Research map / Image generation / GANs
Non-Parametric Priors For 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 |
|---|---|---|---|
| Replace the standard latent prior with the optimized non-parametric prior | DCGAN · CelebA d=100, prior samples | FID: 24.055 -> 17.735 | Helped |
| Replace the standard latent prior with the optimized non-parametric prior | DCGAN · CelebA d=100, mid-point samples | FID: 28.608 -> 19.115 | Helped |
| Replace the standard latent prior with the optimized non-parametric prior | DCGAN · CelebA d=200, prior and mid-point samples | FID: 25.035 -> 14.953; 32.241 -> 19.322 | Helped |
| Replace the standard latent prior with the optimized non-parametric prior | DCGAN · CIFAR10 d=100, prior and mid-point samples | FID: 39.235 -> 34.803; 39.262 -> 37.112 | Helped |
| Replace the standard latent prior with the optimized non-parametric prior | DCGAN · LSUN Bedroom d=100, prior and mid-point samples | FID: 42.998 -> 27.857; 76.412 -> 31.472 | Helped |
| Replace the standard latent prior with the optimized non-parametric prior | DCGAN · LSUN Kitchen d=100, prior and mid-point samples | FID: 39.909 -> 33.194; 51.119 -> 35.074 | Helped |
| Use Cauchy rather than the non-parametric prior | DCGAN · CIFAR10 d=100 | FID: 34.803 -> 180.37; 37.112 -> 180.40 | Hurt |
| Optimize prior-to-mid-point KL divergence rather than use uniform or normal | latent prior distribution · One-dimensional prior and mid-point distributions | KL divergence: 0.3065 -> 0.0075; 0.1544 -> 0.0075 | Helped |
| Increase latent dimension from d=100 to d=200 | DCGAN with non-parametric prior · CelebA prior samples | FID: 17.735 -> 14.953 | Helped |
| Increase latent dimension from d=100 to d=200 | DCGAN with non-parametric prior · CelebA mid-point samples | FID: 19.115 -> 19.322 | No clear effect |
Closest papers by meaning
- A Prior of a Googol Gaussians: a Tensor Ring Induced Prior for Generative Models (NeurIPS 2019 · similarity 0.90)
- Complexity Matters: Rethinking the Latent Space for Generative Modeling (NeurIPS 2023 · similarity 0.90)
- Unveiling the Latent Space Geometry of Push-Forward Generative Models (ICML 2023 · similarity 0.89)
- Improving Generative Adversarial Networks via Adversarial Learning in Latent Space (NeurIPS 2022 · similarity 0.89)
- Variational Annealing of GANs: A Langevin Perspective (ICML 2019 · similarity 0.89)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.88)
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.