Reproduce “Wasserstein GAN” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Martin Arjovsky, Soumith Chintala, Léon Bottou
This work by Arjovsky, Chintala, and Bottou introduces the Wasserstein GAN (WGAN), a generative adversarial network that replaces traditional f-divergences with the Earth Mover's (Wasserstein-1) distance. The WGAN framework stabilizes GAN training, provides a loss metric that correlates with sample quality, and reduces mode collapse, particularly demonstrating robustness across various generator architectures.
Key insights
- The Wasserstein-1 distance induces a 'weaker' topology compared to f-divergences (KL, JS), ensuring continuity and differentiability of the loss function, which resolves the vanishing gradient problem.
- The WGAN critic's loss directly approximates the Earth Mover's distance, providing a reliable and interpretable metric that correlates with the visual quality of generated samples throughout training.
- Training the critic to optimality, as opposed to balancing discriminator and generator, provides stable, non-saturating gradients, facilitating robust generator updates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.