Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Training Generative Adversarial Networks by Solving Ordinary Differential Equations
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 |
|---|---|---|---|
| Use RK4 rather than Euler (ablation) | ODE-GAN on CIFAR-10 with DCGAN · CIFAR-10 unconditional, DCGAN | FID (best): 19.43 ± 0.03 -> 17.66 ± 0.38 | Helped |
| Use RK4 rather than RK2 (ablation) | ODE-GAN on CIFAR-10 with DCGAN · CIFAR-10 unconditional, DCGAN | FID (best): 18.66 ± 0.24 -> 17.66 ± 0.38 | Helped |
| Increase gradient regularisation weight from λ=0.001 to λ=0.01 (ablation) | ODE-GAN (RK4) on CIFAR-10 with DCGAN · CIFAR-10 unconditional, DCGAN | FID (best): 25.09 ± 0.61 -> 17.66 ± 0.38 | Helped |
| Increase gradient regularisation weight from λ=0.005 to λ=0.01 (ablation) | ODE-GAN (RK4) on CIFAR-10 with DCGAN · CIFAR-10 unconditional, DCGAN | FID (best): 18.34 ± 0.24 -> 17.66 ± 0.38 | Helped |
| Use ODE-GAN (RK4) rather than SN-GAN | DCGAN · CIFAR-10 unconditional, DCGAN | FID (best): 21.71 ± 0.61 -> 17.66 ± 0.38 | Helped |
| Use ODE-GAN (RK4) rather than SN-GAN | ResNet · CIFAR-10 unconditional, ResNet | FID (best): 15.97 ± 0.22 -> 11.85 ± 0.21 | Helped |
| Use ODE-GAN (RK4) rather than WGAN-ALP | ResNet · CIFAR-10 unconditional, ResNet | FID (best): 12.96 ± 0.35 -> 11.85 ± 0.21 | Helped |
| Use ODE-GAN (RK4) rather than SN-GAN | ResNet · ImageNet 128×128 conditional | FID (best): 37.05± 0.26 -> 26.16 ± 0.75 | Helped |
| Use ODE-GAN (RK4) rather than SN-GAN | ResNet (large) · ImageNet 128×128 conditional | FID (best): 45.62 -> 22.29 | Helped |
| Use ODE-GAN (RK4) rather than SN-GAN | ResNet (large) · ImageNet 128×128 conditional | IS (best): 24.64 -> 46.17 | Helped |
| Combine RK4 with Adam-adjusted gradients instead of using RK4 alone (ablation) | ODE-GAN on CIFAR-10 with DCGAN · CIFAR-10 unconditional, DCGAN | FID (final): 18.05 ± 0.53 -> 23.20 ± 0.95 | Hurt |
| Combine RK4 with Adam-adjusted gradients instead of using RK4 alone (ablation) | ODE-GAN on CIFAR-10 with ResNet · CIFAR-10 unconditional, ResNet | FID (final): 12.50 ± 0.30 -> 20.32 ± 1.17 | Hurt |
Closest papers by meaning
- Understanding and Stabilizing GANs’ Training Dynamics Using Control Theory (ICML 2020 · similarity 0.93)
- Smoothness and Stability in GANs (ICLR 2020 · similarity 0.93)
- Towards Principled Methods for Training Generative Adversarial Networks (ICLR 2017 · similarity 0.92)
- Which Training Methods for GANs do actually Converge? (ICML 2018 · similarity 0.91)
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.91)
- Unrolled Generative Adversarial Networks (ICLR 2017 · similarity 0.91)
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.