Research map / Image generation / GANs
A Prior of a Googol Gaussians: a Tensor Ring Induced Prior for Generative Models
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 Gaussian prior with a learnable TRIP prior | WGAN-GP · CelebA | FID: 54.71 -> 52.86 | Helped |
| Replace the standard Gaussian prior with a learnable TRIP prior | VAE · CelebA | FID: 86.72 -> 83.54 | Helped |
| Replace the standard Gaussian prior with a learnable TRIP prior | VAE · CelebA | ELBO: -194.16 -> -193.32 | Helped |
| Replace the standard Gaussian prior with a learnable TRIP prior | WGAN-GP · CIFAR-10 | FID: 29.3 -> 16.72 | Helped |
| Use a joint TRIP prior to train with partially missing attributes | VAE conditional generation · CelebA, 90% missing attributes | condition matching accuracy: 74.87 -> 87.08 | Helped |
| Use a joint TRIP prior to train with partially missing attributes | VAE conditional generation · CelebA, 99% missing attributes | condition matching accuracy: 73.68 -> 84.89 | Helped |
| Increase TRIP core size from 1 to 20 (ablation) | VAE-TRIP · CIFAR-10 | ELBO: -89.5 -> -89.1 | Helped |
| Increase TRIP core size from 1 to 20 (ablation) | VAE-TRIP · CelebA | ELBO: -243.40 -> -215.62 | Helped |
| Increase TRIP core size from 20 to 100 (ablation) | TRIP prior · Per-batch sampling on Tesla K80 | sampling time: 312 ± 18 -> 882 ± 15 | Hurt |
| Replace the standard Gaussian prior with a fixed TRIP prior | WGAN-GP · CelebA | FID: 54.71 -> 63.06 | Hurt |
| Replace the standard Gaussian prior with a learnable GMM prior | WGAN-GP · CelebA | FID: 54.71 -> 62.10 | Hurt |
| Combine a TRIP prior with normalizing flows | VAE · CelebA | ELBO: -189.1 -> -184.7 | Helped |
Closest papers by meaning
- Non-Parametric Priors For Generative Adversarial Networks (ICML 2019 · similarity 0.90)
- Variational Annealing of GANs: A Langevin Perspective (ICML 2019 · similarity 0.88)
- On Unifying Deep Generative Models (ICLR 2018 · similarity 0.88)
- Projected GANs Converge Faster (NeurIPS 2021 · similarity 0.87)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.87)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.87)
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.