KRnet and its applications
报告人：Xiaoliang Wan (Louissiana State University)
时间：2022-09-14 10:00-11:00
地点：Tencent
Abstract: In this talk, we discuss the application of KRnet in some scientific computing problems. KRnet defines an invertible transport map from a prior distribution to a target distribution. KRnet has two main favorable features for scientific computing: 1) The exact invertibility results in an explicit model for probability density function (PDF) or conditional PDF; 2) KRnet is able to generate samples efficiently for the induced distribution. These two features merge PDF approximation and sample generation, which are addressed separately by classical approaches. We apply KRnet to approximate Fokker-Planck equations and to improve the training set of deep learning techniques for the approximation of high-dimensional PDEs.
Tencent Information
Link: https://meeting.tencent.com/dm/94F8pt9PdllT
ID: 662-556-201
