Research map / Image generation / Fast diffusion sampling
Learning Normalized Energy Models for Linear Inverse Problems
Nicolas Zilberstein, Santiago Segarra, Eero P Simoncelli, Florentin Guth
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 ULA corrector with MALA (Metropolis-Hastings) corrector, enabled by normalized density (ablation) | A-CSM energy model · CelebA inpainting | LPIPS: 0.093 -> 0.089 | Improved |
| more ULA corrector steps (ablation) | A-CSM energy model · CelebA inpainting | LPIPS: 0.093 -> 0.093 | No clear effect |
| add covariance score matching (dual score matching) to training (ablation) | A-CSM energy model · blind box-size estimation | covariance estimation | Improved |
| energy-guided adaptive covariance schedule instead of geometric schedule | A-CSM energy model · MNIST reconstruction from k pixels | classification error | Improved |
| energy-guided adaptive covariance schedule instead of predefined schedules | A-CSM energy model · CelebA inpainting | reconstruction quality | Hurt |
| single anisotropic energy model vs retrained or likelihood-approximating baselines | A-CSM energy model · CelebA/ImageNet64 inpainting and deblurring | FID: 36.76 -> 34.57 (CelebA inpainting vs DPS) | Improved |
| apply to super-resolution x4 | A-CSM energy model · CelebA super-resolution x4 | FID: 64.75 -> 93.72 (vs RED-Diff) | Hurt |
| out-of-distribution degradations without retraining | A-CSM energy model · CelebA random inpainting, vertical half mask, motion deblurring | PSNR/FID | Mixed |
Closest papers by meaning
- Improving Diffusion Models for Inverse Problems Using Optimal Posterior Covariance (ICML 2024 · similarity 0.91)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.91)
- Exact Posterior Score Estimation for Solving Linear Inverse Problems (NeurIPS 2026 · similarity 0.91)
- Improving Diffusion Posterior Samplers with Lagged Temporal Corrections for Image Restoration (NeurIPS 2026 · similarity 0.90)
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.90)
- Unleashing the Denoising Capability of Diffusion Prior for Solving Inverse Problems (NeurIPS 2024 · similarity 0.90)
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.
