Instability and Local Minima in GAN Training with Kernel Discriminators
This work addresses training instability for GAN practitioners, but it is incremental as it builds on existing analysis frameworks.
The paper tackled the instability and local minima in GAN training by analyzing joint dynamics with kernel discriminators under a discrete, finite sample model, introducing the Isolated Points Model to characterize convergence conditions and explain failure modes like mode collapse and divergence, with numerical simulations replicating these behaviors.
Generative Adversarial Networks (GANs) are a widely-used tool for generative modeling of complex data. Despite their empirical success, the training of GANs is not fully understood due to the min-max optimization of the generator and discriminator. This paper analyzes these joint dynamics when the true samples, as well as the generated samples, are discrete, finite sets, and the discriminator is kernel-based. A simple yet expressive framework for analyzing training called the $\textit{Isolated Points Model}$ is introduced. In the proposed model, the distance between true samples greatly exceeds the kernel width, so each generated point is influenced by at most one true point. Our model enables precise characterization of the conditions for convergence, both to good and bad minima. In particular, the analysis explains two common failure modes: (i) an approximate mode collapse and (ii) divergence. Numerical simulations are provided that predictably replicate these behaviors.