Research map / Graph learning / GNN architectures
Orthogonal Bases for Equivariant Graph Learning with Provable k-WL Expressive Power
Jia He, Maggie Cheng
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 order-3 GNN-a instead of PPGN | Order-3 GNN-a · TUDatasets training time (PROTEINS) | training time per epoch (s): 865 -> 223 | Improved |
| use order-3 GNN-a instead of PPGN | Order-3 GNN-a · MUTAG parameters in linear layers | parameters: 2409600 -> 196500 | Improved |
| use order-2 GNN-b instead of PPGN | Order-2 GNN-b · QM9 | training time per epoch (s): 3300 -> 70 | Improved |
| use order-3 GNN-a instead of PPGN | Order-3 GNN-a · QM9 | training time per epoch (s): 3300 -> 178 | Improved |
| use order-3 GNN-a instead of order-2 GNN-b | Order-3 GNN-a · TUDatasets NCI1 | accuracy (%): 80 -> 85 | Improved |
| use order-3 GNN-a instead of order-2 GNN-b | Order-3 GNN-a · TUDatasets PROTEINS | accuracy (%): 73.9 -> 77.8 | Improved |
| use order-3 GNN-a instead of order-2 GNN-b | Order-3 GNN-a · TUDatasets PTC | accuracy (%): 72.4 -> 72.1 | No clear effect |
| use order-3 GNN-a instead of order-2 GNN-b | Order-3 GNN-a · ZINC12k | MAE: 0.146 -> 0.110 | Improved |
| use order-3 GNN-a instead of order-2 GNN-b | Order-3 GNN-a · QM9 (Δϵ) | MAE: 0.0025 -> 0.0033 | Hurt |
| use order-2 GNN-b instead of PPGN | Order-2 GNN-b · ZINC12k | MAE: 0.256 -> 0.146 | Improved |
| use order-2 GNN-b (linear layers) instead of 2-WL | Order-2 GNN-b · graph isomorphism tests 1-2 | succeeded: N -> Y | Improved |
| use order-2 GNN-a instead of order-2 GNN-b (ablation) | Order-2 GNN-a · graph isomorphism tests 1-2 | succeeded: Y -> N | Hurt |
Closest papers by meaning
- Invariant and Equivariant Graph Networks (ICLR 2019 · similarity 0.93)
- Are High-Degree Representations Really Unnecessary in Equivariant Graph Neural Networks? (NeurIPS 2024 · similarity 0.89)
- On the Generalization of Equivariant Graph Neural Networks (ICML 2024 · similarity 0.89)
- Universally Invariant Learning in Equivariant GNNs (NeurIPS 2025 · similarity 0.88)
- Simple Orthogonal Graph Representation Learning (Student Abstract) (AAAI 2024 · similarity 0.88)
- Schur Nets: exploiting local structure for equivariance in higher order graph neural networks (NeurIPS 2024 · similarity 0.88)
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.
