徐礼虎: Error estimates between SGD with momentum and underdamped Langevin diffusion

报告时间：2024年12月24日（星期二） 14:30-15:30 报告地点：翡翠湖校区科教楼B座1710室 报 告 人：徐礼虎 副教授 工作单位：University of Macau 举办单位：数学学院 报告简介： Stochastic gradient descent with momentum is a popular variant of stochastic gradient descent, which has recently been reported to have a close relationship with the underdamped Langevin diffusion. In this paper, we establish a quantitative error estimate between them in the 1-Wasserstein and total variation distances. 报告人简介： 徐礼虎，2008年博士毕业于Imperial College London，2007-2011在Bonn University做博士后；2014年至今，澳门大学科技学院副教授。主持两项国自然科学面上基金，研究方向主要包括：（1）Applied Probability (analysis of stochastic algorithms in machine learning, heavy tailed distribution, rare events analysis)；（2）Statistics and Data Science (high dimensional robust estimation, high dimensional distribution estimation via DNN, Stein’s method in data science)；（3）Stochastic Dynamical Systems (ergodic theory of SPDEs, numerical methods for SDEs, Malliavin calculus)；在Annals of Statistics, Journal of Machine Learning Research, Annals of Applied Probability, Bernoulli, Mathematics of Operations Research, Probability Theory and Related Fields, Transactions in American Mathematical Society, Stochastic Processes and Their Applications, Electronic Journal of Statistics, Electronic Journal of Probability等期刊共计发表50余篇论文。