Research map / Image generation / GANs
Unrolled Generative Adversarial Networks
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 10 discriminator unrolling steps (ablation) | GAN with a 1/4-size discriminator · 1000-class MNIST | modes generated: 30.6 ± 20.73 -> 327.2 ± 74.67 | Helped |
| Add 10 discriminator unrolling steps (ablation) | GAN with a 1/2-size discriminator · 1000-class MNIST | KL(model || data): 2.58 ±0.751 -> 1.43 ± 0.12 | Helped |
| Add 5 discriminator unrolling steps (ablation) | GAN with a 1/4-layer-size model · Colored MNIST | JS divergence: 0.073 ± 0.0058 -> 0.049 ± 0.0021 | Helped |
| Use 1 discriminator unrolling step rather than 0 (ablation) | GAN on colored MNIST · Colored MNIST, 1/4-layer-size model | JS divergence: 0.073 ± 0.0058 -> 0.142 ± 0.028 | Hurt |
| Increase discriminator unrolling from 0 to 10 steps (ablation) | GAN on colored MNIST · Colored MNIST, 1/1-layer-size model | JS divergence: 0.034 ± 0.0034 -> 0.025 ± 0.00076 | Helped |
| Add 10 discriminator unrolling steps (ablation) | GAN · CIFAR10 | Average reconstruction MSE: 0.0231 ± 0.0024 -> 0.0181 ± 0.0018 | Helped |
| Add 1 discriminator unrolling step (ablation) | GAN · CIFAR10 | Average reconstruction MSE: 0.0231 ± 0.0024 -> 0.0195 ± 0.0021 | Helped |
| Use 1 discriminator unrolling step rather than 0 (ablation) | GAN · CIFAR10 | Percent best rank: 0.63% -> 22.97% | Helped |
| Backpropagate through the unrolled discriminator updates (ablation) | Unrolled GAN with five generator steps per discriminator update · 2D mixture of Gaussians | unrolling steps required for convergence: 10 -> 5 | Helped |
| Use a historical ensemble of discriminators instead of discriminator unrolling (ablation) | GAN · 2D mixture of Gaussians | training stability | No clear effect |
| Increase delays between discriminator snapshots in a historical ensemble (ablation) | GAN with an ensemble of five discriminators · 2D mixture of Gaussians | training stability | Hurt |
| Unroll the discriminator | GAN on a 2D mixture of eight Gaussians · 2D mixture of Gaussians | mode coverage and convergence | Helped |
Closest papers by meaning
- On Stabilizing Generative Adversarial Training With Noise (CVPR 2019 · similarity 0.93)
- Smoothness and Stability in GANs (ICLR 2020 · similarity 0.92)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.91)
- Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · similarity 0.91)
- Towards Principled Methods for Training Generative Adversarial Networks (ICLR 2017 · similarity 0.91)
- Training Generative Adversarial Networks by Solving Ordinary Differential Equations (NeurIPS 2020 · 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.