Research map / Optimization / Optimizers
Which Training Methods for GANs do actually Converge?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply unregularized gradient-based GAN training when distributions may lie on lower-dimensional manifolds | GAN training · Dirac-GAN | local convergence | Hurt |
| Use WGAN with a fixed finite number of discriminator updates per generator update | Dirac-GAN · Dirac-GAN | local convergence | Hurt |
| Use WGAN-GP with a fixed finite number of discriminator updates per generator update | Dirac-GAN · Dirac-GAN | local convergence | Hurt |
| Add Gaussian instance noise | GAN training · Dirac-GAN | local convergence | Helped |
| Add a zero-centered discriminator gradient penalty | GAN training · Dirac-GAN | local convergence | Helped |
| Use consensus optimization | GAN training · Dirac-GAN | local convergence | Helped |
| Replace the generator objective with the non-saturating objective | GAN training · Dirac-GAN | continuous-time convergence | Helped |
| Penalize discriminator gradients on real data only | GAN training · General GANs with lower-dimensional supports | local convergence | Helped |
| Penalize discriminator gradients on generated data only | GAN training · General GANs with lower-dimensional supports | local convergence | Helped |
| Use R1 or R2 gradient regularization instead of unregularized training or WGAN-GP | 2D GANs · Four 2D data distributions | Wasserstein-1-distance | Helped |
| Use R1 regularization | Convolutional GANs · Image datasets, including ImageNet and CelebA-HQ | training stability | Helped |
Closest papers by meaning
- Smoothness and Stability in GANs (ICLR 2020 · similarity 0.92)
- Gradient descent GAN optimization is locally stable (NeurIPS 2017 · similarity 0.92)
- Training Generative Adversarial Networks by Solving Ordinary Differential Equations (NeurIPS 2020 · similarity 0.91)
- Consistency Regularization for Generative Adversarial Networks (ICLR 2020 · similarity 0.91)
- CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 2024 · similarity 0.89)
- Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.