Research map / Graph learning / Knowledge and temporal graphs
Equivariant Spatio-Temporal Attentive Graph Networks to Simulate Physical Dynamics
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 |
|---|---|---|---|
| Use the full ESTAG spatio-temporal equivariant architecture instead of the strongest listed baseline | ESTAG · MD17 (all eight molecules) | MSE: Aspirin: 0.669 -> 0.063; Benzene: 0.106 -> 0.003; Ethanol: 0.245 -> 0.099; Malonaldehyde: 0.427 -> 0.101; Naphthalene: 0 | Helped |
| Use full ESTAG instead of ST_GMN | ESTAG · AdK protein dynamics | MSE: 1.526 -> 1.471 | Helped |
| Use full ESTAG instead of AGL-STAN | ESTAG · CMU Motion Capture Walk | MSE: 0.037 -> 0.040 | Hurt |
| Use full ESTAG instead of EqMotion | ESTAG · CMU Motion Capture Basketball | MSE: 4.893 -> 0.746 | Helped |
| Add EDFT-derived trajectory features (ablation) | ESTAG · MD17, AdK protein dynamics, and CMU Motion Capture | MSE: MD17 Aspirin: 0.079 -> 0.063; protein: 1.490 -> 1.471; motion: 0.053 -> 0.040 | Helped |
| Add ETM attention (ablation) | ESTAG · MD17, AdK protein dynamics, and CMU Motion Capture | MSE: MD17 Aspirin: 0.087 -> 0.063; protein: 1.493 -> 1.471; motion: 0.047 -> 0.040 | Helped |
| Add Euclidean equivariance (ablation) | ESTAG · MD17, AdK protein dynamics, and CMU Motion Capture | MSE: MD17 Aspirin: 0.762 -> 0.063; protein: 2.011 -> 1.471; motion: 0.702 -> 0.040 | Helped |
| Use equivariant temporal pooling instead of simple softmax-weighted past-frame aggregation (ablation) | ESTAG · MD17, AdK protein dynamics, and CMU Motion Capture | MSE: MD17 Aspirin: 0.084 -> 0.063; protein: 1.472 -> 1.471; motion: 0.044 -> 0.040 | Mixed |
| Use learnable frequency filter weights w_k (ablation) | EDFT in ESTAG · MD17 (eight molecules) | MSE: Aspirin: 0.071 -> 0.063; Benzene: 0.003 -> 0.003; Ethanol: 0.102 -> 0.099; Malonaldehyde: 0.104 -> 0.101; Naphthalene: 0 | Helped |
| Set the ESTAG layer count L to 2 rather than 1 (ablation) | ESTAG · Ethanol | MSE: 1.25 -> 0.990 | Helped |
| Increase the number of past timestamps from 3 to 10 (ablation) | ESTAG · Ethanol | MSE: 1.760 -> 0.990 | Helped |
Closest papers by meaning
- Equivariant Graph Neural Operator for Modeling 3D Dynamics (ICML 2024 · similarity 0.91)
- SEGNO: Generalizing Equivariant Graph Neural Networks with Physical Inductive Biases (ICLR 2024 · similarity 0.90)
- Universally Invariant Learning in Equivariant GNNs (NeurIPS 2025 · similarity 0.87)
- E(n) Equivariant Topological Neural Networks (ICLR 2025 · similarity 0.87)
- PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems (ICLR 2025 · similarity 0.87)
- Are High-Degree Representations Really Unnecessary in Equivariant Graph Neural Networks? (NeurIPS 2024 · similarity 0.87)
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.