Realizing GANs via a Tunable Loss Function

2021/06/09 by Kurri, Gowtham R., Sypherd, Tyler, Sankar, Lalitha · 1 citation
#FOS: Computer and information sciences #Information Theory (cs.IT) #Machine Learning (cs.LG) #Machine Learning (stat.ML)

paper · doi:10.48550/arxiv.2106.05232

Abstract

We introduce a tunable GAN, called α-GAN, parameterized by α∈ (0,∞], which interpolates between various f-GANs and Integral Probability Metric based GANs (under constrained discriminator set). We construct α-GAN using a supervised loss function, namely, α-loss, which is a tunable loss function capturing several canonical losses. We show that α-GAN is intimately related to the Arimoto divergence, which was first proposed by Österriecher (1996), and later studied by Liese and Vajda (2006). We also study the convergence properties of α-GAN. We posit that the holistic understanding that α-GAN introduces will have practical benefits of addressing both the issues of vanishing gradients and mode collapse.

Cited by

• α-GAN by Rényi Cross Entropy

Related

• α-GAN: Convergence and Estimation Guarantees
• f-GANs in an Information Geometric Nutshell
• (α_D,α_G)-GANs: Addressing GAN Training Instabilities via Dual Objectives
• Addressing GAN Training Instabilities via Tunable Classification Losses
• GAN-QP: A Novel GAN Framework without Gradient Vanishing and Lipschitz Constraint