"/><abstract><title>Abstract</title><p>Machine learning and deep learning models have become essential in the recent fast development of artificial intelligence in many sectors of the society. It is now widely acknowledge that the development of these models has an environmental cost that has been analyzed in many studies. Several online and software tools have been developed to track energy consumption while training machine learning models. In this paper, we propose a comprehensive introduction and comparison of these tools for AI practitioners wishing to start estimating the environmental impact of their work. We review the specific vocabulary, the technical requirements for each tool. We compare the energy consumption estimated by each tool on two deep neural networks for image processing and on different types of servers. From these experiments, we provide some advice for better choosing the right tool and infrastructure.</p></abstract><kwd-group kwd-group-type="author"><kwd>AI carbon footprint</kwd><kwd>measuring electrical consumption of AI</kwd><kwd>environmental impacts of deep learning</kwd></kwd-group><funding-group><award-group xlink:type="simple"><funding-source xlink:type="simple">Agence Nationale de la Recherche<named-content content-type="funder-id" xlink:type="simple">https://doi.org/10.13039/501100001665</named-content>
               </funding-source><award-id>ANR-19-CE23-0027-01</award-id></award-group><award-group xlink:type="simple"><funding-source xlink:type="simple">Cambridge British Heart Foundation Centre of Research Excellence</funding-source><award-id>RE/18/1/34212</award-id></award-group><award-group xlink:type="simple"><funding-source xlink:type="simple">NIHR Cambridge Biomedical Research Centre<named-content content-type="funder-id" xlink:type="simple">https://doi.org/10.13039/501100018956</named-content>
               </funding-source><award-id>BRC-1215-20014</award-id><award-id>NIHR203312</award-id></award-group><award-group xlink:type="simple"><funding-source xlink:type="simple">British Heart Foundation<named-content content-type="funder-id" xlink:type="simple">https://doi.org/10.13039/501100000274</named-content>
               </funding-source><award-id>CH/12/2/29428</award-id><award-id>RG/18/13/33946</award-id></award-group></funding-group><counts><page-count count="26"/></counts><custom-meta-group><custom-meta xlink:type="simple"><meta-name>ccc</meta-name><meta-value>2515-7620/23/115014+26$33.00</meta-value></custom-meta><custom-meta xlink:type="simple"><meta-name>crossmark</meta-name><meta-value>yes</meta-value></custom-meta></custom-meta-group></article-meta></front></article>