Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Statistical Benchmark for Diffusion Posterior Sampling Algorithms
Martin Zach, Youssef Haouchat, Michael Unser
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 learned denoiser with oracle Gibbs denoiser (ablation) | DPnP · Imputation, Student-t(1) jumps | MMSE optimality gap (dB): -10.46 ± 5.70 | Improved |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | DPnP · Deconvolution, Student-t(1) jumps | MMSE optimality gap (dB): -4.98 ± 3.86 | Improved |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | DiffPIR · Deconvolution, Student-t(1) jumps | MMSE optimality gap (dB): 13.56 ± 9.90 | Hurt |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | C-DPS · Deconvolution, Student-t(1) jumps | MMSE optimality gap (dB): 6.64 ± 3.21 | Hurt |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | C-DPS · Imputation, Student-t(1) jumps | MMSE optimality gap (dB): 3.41 ± 4.99 | Hurt |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | DPnP · Denoising, Gaussian jumps | MMSE optimality gap (dB): 0.04 ± 0.27 | No clear effect |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | DiffPIR · Denoising, Gaussian jumps | MMSE optimality gap (dB): 0.00 ± 0.13 | No clear effect |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | DPnP · Fourier, Student-t(1) jumps | MMSE optimality gap (dB): -1.32 ± 3.18 | Improved |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | DPnP · Denoising, Bernoulli-Laplace(0.1, 1) jumps | MMSE optimality gap (dB): -0.55 ± 1.31 | Improved |
| replace learned denoiser with oracle Gibbs denoiser (ablation) | C-DPS · Imputation, Bernoulli-Laplace(0.1, 1) jumps | MMSE optimality gap (dB): 0.41 ± 1.51 | Hurt |
| add brief hand-tuning of DiffPIR parameters (zeta=0.7, rho=4) on the validation set (ablation) | DiffPIR with oracle denoiser · Heavy-tailed jumps | MMSE optimality gap (dB): almost 10 dB lower | Improved |
| use DiffPIR instead of C-DPS | DPS algorithms · Denoising, Bernoulli-Laplace(0.1, 1) jumps | MMSE optimality gap (dB): 2.22 -> 0.72 | Improved |
Closest papers by meaning
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.93)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.92)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.91)
- Split Gibbs Discrete Diffusion Posterior Sampling (NeurIPS 2025 · similarity 0.91)
- A Mixture-Based Framework for Guiding Diffusion Models (ICML 2025 · similarity 0.91)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 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.
