Research map / Image generation / GANs
Learning disconnected manifolds: a no GAN’s land
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 |
|---|---|---|---|
| keep only the 90% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · MNIST | precision: 91.2 {\scriptstyle \pm 0.3} -> 92.5 {\scriptstyle \pm 0.5} | Helped |
| keep only the 80% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · MNIST | precision: 91.2 {\scriptstyle \pm 0.3} -> 93.3 {\scriptstyle \pm 0.3} | Helped |
| keep only the 90% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · MNIST | recall: 93.7 {\scriptstyle \pm 0.5} -> 92.9 {\scriptstyle \pm 0.3} | Hurt |
| keep only the 80% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · MNIST | recall: 93.7 {\scriptstyle \pm 0.5} -> 91.8 {\scriptstyle \pm 0.4} | Hurt |
| keep only the 90% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · MNIST | FID: 24.3 {\scriptstyle \pm 0.3} -> 26.9 {\scriptstyle \pm 0.5} | Hurt |
| keep only the 80% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · MNIST | FID: 24.3 {\scriptstyle \pm 0.3} -> 33.1 {\scriptstyle \pm 0.3} | Hurt |
| keep only the 90% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · F-MNIST | precision: 86.3 {\scriptstyle \pm 0.4} -> 88.6 {\scriptstyle \pm 0.6} | Helped |
| keep only the 80% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · F-MNIST | precision: 86.3 {\scriptstyle \pm 0.4} -> 89.8 {\scriptstyle \pm 0.4} | Helped |
| keep only the 90% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · F-MNIST | recall: 88.2 {\scriptstyle \pm 0.2} -> 86.6 {\scriptstyle \pm 0.5} | Hurt |
| keep only the 80% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · F-MNIST | recall: 88.2 {\scriptstyle \pm 0.2} -> 84.9 {\scriptstyle \pm 0.5} | Hurt |
| keep only the 90% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · F-MNIST | FID: 259.7 {\scriptstyle \pm 3.5} -> 257.4 {\scriptstyle \pm 3.0} | Helped |
| keep only the 80% of WGAN samples with lowest Jacobian norm (ablation) | WGAN · F-MNIST | FID: 259.7 {\scriptstyle \pm 3.5} -> 396.2 {\scriptstyle \pm 6.4} | Hurt |
Closest papers by meaning
- Disconnected Manifold Learning for Generative Adversarial Networks (NeurIPS 2018 · similarity 0.95)
- Manifold Learning Benefits GANs (CVPR 2022 · similarity 0.88)
- Global Versus Localized Generative Adversarial Nets (CVPR 2018 · similarity 0.87)
- A Geometric Analysis of Deep Generative Image Models and Its Applications (ICLR 2021 · similarity 0.86)
- What Secrets Do Your Manifolds Hold? Understanding the Local Geometry of Generative Models (ICLR 2025 · similarity 0.84)
- Learning Distributions on Manifolds with Free-Form Flows (NeurIPS 2024 · similarity 0.84)
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.