GANs beyond divergence minimization
This work addresses a theoretical inconsistency in GANs for machine learning researchers, offering a broader framework that could lead to more flexible model designs, though it is incremental in refining existing theory.
The paper challenges the prevailing view that GANs operate by minimizing a divergence, showing that common generator loss functions are not divergences and do not require such minimization to generate realistic data. It introduces four new classes of loss functions, most of which are divergences, and demonstrates on synthetic and CIFAR-10 datasets that these converge well and achieve comparable quality to traditional methods.
Generative adversarial networks (GANs) can be interpreted as an adversarial game between two players, a discriminator D and a generator G, in which D learns to classify real from fake data and G learns to generate realistic data by "fooling" D into thinking that fake data is actually real data. Currently, a dominating view is that G actually learns by minimizing a divergence given that the general objective function is a divergence when D is optimal. However, this view has been challenged due to inconsistencies between theory and practice. In this paper, we discuss of the properties associated with most loss functions for G (e.g., saturating/non-saturating f-GAN, LSGAN, WGAN, etc.). We show that these loss functions are not divergences and do not have the same equilibrium as expected of divergences. This suggests that G does not need to minimize the same objective function as D maximize, nor maximize the objective of D after swapping real data with fake data (non-saturating GAN) but can instead use a wide range of possible loss functions to learn to generate realistic data. We define GANs through two separate and independent D maximization and G minimization steps. We generalize the generator step to four new classes of loss functions, most of which are actual divergences (while traditional G loss functions are not). We test a wide variety of loss functions from these four classes on a synthetic dataset and on CIFAR-10. We observe that most loss functions converge well and provide comparable data generation quality to non-saturating GAN, LSGAN, and WGAN-GP generator loss functions, whether we use divergences or non-divergences. These results suggest that GANs do not conform well to the divergence minimization theory and form a much broader range of models than previously assumed.