报告人
杨俊驰 助理教授
主持人
张思奇 助理教授
时间
12月26日(周四) 16:30-18:30
地点
协鑫楼108
Near-Optimal and Tuning-Free Min-Max Optimization
报告摘要
Machine learning applications have introduced new optimization challenges, particularly in the realm of trustworthy machine learning, which often involves min-max formulations aimed at finding equilibrium solutions. In this talk, we will first provide an overview of min-max optimization and its key applications, including federated learning, adversarial learning, and beyond. We will then discuss the limitations of classical algorithms when addressing challenges such as nonconvexity and asymmetry, and introduce a novel algorithmic framework that achieves near-optimal complexity. Finally, we will present adaptive algorithm designs for min-max optimization that achieve near-optimal complexity without the need for tedious stepsize tuning.
(In the last 30 minutes, I will introduce The Chinese University of Hong Kong, Shenzhen (CUHK-Shenzhen), and provide an overview of the Master's and Ph.D. programs offered by the School of Data Science (SDS).)
报告人简介
杨俊驰博士现任香港中文大学（深圳）数据科学学院助理教授。他分别于瑞士苏黎世联邦理工学院（ETH Zurich）获得计算机科学博士学位、伊利诺伊大学香槟分校（UIUC）获得工业工程硕士学位，以及加州大学洛杉矶分校（UCLA）获得应用数学与经济学学士学位。在加入香港中文大学（深圳）之前，杨博士曾在阿贡国家实验室担任博士后研究员。他的研究致力于连续优化算法的理论和实践，以解决复杂的决策问题，特别是在机器学习和能源领域的应用。他的研究成果已发表在NeurIPS和ICLR等顶尖学术会议上。此外，杨博士担任多个顶级会议和期刊的审稿人，包括机器学习（如NeurIPS）、数学优化（如Mathematical Programming）和电力系统（如IEEE Transactions on Smart Grid）领域的权威刊物。
美编 | 朱杼慧
责编 | 李梦爽、唐迪明
