Research map / Graph learning / GNN architectures
Enhancing the Inductive Biases of Graph Neural ODE for Modeling Physical Systems
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 explicit constraints (CGnode: predict nodal force, compute acceleration via constrained Lagrangian dynamics) (ablation) | Gnode on pendulum · n-pendulum (5-link trained, 4 and 5 link tested) | energy violation and rollout error: ~4 orders of magnitude | Helped |
| decouple internal (local, topology-based) and body (global position/velocity) forces (ablation) | CGnode (CDGnode) · pendulum and spring systems | energy violation and rollout error: ~1 order of magnitude | Helped |
| enforce Newton's third law (edge-based symmetric forces, momentum conserving) (ablation) | CDGnode (MCGnode) · spring system | momentum error: zero momentum error | Helped |
| enforce Newton's third law (MCGnode) (ablation) | CDGnode on pendulum · pendulum system | momentum error | No clear effect |
| MCGnode vs Gnode-style baseline on large unseen systems | MCGnode vs LGN/HGN · generalization to 20/50-link spring and 10/20-link pendulum | energy and rollout error | Helped |
| full graph architecture for LGN/HGN vs the Gnode GNN backbone (ablation) | LGN and HGN · n-pendulum and spring | performance | Helped |
| removing velocity as input (MCGnode-woV) (ablation) | MCGnode · 5-link pendulum | energy violation and rollout error | Hurt |
| use relative velocity instead of relative position as edge feature (ablation) | Gnode · pendulum training | training loss convergence | Hurt |
| MCGnode vs Node trained per system (training efficiency) | MCGnode · 5-spring and 5-pendulum, dataset sizes 100 to 10000 | energy and rollout error: orders of magnitude | Helped |
| Gnode without additional biases vs Node trained per system (ablation) | Gnode · 3,4,5 pendulum and spring | energy and rollout error | No clear effect |
Closest papers by meaning
- SEGNO: Generalizing Equivariant Graph Neural Networks with Physical Inductive Biases (ICLR 2024 · similarity 0.90)
- Towards Cross Domain Generalization of Hamiltonian Representation via Meta Learning (ICLR 2024 · similarity 0.86)
- ControlSynth Neural ODEs: Modeling Dynamical Systems with Guaranteed Convergence (NeurIPS 2024 · similarity 0.86)
- Physics-Informed DeepONets for drift-diffusion on metric graphs: simulation and parameter identification (ICML 2025 · similarity 0.86)
- Improving equilibrium propagation without weight symmetry through Jacobian homeostasis (ICLR 2024 · similarity 0.86)
- Stable Port-Hamiltonian Neural Networks (NeurIPS 2025 · similarity 0.86)
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.