美国纽约州立大学石溪分校计算机系顾险峰教授来校做学术报告

活动信息

  • 开始时间:网上投稿
  • 活动地点:学术报告
  • 主讲人:雷娜

活动简介

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.

主讲人介绍

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 connect...