-
일시: 2024년 11월 7일(목), 10:30-12:30
-
장소: 판교 테크노밸리 산업수학혁신센터 세미나실
- 경기 성남시 수정구 대왕판교로 815, 기업지원허브 231호 국가수리과학연구소
- 무료주차는 2시간 지원됩니다.
-
발표자: 손지용 교수(연세대학교)
-
주요내용: Optimizing Embeddings using Contrastive Loss
In this presentation, I will delve into recent research papers that provide theoretical analyses of embeddings trained using contrastive losses, which have gained significant traction across a broad range of modern applications. Specifically, I will focus on two widely-used losses: the InfoNCE loss and the sigmoid loss. In Part 1, I will explore the dynamics of mini-batch optimization within the context of contrastive learning using the InfoNCE loss, highlighting key findings and behaviors. In Part 2, I will shift focus to the behavior and properties of embeddings trained with contrastive learning utilizing the sigmoid loss. By examining both optimization approaches, this presentation will offer deeper insights into the fundamental principles and performance implications of contrastive learning techniques in the domains of machine learning and representation learning.
