Generalization Error of GAN from the Discriminator's Perspective

07/08/2021
by   Hongkang Yang, et al.
4

The generative adversarial network (GAN) is a well-known model for learning high-dimensional distributions, but the mechanism for its generalization ability is not understood. In particular, GAN is vulnerable to the memorization phenomenon, the eventual convergence to the empirical distribution. We consider a simplified GAN model with the generator replaced by a density, and analyze how the discriminator contributes to generalization. We show that with early stopping, the generalization error measured by Wasserstein metric escapes from the curse of dimensionality, despite that in the long term, memorization is inevitable. In addition, we present a hardness of learning result for WGAN.

READ FULL TEXT

page 1

page 2

page 3

page 4

research
05/25/2022

Learning Distributions by Generative Adversarial Networks: Approximation and Generalization

We study how well generative adversarial networks (GAN) learn probabilit...
research
11/29/2020

Generalization and Memorization: The Bias Potential Model

Models for learning probability distributions such as generative models ...
research
12/22/2022

A Mathematical Framework for Learning Probability Distributions

The modeling of probability distributions, specifically generative model...
research
01/23/2017

Loss-Sensitive Generative Adversarial Networks on Lipschitz Densities

*New Theory Result* We analyze the generalizability of the LS-GAN, showi...
research
09/08/2021

AWGAN: Empowering High-Dimensional Discriminator Output for Generative Adversarial Networks

Empirically multidimensional discriminator (critic) output can be advant...
research
02/26/2018

PBGAN: Partial Binarization of Deconvolution Based Generators

The generator is quite different from the discriminator in a generative ...
research
11/07/2017

On the Discrimination-Generalization Tradeoff in GANs

Generative adversarial training can be generally understood as minimizin...

Please sign up or login with your details

Forgot password? Click here to reset