Research map / Image generation / GANs
Understanding and Stabilizing GANs’ Training Dynamics Using Control Theory
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 closed-loop control via squared discriminator-output regularization | WGAN · CIFAR10 | FID: 105.21 -> 21.14± 1.84 | Helped |
| Add closed-loop control via squared discriminator-output regularization | SGAN · CIFAR10 | FID: 28.51 -> 21.52± 0.96 | Helped |
| Add closed-loop control regularization | WGAN · CIFAR10 | IS: 7.82 -> 8.49± .07 | Helped |
| Add closed-loop control regularization | SGAN · CIFAR10 | IS: 7.37 -> 8.47± .09 | Helped |
| Add closed-loop control regularization | Hinge GAN · CIFAR10 | IS: 8.22 -> 8.54± .03 | Helped |
| Apply CLC to a spectral-normalized GAN | SN-GAN with WGAN objective · CIFAR10 | IS: 3.29 -> 8.14± .02 | Helped |
| Vary the feedback coefficient across 2, 5, and 10 (ablation) | CLC-GAN with WGAN objective · CIFAR10 | IS: 8.42± .06; 8.49± .07; 8.38± .10 | No clear effect |
| Vary the feedback coefficient across 2, 5, and 10 (ablation) | CLC-GAN with Hinge objective · CIFAR10 | IS: 8.49± .08; 8.54± .03; 8.46± .00 | No clear effect |
| Use CLC regularization instead of Reg-GAN regularization | GANs on CelebA · CelebA | training iterations per second: 4 -> 8 | Helped |
Closest papers by meaning
- Smoothness and Stability in GANs (ICLR 2020 · similarity 0.93)
- Stabilizing GANs' Training with Brownian Motion Controller (ICML 2023 · similarity 0.93)
- Training Generative Adversarial Networks by Solving Ordinary Differential Equations (NeurIPS 2020 · similarity 0.93)
- On the Limitations of First-Order Approximation in GAN Dynamics (ICML 2018 · similarity 0.91)
- Which Training Methods for GANs do actually Converge? (ICML 2018 · similarity 0.90)
- On Stabilizing Generative Adversarial Training With Noise (CVPR 2019 · 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.