Research map / Graph learning / GNN architectures
Fast and Distributed Equivariant Graph Neural Networks by Virtual Node Learning
Yuelin Zhang, Jiacheng Cen, Jiaqi Han, Wenbing Huang
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 virtual nodes (FastEGNN-<3, 0.00>) | EGNN · Water-3D | MSE (x1e-4): 6.00 -> 2.58 | Improved |
| add virtual nodes (FastEGNN-<3, 0.00>) | EGNN · N-body | MSE (x1e-2): 1.41 -> 1.06 | Improved |
| add MMD loss on virtual nodes (ablation) | FastEGNN · Protein Dynamics | MSE: 1.857 -> 1.821 | Improved |
| model virtual nodes as ordered set instead of global nodes (ablation) | FastEGNN · Protein Dynamics | MSE: 1.950 -> 1.821 | Improved |
| increase virtual nodes from 3 to 10 (ablation) | FastEGNN · Water-3D | MSE (x1e-4): 2.58 -> 2.66 | Hurt |
| drop all edges with 3 virtual nodes (ablation) | FastEGNN · Water-3D | MSE (x1e-4): 2.58 -> 3.60 | Mixed |
| add edge dropping 0.75 with 10 virtual nodes (ablation) | FastEGNN · N-body | Relative inference time: 1.36 -> 0.53 | Improved |
| add virtual nodes to TFN | TFN · N-body (no edge dropping) | MSE (x1e-2): 1.62 -> 3.42 | Hurt |
| add virtual nodes to SchNet | SchNet · Protein Dynamics | MSE: 2.56 -> 1.98 | Improved |
| distribute with virtual nodes (DistEGNN-3) over 8 devices instead of EGNN | EGNN · Water-3D, 8 devices | MSE (x1e-4): 11.05 -> 3.27 | Improved |
| use METIS partitioning instead of random (ablation) | DistEGNN-3 · Water-3D | MSE (x1e-4): 3.46 -> 3.04 | No clear effect |
| dynamically enlarge cutoff radius (ablation) | DistEGNN-1 · Water-3D, 8 devices | MSE (x1e-4): 3.85 -> 3.27 | Improved |
Closest papers by meaning
- Improving Equivariant Graph Neural Networks on Large Geometric Graphs via Virtual Nodes Learning (ICML 2024 · similarity 0.97)
- Universally Invariant Learning in Equivariant GNNs (NeurIPS 2025 · similarity 0.90)
- Are High-Degree Representations Really Unnecessary in Equivariant Graph Neural Networks? (NeurIPS 2024 · similarity 0.90)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · similarity 0.90)
- On Explaining Equivariant Graph Networks via Improved Relevance Propagation (ICML 2025 · similarity 0.89)
- On the Generalization of Equivariant Graph Neural Networks (ICML 2024 · similarity 0.89)
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.
