Research map / Probabilistic and scientific modeling / Molecules and proteins
On the Generalization in Topology Optimization via Sensitivity-Conditioned Bernoulli Flow Matching
Mohammad Rashed, Duarte F. Valoroso Madeira, Babak Gholami, Caglar Guerbuez, Yunjia Yang, Nils Thuerey
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 |
|---|---|---|---|
| condition on pseudo-sensitivity (velocity magnitude) instead of pressure (ablation) | Bernoulli flow model, CFD-TO · CFD-TO OOD-Hard (3 outlets) | median pressure-drop error (%): 100.0 -> 2.7 | Improved |
| condition on pressure instead of sensitivity or pseudo-sensitivity (ablation) | CFD-TO surrogate · CFD-TO out-of-distribution | median pressure drop error (%): 3.34 -> 11.51 ID to OOD | Hurt |
| condition on strain energy density instead of true sensitivity (ablation) | Bernoulli flow model, structural TO · Structural TO OOD | median compliance error (%): 0.54 -> 0.53 | No clear effect |
| condition on displacement instead of sensitivity (ablation) | Bernoulli flow model, structural TO · Structural TO OOD | median compliance error (%): 0.54 -> 1.30 | Hurt |
| condition on physical parameters only instead of sensitivity (ablation) | Bernoulli flow model, structural TO · Structural TO OOD | median compliance error (%): 0.54 -> 14.0 | Hurt |
| use a deterministic model instead of a generative one, with sensitivity conditioning (ablation) | structural TO surrogate · Structural TO OOD | median compliance error (%): 0.54 -> 2.27 | Hurt |
| add sensitivity conditioning to a deterministic model versus physical parameters (ablation) | deterministic surrogate · Structural TO OOD | median compliance error (%): 14.0 -> 2.27 | Improved |
| use Bernoulli flow matching instead of DiT/UDiT/PDE-T backbones | sensitivity-conditioned surrogate · CFD-TO ID test | relative pressure drop error (%): 2.37 -> 3.68 (UDiT vs Ours) | Hurt |
| use Bernoulli flow matching instead of DiT/UDiT/PDE-T backbones | sensitivity-conditioned surrogate · CFD-TO OOD-Hard (3 outlets) | accuracy within 10% error (%): 68.6 -> 74.5 vs PDE-T | Improved |
| replace sensitivity-conditioned models with one STAR-CCM+ optimization iteration | CFD-TO baseline · CFD-TO OOD-Hard (3 outlets) | relative pressure drop error (%): 6.12 -> 26.51 | Hurt |
| use Bernoulli flow model rather than TopoDiffusion for structural OOD | structural TO · Structural TO OOD testing | mean compliance error (%): 8.57 -> 5.73 | Improved |
| use pseudo-sensitivity instead of exact adjoint sensitivity | CFD-TO conditioning · CFD-TO compute cost | physics time per sample: ~8.5 s -> ~3.2 s | Improved |
Closest papers by meaning
- Optimize Any Topology: A Foundation Model for Shape- and Resolution-Free Structural Topology Optimization (NeurIPS 2025 · similarity 0.87)
- Physics-Informed Diffusion Models (ICLR 2025 · similarity 0.85)
- Diffeomorphic interpolation for efficient persistence-based topological optimization (NeurIPS 2024 · similarity 0.85)
- Boosting Offline Optimizers with Surrogate Sensitivity (ICML 2024 · similarity 0.84)
- Sharper Convergence Rates for Nonconvex Optimisation via Reduction Mappings (NeurIPS 2025 · similarity 0.84)
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.84)
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.
