报告人：Xianfeng Gu  （State University of New York at Stony Brook)

题目：Optimal Mass Transportation and Deep Learning

摘要：The fundamental principle behind generative adversarial networks (GANs) is to manipulate probability measures, such as to transform distributions, measure the Wasserstein distance between distributions and so on. Optimal mass transportation theory offers a geometric framework to handle probability measures, which gives a unique point of view of interpreting GAN models. In this talk, the connection between Optimal mass transportation theory and the convex geometry will be discussed, a variational approach will be given which leads to the solution to the classical Monge-Ampere equation and the Wasserstein distance between distributions. The similarities between these computational approach and GAN model will be analyzed.