Research map / Optimization / Optimizers
On gradient regularizers for MMD GANs
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 spectral normalization parametrization (ablation) | SMMDGAN · CelebA 160x160 | FID: 18.4±0.2 -> 12.4±0.2 | Helped |
| add spectral normalization parametrization (ablation) | SMMDGAN · ImageNet 64x64 | FID: 38.4±0.3 -> 36.6±0.2 | Helped |
| add spectral normalization parametrization (ablation) | SMMDGAN · CIFAR-10 | FID: 31.5±0.4 -> 25.0±0.3 | Helped |
| replace additive gradient penalty with scaled MMD (SMMD) critic constraint | MMDGAN-GP-L2 · CelebA 160x160 | FID: 20.5±0.2 -> 18.4±0.2 | Helped |
| scaled MMD with spectral normalization vs SN-GAN | SN-GAN · CelebA 160x160 | FID: 22.6±0.1 -> 12.4±0.2 | Helped |
| scaled MMD with spectral normalization vs SN-GAN | SN-GAN · CIFAR-10 | FID: 26.7±0.2 -> 25.0±0.3 | Mixed |
| scaled MMD with spectral normalization vs SN-GAN | SN-GAN · ImageNet 64x64 | FID: 47.5±0.1 -> 36.6±0.2 | Helped |
| use Scaled WGAN (linear kernel SMMD) with SN instead of WGAN-GP | WGAN-GP · CelebA 160x160 | FID: 29.2±0.2 -> 14.1±0.2 | Helped |
| add spectral normalization alone to WGAN, Sobolev-GAN, MMDGAN without scaled gradient control (ablation) | WGAN/Sobolev-GAN/MMDGAN · CelebA, CIFAR-10 | training stability | Hurt |
| add learned gamma with gradient and L2 penalty to SN variants (ablation) | MMDGAN · CIFAR-10, CelebA | sample quality | No clear effect |
| increase critic output dimension or change kernel (ablation) | SMMD · image generation | sample quality | No clear effect |
Closest papers by meaning
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.92)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.91)
- Demystifying MMD GANs (ICLR 2018 · similarity 0.91)
- Improved Consistency Regularization for GANs (AAAI 2021 · similarity 0.90)
- MS$^3$D: A RG Flow-Based Regularization for GAN Training with Limited Data (ICML 2024 · similarity 0.90)
- Consistency Regularization for Generative Adversarial Networks (ICLR 2020 · 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.