BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Toward a Theoretical Understanding of Self-Supervised Learning in 
 the Foundation Model Era - Yisen Wang
DTSTART:20260128T170000Z
DTEND:20260128T180000Z
UID:TALK243901@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:Self-supervised learning (SSL) has become the cornerstone of m
 odern foundation models\, enabling them to learn powerful representations 
 from vast amounts of unlabeled data. By designing auxiliary tasks on raw i
 nputs\, SSL removes the reliance on human-provided labels and underpins th
 e pretraining–finetuning paradigm that has reshaped machine learning bey
 ond the traditional empirical risk minimization framework. Despite its rem
 arkable empirical success\, its theoretical foundations remain relatively 
 underexplored. This gap raises fundamental questions about when and why SS
 L works\, and what governs its generalization and robustness. In this talk
 \, I will introduce representative SSL methodologies widely used in founda
 tion models\, and then present a series of our recent works on the theoret
 ical understanding of SSL\, with a particular focus on contrastive learnin
 g\, masked autoencoders and autoregressive learning.\n
LOCATION:Computer Laboratory\, William Gates Building\, Room FW26
END:VEVENT
END:VCALENDAR