Investigating Shifts in GAN Output-Distributions
This addresses a fundamental unsolved question in GAN research, providing insights into limitations of current algorithms, though it is incremental in nature.
The paper tackles the problem of whether Generative Adversarial Networks (GANs) can truly capture real data distributions by investigating shifts between real and generated data distributions, finding large shifts in experiments across datasets and state-of-the-art GAN architectures.
A fundamental and still largely unsolved question in the context of Generative Adversarial Networks is whether they are truly able to capture the real data distribution and, consequently, to sample from it. In particular, the multidimensional nature of image distributions leads to a complex evaluation of the diversity of GAN distributions. Existing approaches provide only a partial understanding of this issue, leaving the question unanswered. In this work, we introduce a loop-training scheme for the systematic investigation of observable shifts between the distributions of real training data and GAN generated data. Additionally, we introduce several bounded measures for distribution shifts, which are both easy to compute and to interpret. Overall, the combination of these methods allows an explorative investigation of innate limitations of current GAN algorithms. Our experiments on different data-sets and multiple state-of-the-art GAN architectures show large shifts between input and output distributions, showing that existing theoretical guarantees towards the convergence of output distributions appear not to be holding in practice.