一、题目

Learn the Dynamics and Dynamics in Deep Learning

二、报告人

Qi Meng (Microsoft Research Asia)

三、摘要

In the big data and AI era, researchers are investigating the theoretical foundation of deep learning, to understand why deep learning works and further improve the algorithms. We leverage the mathematical tools from dynamic system to analyze the dynamics of optimization algorithms in deep learning. Our study on power-law dynamic and max-margin solution explains the implicit bias that optimization algorithms (e.g., SGD, Adam) can converge to the solution with low complexity, to guarantee the generalization. Except for understanding the dynamics in deep learning, using deep learning to model the complex dynamics or accelerate dynamics simulation attracts much attention recently. Since symmetry plays an essential role in physical systems, we design symmetry-preserving deep learning models NNPhD, EVFN for physical system modeling.

四、主讲人介绍

Qi Meng (孟琪) is a senior researcher in Computational and Learning Group (COLT) in Microsoft Research Asia (MSRA). Before she joined Microsoft in July 2018, she obtained her Ph.D. in probability and mathematical statistics from Peking University, supervised by Prof. Zhi-Ming Ma (马志明), Chinese Academy of Sciences. Her main research interests include deep learning theory, learning dynamics, optimization theory and statistics.

五、邀请人

王光辉 数学学院教授

六、时间

12月13日（周一）19：00

七、地点

腾讯会议：709 946 788 （password 123456）

八、主办方

山东大学数学学院

山东大学数据科学研究院