Speaker: Bohan Wang University of Science and Technology of China and Microsoft Research Asia (COLT Group)
Time: 2021-12-20 14:30-2021-12-20 15:30
Venue: Tencent Meeting (ID: 756-289-160)

Abstract:

This talk presents my recent works on the interface. Specifically, this talk is divided into two parts: (1). in the first part, I will introduce the effect brought by the optimizer to the generalization, i.e., the implicit bias of optimizers; (2). in the second part, I will talk about our approach to apply optimization to analyze generalization, i.e., the optimal noise covariance of SGLD in terms of generalization.

Short Bio:

Bohan Wang is a first-year student of the joint Ph.D. program of University of Science and Technology of China and Microsoft Research Asia (COLT Group). He is broadly interested in deep learning theory, especially in the interface between optimization and generalization. So far, he has several papers published in flagship conferences and journal, including ICLR, Neurips, UAI, and Physics Review E, with an oral presentation work on ICML.