Research map / Probabilistic and scientific modeling / Dynamical systems
Neural Modular Physics for Elastic Simulation
Yifei Li, Haixu Wu, Zeyi Xu, Tuur Stuyck, Wojciech Matusik
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 volume preservation loss (weight 1) (ablation) | NMP · Cube real-world scenario | 1000-step RMSE: 0.397 -> 0.337 | Improved |
| increase volume loss weight from 1 to 1000 (ablation) | NMP · Cube real-world scenario | 1000-step RMSE: 0.337 -> 2.579 | Hurt |
| increase volume loss weight from 1 to 10 (ablation) | NMP · Cube real-world scenario | 1000-step RMSE: 0.337 -> 0.604 | Hurt |
| add local volume preservation loss (ablation) | NMP · Cube | tetrahedra exceeding 5% volume change: reduces maximum violations by 10% | Improved |
| use larger constitutive network (96/80 vs 64/40 hidden) (ablation) | NMP neural constitutive model · Cube | 1000-step RMSE: 0.337 -> 0.220 | Improved |
| train both modules jointly instead of two-stage (ablation) | NMP · Cube | internal force quality | Hurt |
| predict 100 steps at once instead of one-step | Transolver · Bob | 1000-step RMSE: 1.782 -> 0.75 | Improved |
| predict 100 steps at once instead of one-step | MeshGraphNet · Bob | 1000-step RMSE: 2.104 -> 2.61 | Hurt |
| use NMP instead of MeshGraphNet | simulator · Cube | 1000-step RMSE: 1.554 -> 0.535 | Improved |
| use NMP instead of MeshGraphNet | simulator · Bob | 100-step RMSE: 0.048 -> 0.050 | Hurt |
| use NMP instead of physics simulator | simulator · Bob | inference time (ms): 90275 -> 2250 | 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.
- Transolver — Transolver: A Fast Transformer Solver for PDEs on General Geometries (ICML 2024) · better on 9 of 9 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
- Diffusion-Based Hierarchical Graph Neural Networks for Simulating Nonlinear Solid Mechanics (NeurIPS 2025 · similarity 0.85)
- Scaling physics-informed hard constraints with mixture-of-experts (ICLR 2024 · similarity 0.84)
- Parameterized Physics-informed Neural Networks for Parameterized PDEs (ICML 2024 · similarity 0.84)
- UniPhy: Learning a Unified Constitutive Model for Inverse Physics Simulation (CVPR 2025 · similarity 0.84)
- Interpolating Neural Network-Tensor Decomposition (INN-TD): a scalable and interpretable approach for large-scale physics-based problems (ICML 2025 · similarity 0.84)
- Harnessing the Power of Neural Operators with Automatically Encoded Conservation Laws (ICML 2024 · 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.
