Research map / Image generation / Fast diffusion sampling
Supervised Guidance Training for Infinite-Dimensional Diffusion Models
Elizabeth Louise Baker, Alexander Denker, Jes Frellsen
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 FunDPS Tweedie-approximate guidance with learned SGT guidance (k=1) | FunDPS · Sparse observations | RMSE: 0.456 -> 0.373 | Improved |
| replace FunDPS guidance with learned SGT guidance (k=1) | FunDPS · Sparse observations | energy score: 3.709 -> 3.009 | Improved |
| use SGT guidance with a smaller network instead of a same-size conditional diffusion | Cond. Diff. (SGT size) · Sparse observations | RMSE: 0.406 -> 0.373 | Improved |
| use full-size conditional diffusion instead of SGT | SGT k=1 · Sparse observations | RMSE: 0.373 -> 0.347 | Mixed |
| add smoothness preconditioner C^k, k from 0 to 1 (ablation) | SGT · Sparse observations | RMSE: 0.390 -> 0.373 | Improved |
| add smoothness preconditioner C^k, k from 0 to 1 (ablation) | SGT · Sparse observations | energy score: 3.148 -> 3.009 | Improved |
| add smoothness preconditioner C^k (ablation) | SGT · Heat equation | RMSE: 0.0443 -> 0.0448 | No clear effect |
| use SGT instead of FunDPS | FunDPS · Heat equation | RMSE: 0.0519 -> 0.0443 | Improved |
| use SGT with a randomly initialised base model instead of FunDPS | FunDPS · Imperfect base model, random | energy score: 20.685 -> 6.591 | Improved |
| use SGT with a partly trained base model instead of FunDPS | FunDPS · Imperfect base model, partly trained | energy score: 3.816 -> 3.230 | Improved |
| use SGT with a fully trained base model instead of FunDPS | FunDPS · Imperfect base model, fully trained | energy score: 3.608 -> 3.190 | Improved |
| use SGT instead of conditional diffusion, 4 EFD coefficients | Cond. Diff. · Shape inpainting (4 EFD coefficients) | RMSE: 0.72 -> 0.63 | Improved |
Closest papers by meaning
- Guided Diffusion Sampling on Function Spaces with Applications to PDEs (NeurIPS 2025 · similarity 0.91)
- Learning Diffusion Priors from Observations by Expectation Maximization (NeurIPS 2024 · similarity 0.90)
- Tempered Guided Diffusion (NeurIPS 2026 · similarity 0.89)
- A Mixture-Based Framework for Guiding Diffusion Models (ICML 2025 · similarity 0.89)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.88)
- TFG: Unified Training-Free Guidance for Diffusion Models (NeurIPS 2024 · similarity 0.88)
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.
