Research map / Image generation / Fast diffusion sampling
Physics-Informed Distillation of Diffusion Models for PDE-Constrained Generation
Yi Zhang, Peng Wang, Difan Zou
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 |
|---|---|---|---|
| add hard-constraint replacement of masked entries at inference (ablation) | PIDDM · Darcy forward | PDE error (x1e-4): 0.354 -> 0.145 | Improved |
| add hard-constraint replacement at inference (ablation) | PIDDM · Darcy forward | MSE (x1e-1): 0.705 -> 0.316 | Improved |
| add two rounds of reflowing (ablation) | PIDDM · Darcy forward | MSE (x1e-1): 0.316 -> 0.127 | Improved |
| add consistency model distillation (ablation) | PIDDM · Darcy forward | PDE error (x1e-4): 0.145 -> 0.083 | Improved |
| add distribution matching distillation (ablation) | PIDDM · Darcy forward | MSE (x1e-1): 0.316 -> 0.255 | Improved |
| use VP diffusion process instead of linear (ablation) | PIDDM · Darcy forward | MSE (x1e-1): 0.316 -> 0.398 | Hurt |
| use sub-VP diffusion process instead of linear (ablation) | PIDDM · Darcy forward | MSE (x1e-1): 0.316 -> 0.372 | Hurt |
| add optional refinement of the latent noise (PIDDM-ref) (ablation) | PIDDM-1 · Darcy generation | PDE error (x1e-4): 0.226 -> 0.148 | Improved |
| add refinement of the latent noise (PIDDM-ref) (ablation) | PIDDM-1 · Darcy generation | FPD: 0.754 -> 0.385 | Improved |
| increase teacher sampling steps N_s (ablation) | PIDDM student · Darcy | MMSE, SMSE, PDE error | Improved |
| vary PDE loss weight in distillation and inference (ablation) | PIDDM · Darcy | all metrics | No clear effect |
| enforce PDE on final sample via distillation instead of posterior mean (PIDM) | PIDDM-1 · Darcy generation | PDE error (x1e-4): 1.236 -> 0.226 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- D-Flow — DFlow: A Generative Model Combining Denoising AutoEncoder and Normalizing Flow for High Fidelity Waveform Generation (ICML 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Ultra Fast PDE Solving via Physics Guided Few-step Diffusion (NeurIPS 2026 · similarity 0.93)
- Physics-Informed Diffusion Models (ICLR 2025 · similarity 0.91)
- DiffusionPDE: Generative PDE-Solving under Partial Observation (NeurIPS 2024 · similarity 0.89)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.89)
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.89)
- Simple Distillation for One-Step Diffusion Models (NeurIPS 2025 · 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.
