BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:On theory of contrastive self-supervised learning
DTSTART:20230816T130000
DTEND:20230816T150000
DTSTAMP:20260510T025351Z
UID:7f560b9aae3d238f013b45c5f1669168d6dc2f6fa552d744ec8150e5
CATEGORIES:Conferences - Seminars
DESCRIPTION:Oguz Yüksel\nEDIC candidacy exam\nExam president: Prof. Volka
 n Cevher\nThesis advisor: Prof. Nicolas Flammarion\nCo-examiner: Prof. Len
 ka Zdeborova\n\nAbstract\nSelf-supervised learning has taken over classica
 l supervised\nlearning by enabling the use of large datasets needed to\ntr
 ain big models through a pretext task. These pretext tasks allow\nthe lear
 ning of representations that\, after adaptation with limited\ntask-specifi
 c labeled samples\, yield good downsteam performance\nacross various tasks
 . Contrastive learning\, in particular\, has\nbeen particularly effective 
 and has seen wide adoption from\npractitioners due to its simplicity and s
 uccess. As a result\, there\nhas been growing interest in developing a the
 ory of contrastive\nlearning. In this write-up\, some of the recent progre
 ss on the\ntheory of contrastive learning is summarized\, along with a\nfo
 llow-up on important open questions and proposals for future\ninvestigatio
 ns.\n\nBackground papers\n[1] T. Chen\, S. Kornblith\, M. Norouzi\, and G.
  Hinton\, "A simple framework for contrastive learning of visual represent
 ations\," in International conference on machine learning. PMLR\, 2020\, p
 p. 1597–1607.\n[2] N. Saunshi\, O. Plevrakis\, S. Arora\, M. Khodak\, an
 d H. Khandeparkar\, "A theoretical analysis of contrastive unsupervised re
 presentation learning\," in International Conference on Machine Learning. 
 PMLR\, 2019\, pp. 5628–5637.\n[3] J. Z. HaoChen\, C. Wei\, A. Gaidon\, a
 nd T. Ma\, "Provable guarantees for self-supervised deep learning with spe
 ctral contrastive loss\," Advances in Neural Information Processing System
 s\, vol. 34\, pp. 5000–5011\, 2021.
LOCATION:BC 133 https://plan.epfl.ch/?room==BC%20133
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR