主讲人
Yisen Wang
Peking University
时间
2026年4月20日 星期一
上午 10:00-11:00
地点
学院104会议室
Abstract
Despite remarkable empirical success of Self-Supervised Learning (SSL), its theoretical foundations remain relatively underexplored. This gap raises fundamental questions about when and why SSL works, and what governs its generalization and robustness. In this talk, I will introduce representative SSL methodologies widely used in foundation models, and then present a series of our recent works on the theoretical understanding of SSL, with a particular focus on contrastive learning, masked modeling and autoregressive learning.
Biography
Yisen Wang is an Assistant Professor at Peking University (PKU). His research focuses on the theoretical foundations of representation learning and safety. He has published around 50 JMLR/TPAMI/ICML/NeurIPS/ICLR papers and earned 14k+ Google Scholar citations. He also received 5 Best Paper or Runner-up honors. Additionally, he serves as Senior Area Chair for NeurIPS and Associate Editor for TPAMI.
