Deep learning approach for Bayesian inverse problems

Speaker:    Liang Yan, Southeast University

Time:         10:00-11:00 , May.27

Location:   Tencent Meeting

    Meeting ID：348-152-382

     Link：https://meeting.tencent.com/dm/n332A5U4oQQM

Host:          Yue Qiu

Abstract:

Obtaining samples from the posterior distribution of Bayesian inverse problems (BIPs) is a long-standing challenging, especially when the forward operator is modeled by partial differential equation (PDE). In this talk, we will show you how to leverage the deep learning’s capabilities to tackle this challenge. Several fast and efficient deep neural network (DNN)-based approaches for accelerating simulations in sample generation will be described. A novel framework based on invertible neural networks using normalizing flow is also demonstrated.

Bio:

Dr. Liang Yan obtained both his bachelor degree and doctoral degrees in mathematics from Lanzhou University in 2006 and 2011, respectively. He started working at Southeast University first as assistant professor (2011-2006) then as associate professor. Dr. Yan’s research interest lies in Bayesian modeling and computing, uncertainty quantification and inverse problems and he has published more than 30 papers in top tier journals such as IPs, SIAM journals, CMAME, et al. His research has been continuously supported by NSFC.