Research map / Image generation / Fast diffusion sampling
Strictly Constrained Generative Modeling via Split Augmented Langevin Sampling
Matthieu Blanke, Yongquan Qu, Sara Shamekh, Pierre Gentine
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 |
|---|---|---|---|
| vary rho over iterations (linear 2 to 20) instead of constant rho_min (ablation) | CASAL · Field generation, 1000 steps | wrong-mode proportion: 42.4% -> 0.04% | Improved |
| vary rho linearly instead of constant rho_max (ablation) | CASAL · Field generation, 1000 steps | wrong-mode proportion: 22% -> 0.04% | Improved |
| use linear schedule instead of logarithmic (ablation) | CASAL · Field generation, 1000 steps | wrong-mode proportion: 3.8% -> 0.04% | Improved |
| increase Langevin steps from 1000 to 5000 with linear schedule (ablation) | CASAL · Field generation | wrong-mode proportion: 0.04% -> 0.001% | Improved |
| add CASAL splitting vs projected diffusion | projected diffusion · Darcy flow | reconstruction error: 22% -> 10% | Improved |
| use CASAL instead of penalty guidance | penalty diffusion · Darcy flow | constraint violation: 25% -> 2e-5% | Improved |
| use CASAL instead of projected diffusion | projected diffusion · Navier-Stokes | reconstruction error: 10% -> 8% | Improved |
| use CASAL instead of projected diffusion | projected diffusion · Poisson equation | reconstruction error: 15% -> 5% | Improved |
| use CASAL instead of projected Langevin | projected Langevin · Data assimilation runtime | seconds per 10^6 steps: 2.46 -> 0.31 | Improved |
| use CASAL instead of unconstrained Langevin | unconstrained Langevin · Fluid generation runtime | seconds per 10^6 steps: 0.37 -> 0.45 | Hurt |
| use CASAL instead of projected Langevin | projected Langevin · Energy-constrained field generation | match to conditional distribution | Improved |
| use CASAL instead of penalty guidance | penalty guidance · Quadrotor feasibility | success rate | Improved |
Closest papers by meaning
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.89)
- Efficient Integrators for Diffusion Generative Models (ICLR 2024 · similarity 0.88)
- Wasserstein Convergence of Critically Damped Langevin Diffusions (NeurIPS 2025 · similarity 0.87)
- Constrained Exploration via Reflected Replica Exchange Stochastic Gradient Langevin Dynamics (ICML 2024 · similarity 0.87)
- The Poisson Midpoint Method for Langevin Dynamics: Provably Efficient Discretization for Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Generating Physical Dynamics under Priors (ICLR 2025 · similarity 0.87)
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.
