Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Physical Constraints with Neural Projections
Shuqi Yang, Xingzhe He, Bo Zhu
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 |
|---|---|---|---|
| Use iterative neural projection instead of direct next-step prediction with a simple MLP | Neural physics simulation · Rigid bodies, ropes, articulated bodies, and collisions | Position trajectory error: 10^{-2} -> 10^{-4} and 10^{-3} | Improved |
| Use iterative neural projection instead of an Interaction Net | Neural physics simulation · Rigid bodies, ropes, articulated bodies, and collisions | Position trajectory error: 10^{-2} -> 10^{-4} and 10^{-3} | Improved |
| Use iterative neural projection instead of the comparison models | Neural physics simulation · Rigid bodies, ropes, articulated bodies, and collisions | Constraint error: around 10^{-4}; other models two orders of magnitude higher | Improved |
| Train Interaction Net with Δt = 0.001s instead of Δt = 0.01s | Interaction Net · 200 test samples | Constraint error: 0.026 -> 5e-5 | Improved |
| Use neural projection with Δt = 0.1s instead of Interaction Net with Δt = 0.01s | Neural physics simulation · 200 test samples | Constraint error: 0.026 -> 5.3e-5 | Improved |
| Use neural projection with Δt = 0.1s instead of Interaction Net with Δt = 0.001s | Neural physics simulation · 200 test samples | Constraint error: 5e-5 vs. 5.3e-5 | No clear effect |
Closest papers by meaning
- Scaling physics-informed hard constraints with mixture-of-experts (ICLR 2024 · similarity 0.87)
- A Compositional Object-Based Approach to Learning Physical Dynamics (ICLR 2017 · similarity 0.86)
- Learning Physics Constrained Dynamics Using Autoencoders (NeurIPS 2022 · similarity 0.86)
- Learning Articulated Rigid Body Dynamics with Lagrangian Graph Neural Network (NeurIPS 2022 · similarity 0.86)
- RIGNO: A Graph-based Framework For Robust And Accurate Operator Learning For PDEs On Arbitrary Domains (NeurIPS 2025 · similarity 0.84)
- ConFIG: Towards Conflict-free Training of Physics Informed Neural Networks (ICLR 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.
