Wasserstein GAN
Martín Arjovsky, Soumith Chintala, Léon Bottou · arXiv (Cornell University) · 2017
We introduce a new algorithm named WGAN, an alternative to traditional GAN training. In this new model, we show that we can improve the stability of learning, get rid of problems like mode collapse, and provide meaningful learning curves useful for debugging and hyperparameter searches. Furthermore, we show that the corresponding optimization problem is sound, and provide extensive theoretical work highlighting the deep connections to other distances between distributions.
Cited by 367 works.
Authors
References
- AutoResearch Lite: a reproducible paper-inspired experiment runner (Adam mechanism reproduction and convergence-speed comparison)
- GAN(Generative Adversarial Nets)
- Auto-Encoding Variational Bayes
- Annealed importance sampling
- Optimal Transport: Old and New
- A kernel two-sample test
- Generative Moment Matching Networks
- Concrete Representation of Abstract (M)-Spaces (A characterization of the Space of Continuous Functions)
- Envelope Theorems for Arbitrary Choice Sets
- Unrolled Generative Adversarial Networks