Research map / Image generation / Fast diffusion sampling
Learning Energy-Based Prior Model with Diffusion-Amortized MCMC
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 |
|---|---|---|---|
| Progressively amortize short-run LD chain segments instead of directly distilling the long-run chain (NALR) (ablation) | DAMC-LEBM · CIFAR-10 | FID: 64.38 -> 57.72 | Helped |
| Learn LEBM jointly instead of a Gaussian prior (DAMC-G) (ablation) | DAMC · CIFAR-10 | FID: 66.93 -> 57.72 | Helped |
| Use DAMC sampler instead of vanilla sampling (ablation) | DAMC-LEBM · CIFAR-10 | FID: 60.89 -> 57.72 | Helped |
| Use DAMC sampler instead of vanilla sampling (ablation) | NALR-LEBM · CIFAR-10 | FID: 68.52 -> 64.38 | Helped |
| Use DAMC sampler instead of vanilla sampling (ablation) | DAMC-G · CIFAR-10 | FID: 90.30 -> 66.93 | Helped |
| Train LEBM with DAMC samples instead of short-run LD | LEBM · CIFAR-10 | FID: 70.15 -> 60.89 | Helped |
| Train LEBM with DAMC samples instead of short-run LD | LEBM · CelebA-HQ | FID: 133.07 -> 89.54 | Helped |
| DAMC-based learning instead of variational learning (VI-LEBM) (ablation) | LEBM · CIFAR-10 | FID: 78.06 -> 57.72 | Helped |
| Use DAMC sampler instead of LEBM-based inversion with 4x iterations | StyleGAN inversion · FFHQ | FID: 84.64 -> 52.85 | Helped |
| Use DAMC sampler instead of LEBM-based inversion with 4x iterations | StyleGAN inversion · LSUN-Tower | FID: 111.53 -> 80.42 | Helped |
| Increase LEBM inversion optimization iterations from 100 to 400 | LEBM-based inversion · FFHQ | FID: 149.21 -> 84.64 | Helped |
| Replace encoder with DAMC sampler for inversion | Enc. baseline · FFHQ | MSE: 0.028 -> 0.025 | Helped |
Closest papers by meaning
- No MCMC for me: Amortized sampling for fast and stable training of energy-based models (ICLR 2021 · similarity 0.90)
- Learning Latent Space Energy-Based Prior Model (NeurIPS 2020 · similarity 0.90)
- Learning Latent Space Hierarchical EBM Diffusion Models (ICML 2024 · similarity 0.90)
- Learning Energy-Based Models by Cooperative Diffusion Recovery Likelihood (ICLR 2024 · similarity 0.89)
- Outsourced Diffusion Sampling: Efficient Posterior Inference in Latent Spaces of Generative Models (ICML 2025 · similarity 0.86)
- Inverse Problem Sampling in Latent Space Using Sequential Monte Carlo (ICML 2025 · similarity 0.86)
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.