Research map / Image generation / Fast diffusion sampling
Sample-efficient evidence estimation of score based priors for model selection
Frederic Wang, Katherine Bouman
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 |
|---|---|---|---|
| use improved prior-aware posterior covariance instead of the sigma_t^2 heuristic in Gaussian-approximation DAPS (ablation) | DiME · Gaussian mixture, in-distribution x* | relative error (%): 146% -> 1.5% | Improved |
| use improved covariance instead of original heuristic (ablation) | DiME · Gaussian mixture, out-of-distribution x* | relative error (%): 3.3% -> 0.3% | Improved |
| use PnP-DM marginals (DiME-tilted) instead of standard DAPS marginals (ablation) | DiME · Gaussian mixture, out-of-distribution x* | relative error (%): 0.3% -> 13.8% | Hurt |
| use PnP-DM marginals instead of standard marginals (ablation) | DiME · Gaussian mixture, in-distribution x* | relative error (%): 1.5% -> 2.4% | No clear effect |
| estimate evidence along posterior time-marginals (DiME) instead of power-posterior SMC | evidence estimator · Gaussian mixture, saddle point x* | relative error (%): 0.7% -> 0.8% | No clear effect |
| use DiME instead of naive Monte Carlo | evidence estimation · Gaussian mixture, in-distribution x* | relative error (%): 2263% -> 1.5% | Improved |
| use DiME instead of SMC with learned score as surrogate prior score | MNIST diffusion priors, Gaussian and Fourier phase retrieval model selection · MNIST digit prior selection | correct model selected | Improved |
| use Gaussian-approximation DAPS instead of exact DAPS (ablation) | DiME on M87* · M87* model selection | compute time: 30 min -> 4 min (about 7x) | Improved |
Closest papers by meaning
- Principled Probabilistic Imaging using Diffusion Models as Plug-and-Play Priors (NeurIPS 2024 · similarity 0.91)
- Divide-and-Conquer Posterior Sampling for Denoising Diffusion priors (NeurIPS 2024 · similarity 0.90)
- Rethinking Diffusion Posterior Sampling: From Conditional Score Estimator to Maximizing a Posterior (ICLR 2025 · similarity 0.90)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.90)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.89)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.89)
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.
