Research map / Learning theory and algorithms / Prediction and decision losses
Are GANs Created Equal? A Large-Scale Study
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 different GAN objectives with extensive hyperparameter search | GAN algorithms in the comparison · MNIST, Fashion-MNIST, CIFAR, CelebA | FID | Mixed |
| Increase the hyperparameter-search computational budget | GAN algorithms in the comparison · GAN benchmarks | FID | Helped |
| Report score distributions rather than only the best seed | GAN evaluation · GAN benchmarks | FID | Helped |
| Transfer narrow hyperparameter ranges tuned on Fashion-MNIST to other datasets (ablation) | GAN algorithms · Other datasets | FID | Mixed |
| Replace Adam with RMSProp (ablation) | WGAN · CIFAR, CelebA | FID: CIFAR10: 154.5 vs 161.2; Celeba: 97.9 vs 216.3 | Mixed |
| Vary Adam's β1 (ablation) | GAN algorithms · Fashion-MNIST | FID | No clear effect |
| Use RMSProp rather than Adam across the broad WGAN search (ablation) | WGAN · CIFAR, CelebA | FID | No clear effect |
| Compute FID using VGG rather than InceptionNet embeddings (ablation) | FID evaluation · CelebA | Spearman rank correlation: ρ=0.9 | No clear effect |
| Drop data modes (ablation) | FID evaluation · Mode-dropping experiment | FID | Hurt |
Closest papers by meaning
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.93)
- Quantitatively Evaluating GANs With Divergences Proposed for Training (ICLR 2018 · similarity 0.93)
- Variational Annealing of GANs: A Langevin Perspective (ICML 2019 · similarity 0.91)
- A Neural Tangent Kernel Perspective of GANs (ICML 2022 · similarity 0.91)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.91)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 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.