一楼一凤

Heavy-Tailed Methods in Machine Learning

活动时间:2026-07-12 15:00

活动地点:2号学院楼2432

主讲人:朱凌炯

主讲人中文简介:

朱凌炯,Florida State University数学系讲席教授。2008年本科毕业于英国剑桥大学数学系,后在美国纽约大学Courant数学研究所师从著名数学家S.R.S.Varadhan,2013年获数学博士学位。2014-2015年在美国明尼苏达大学担任Dunham Jackson助理教授。2015年加入Florida State University,历任助理教授,副教授,教授与思考机器杰出学者。他在概率论与随机过程期刊(AAP,SPA,AIHP等),运筹学期刊(OR,POM等),机器学习期刊与会议(JMLR,ICML,NIPS等),金融工程期刊(FS,SIFIN等)顶级期刊发表论文数篇.

活动内容摘要:

Recent studies have shown that heavy tails can emerge in stochastic optimization and that the heaviness of the tails have links to the generalization error in machine learning. We first study the generalization performance of heavy tailed SGD through the lens of algorithmic stability, and show rigorously that heavy tails can help generalization performance. We will also study differential privacy for heavy tailed SGD. Finally, we will discuss fractional Langevin algorithms and the implications in optimization when the SGD has heavy tails.

主持人:吕吴俊