Research map / Image generation / GANs
Stabilizing Training of Generative Adversarial Networks through Regularization
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 |
|---|---|---|---|
| Add the weighted discriminator gradient-norm regularizer | GAN · 2D submanifold mixture of Gaussians in 3D | training stability: around 50k iterations -> well beyond 200k iterations | Helped |
| Add the weighted discriminator gradient-norm regularizer | DCGAN · CIFAR-10 | cross-testing false-positive rate: 1.0 -> 0.0 | Helped |
| Add the weighted discriminator gradient-norm regularizer | GAN · GAN training | training time: roughly 1.4 | Hurt |
| Use a non-zero regularization level without annealing (ablation) | GAN · 2D submanifold mixture of Gaussians in 3D | training stability | Helped |
| Exponentially anneal the regularization level (ablation) | GAN · 2D submanifold mixture of Gaussians in 3D | training stability | Helped |
| Apply the regularized objective to the generator near convergence (ablation) | Regularized GAN · GAN training | training stability | Hurt |
| Use the non-saturating alternative generator objective with the regularized discriminator (ablation) | Regularized JS-GAN · GAN training | training stability | Helped |
| Add noise explicitly to discriminator training examples instead of using the analytic regularizer (ablation) | DCGAN · CIFAR-10 | sample quality | Hurt |
Closest papers by meaning
- Consistency Regularization for Generative Adversarial Networks (ICLR 2020 · similarity 0.94)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.93)
- Regularizing Generative Adversarial Networks Under Limited Data (CVPR 2021 · similarity 0.93)
- Mode Regularized Generative Adversarial Networks (ICLR 2017 · similarity 0.93)
- On Stabilizing Generative Adversarial Training With Noise (CVPR 2019 · similarity 0.92)
- MS$^3$D: A RG Flow-Based Regularization for GAN Training with Limited Data (ICML 2024 · similarity 0.90)
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.