ODE-based Learning to Optimize
科研大讨论系列报告
报告题目(Title):ODE-based Learning to Optimize
报告人(Speaker):文再文 教授(北京大学)
地点(Place):后主楼1124
时间(Time):3月26日(星期二),10:00am-11:00am
邀请人(Inviter):蔡勇勇
报告摘要
In recent years, ordinary differential equation (ODE) has become a promising starting point to understand the nature of acceleration methods. However, there still exits gap between ODE and optimization methods. In this talk, we introduce the idea behind learn to optimize and optimization-inspired ODE, and try to provide a framework that automatically looks for efficient problem-orientational optimization methods with a guarantee of worst-case convergence.
主讲人简介
文再文,北京大学北京国际数学研究中心教授,主要研究最优化算法与理论及其在机器学习、人工智能中的应用。2016年获中国青年科技奖。2020年获国家万人计划科技创新领军人才,现为中国运筹学会常务理事,中国运筹学会数学规划分会副理事长。