机器学习与数据科学博士生系列论坛（第九十八期）—— Optimal Transport: From Foundations and Algorithms to Generative Applications.
报告人：戴锦阳（伊人直播
）
时间：2026-03-19 16:00-17:00
地点：腾讯会议：361-166-556
摘要：
Establishing a rigorous and computable metric between probability distributions is a central problem in evaluating and training machine learning models. Optimal Transport (OT) offers a powerful mathematical framework to address this, connecting optimization, probability, and partial differential equations. However, traditional OT formulations historically suffered from severe computational bottlenecks.
In this talk, we will explore how algorithmic breakthroughs have transformed OT into a highly scalable tool for modern AI. We will begin with the foundational Monge and Kantorovich problems, and demonstrate how entropic regularization and Sinkhorn's algorithm can be utilized to overcome traditional computational bottlenecks. We will also introduce the applications of OT theory in generative models, such as Wasserstein gradient flows and flow matching.
论坛简介：该线上论坛是由张志华教授机器学习实验室组织，每两周主办一次（除了公共假期）。论坛每次邀请一位博士生就某个前沿课题做较为系统深入的介绍，主题包括但不限于机器学习、高维统计学、运筹优化和理论计算机科学。
