Show simple item record

About contrastive unsupervised representation learning for classification and its convergence

dc.contributor.authorMerad, Ibrahim
dc.contributor.authorYu, Yiyang
hal.structure.identifierCEntre de REcherches en MAthématiques de la DEcision [CEREMADE]
dc.contributor.authorBacry, Emmanuel
HAL ID: 735850
ORCID: 0000-0001-5997-1942
hal.structure.identifierCentre de Mathématiques Appliquées - Ecole Polytechnique [CMAP]
hal.structure.identifierLaboratoire de Probabilités, Statistique et Modélisation [LPSM (UMR_8001)]
dc.contributor.authorGaïffas, Stéphane
dc.date.accessioned2022-02-22T15:36:28Z
dc.date.available2022-02-22T15:36:28Z
dc.date.issued2021
dc.identifier.urihttps://basepub.dauphine.psl.eu/handle/123456789/22718
dc.language.isoenen
dc.subjectunsupervised learningen
dc.subjectconstrative learningen
dc.subjectdeep neural networksen
dc.subjecttheoretical guaranteesen
dc.subject.ddc006.3en
dc.titleAbout contrastive unsupervised representation learning for classification and its convergenceen
dc.typeDocument de travail / Working paper
dc.description.abstractenContrastive representation learning has been recently proved to be very efficient for selfsupervised training. These methods have been successfully used to train encoders which perform comparably to supervised training on downstream classification tasks. A few works have started to build a theoretical framework around contrastive learning in which guarantees for its performance can be proven. We provide extensions of these results to training with multiple negative samples and for multiway classification. Furthermore, we provide convergence guarantees for the minimization of the contrastive training error with gradient descent of an overparametrized deep neural encoder, and provide some numerical experiments that complement our theoretical findings.en
dc.publisher.cityParisen
dc.identifier.citationpages17en
dc.relation.ispartofseriestitleCahier de recherche CEREMADE, Université Paris Dauphine-PSLen
dc.identifier.urlsitehttps://hal.archives-ouvertes.fr/hal-03438767en
dc.subject.ddclabelIntelligence artificielleen
dc.identifier.citationdate2021
dc.description.ssrncandidatenon
dc.description.halcandidatenonen
dc.description.readershiprechercheen
dc.description.audienceInternationalen
dc.date.updated2022-02-22T15:25:11Z
hal.author.functionaut
hal.author.functionaut
hal.author.functionaut
hal.author.functionaut


Files in this item

Thumbnail
Name:
2012.01064.pdf
Size:
675.6Kb
Format:
PDF
View/Open

This item appears in the following Collection(s)

Show simple item record