Research map / Probabilistic and scientific modeling / Molecules and proteins
Inverse Design for Fluid-Structure Interactions using Graph Network Simulators
Kelsey R Allen, Tatiana Lopez-Guavara, Kim Stachenfeld, Alvaro Sanchez-Gonzalez, Peter Battaglia, Jessica B Hamrick, Tobias Pfaff
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 |
|---|---|---|---|
| use gradient descent on learned simulator instead of CEM on learned simulator | GD-M · 2D Fluid Tools | mean reward: 16.1–118.9% | Improved |
| use gradient descent on learned simulator instead of CEM on ground-truth simulator | GD-M · 2D Fluid Tools | mean reward: 3.9–37.5% | Improved |
| use GD-M instead of CEM-M | GD-M · 3D Watercourse | mean reward: 30–85× | Improved |
| increase number of tool joints to 48 | GD-M vs CEM · Contain | mean reward: 154.9% | Improved |
| increase number of tools to 36 joints | GD-M vs CEM · Maze | mean reward: 207.4% | Improved |
| lengthen rollout up to 225 steps | GD-M · 2D Fluid Tools | mean reward: 225 | Mixed |
| remove noise from simulator training (ablation) | GD-M simulator · 2D Fluid Tools | design quality | Hurt |
| encode absolute instead of relative particle coordinates (ablation) | GD-M simulator · 2D Fluid Tools | design quality | Hurt |
| increase ensemble size of learned simulators | GD-M · AIRFOIL | drag coefficient: 0.01919 -> 0.01898 | Improved |
| use size-5 ensemble versus DAFoam | GD-M · AIRFOIL | drag coefficient: 0.01902 | No clear effect |
| use end-to-end surrogate reward model instead of GNN simulator | GD-E2E · 3D Watercourse Direction | actual reward: 1.21 ± 0.18 | Hurt |
| increase design dimensionality (more joints) | CEM-M and CEM-S · Contain | design quality | Hurt |
Closest papers by meaning
- 3DID: Direct 3D Inverse Design for Aerodynamics with Physics-Aware Optimization (NeurIPS 2025 · similarity 0.90)
- AFBench: A Large-scale Benchmark for Airfoil Design (NeurIPS 2024 · similarity 0.88)
- EngiBench: A Framework for Data-Driven Engineering Design Research (NeurIPS 2025 · similarity 0.88)
- Compositional Generative Inverse Design (ICLR 2024 · similarity 0.88)
- Learning Distributions of Complex Fluid Simulations with Diffusion Graph Networks (ICLR 2025 · similarity 0.85)
- PIORF: Physics-Informed Ollivier-Ricci Flow for Long–Range Interactions in Mesh Graph Neural Networks (ICLR 2025 · similarity 0.85)
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.
