Research map / Image generation / Fast diffusion sampling
Aligning Optimization Trajectories with Diffusion Models for Constrained Design Generation
Giorgio Giannone, Akash Srivastava, Ole Winther, Faez Ahmed
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 trajectory alignment (clean-space) to DOM (ablation) | DOM with kernel conditioning · in-distribution topology generation | average compliance error (%): 13.61 -> 4.44 | Improved |
| add trajectory alignment (clean-space) to DOM (ablation) | DOM with kernel conditioning · ablation, out-of-distribution | median compliance error (%): 5.62 -> 3.47 | Improved |
| add dense kernel relaxation (ablation) | DOM with clean TA · ablation, out-of-distribution | median compliance error (%): 9.01 -> 3.47 | Improved |
| add dense kernel relaxation (ablation) | DOM with clean TA · ablation, in-distribution | median compliance error (%): 1.11 -> 0.74 | Improved |
| match trajectories in clean space instead of noisy, performance, or multi-resolution space (ablation) | DOM with kernel and TA · in-distribution | median compliance error (%): 0.74 vs 2.08 (NOISY), 2.41 (PERF), 34.95 (MULTI) | Improved |
| replace DOM with Consistency Model objective for kernel conditioned model (ablation) | kernel-conditioned diffusion · in-distribution | median compliance error (%): 0.74 -> 2.20 | Hurt |
| use DOM instead of TopoDiff-G with only 2 sampling steps | few-step sampling · in-distribution | median compliance error (%): 436.83 -> 1.46 | Improved |
| use DOM instead of TopoDiff-G with 10 sampling steps | few-step sampling · out-of-distribution | median compliance error (%): 2.55 -> 3.69 | Hurt |
| add 10 steps of SIMP post-optimization to DOM (ablation) | DOM · out-of-distribution | median compliance error (%): 3.47 -> 1.15 | Improved |
| add only 5 steps of SIMP post-optimization (ablation) | DOM · out-of-distribution | floating material (%): 8.02 -> 10.19 | Hurt |
| remove FEM preprocessing via kernel relaxation (DOM vs TopoDiff) | inference pipeline · 64x64 inference | inference time (s): 5.54 -> 2.35 | Improved |
| DOM versus TopoDiff on floating material with 100 steps | DOM · in-distribution | floating material (%): 5.79 -> 6.72 | Hurt |
Closest papers by meaning
- Optimize Any Topology: A Foundation Model for Shape- and Resolution-Free Structural Topology Optimization (NeurIPS 2025 · similarity 0.89)
- Guided Trajectory Generation with Diffusion Models for Offline Model-based Optimization (NeurIPS 2024 · similarity 0.89)
- Training-Free Constrained Generation With Stable Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Physics-Informed Diffusion Models (ICLR 2025 · similarity 0.88)
- Constrained Synthesis with Projected Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Model-based Diffusion for Trajectory Optimization (NeurIPS 2024 · 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.
