FLUID FLOW MASS TRANSPORT FOR GENERATIVE NETWORKS

Jingrong Lin, Keegan Lensink, Eldad Haber

Keywords: generative network, optimal mass transport, gaussian mixture, model matching

Abstract: Generative Adversarial Networks have been shown to be powerful tools for generating content resulting in them being intensively studied in recent years. Training these networks requires maximizing a generator loss and minimizing a discriminator loss, leading to a difficult saddle point problem that is slow and difficult to converge. Motivated by techniques in the registration of point clouds and the fluid flow formulation of mass transport, we investigate a new formulation that is based on strict minimization, without the need for the maximization. This formulation views the problem as a matching problem rather than an adversarial one, and thus allows us to quickly converge and obtain meaningful metrics in the optimization path.

Code: https://github.com/Jingrong-LIN/cGAN/blob/master/GANS2Dsubmision.ipynb