On Relativistic $f$-Divergences

Abstract

This paper provides a more rigorous look at Relativistic Generative Adversarial Networks (RGANs). We prove that the objective function of the discriminator is a statistical divergence for any concave function $f$ with minimal properties ($f(0)=0$, $f^{\prime }(0)\ne 0$, ${\sup }_{x}f(x)>0$). We also devise a few variants of relativistic $f$-divergences. Wasserstein GAN was originally justified by the idea that the Wasserstein distance (WD) is most sensible because it is weak (i.e., it induces a weak topology). We show that the WD is weaker than $f$-divergences which are weaker than relativistic $f$-divergences. Given the good performance of RGANs, this suggests that WGAN does not performs well primarily because of the weak metric, but rather because of regularization and the use of a relativistic discriminator. We also take a closer look at estimators of relativistic $f$-divergences. We introduce the minimum-variance unbiased estimator (MVUE) for Relativistic paired GANs (RpGANs; originally called RGANs which could bring confusion) and show that it does not perform better. Furthermore, we show that the estimator of Relativistic average GANs (RaGANs) is only asymptotically unbiased, but that the finite-sample bias is small. Removing this bias does not improve performance.

arXiv commentCode is available on: https://github.com/AlexiaJM/relativistic-f-divergences

References in corpus (6)

• Adam: A Method for Stochastic Optimization
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Spectral Normalization for Generative Adversarial Networks
• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• The relativistic discriminator: a key element missing from standard GAN
• Towards Principled Methods for Training Generative Adversarial Networks

Cited by in corpus (4)

• Adversarial score matching and improved sampling for image generation
• Gradient penalty from a maximum margin perspective
• The Benefits of Pairwise Discriminators for Adversarial Training
• Class Balancing GAN with a Classifier in the Loop