Research map / Graph learning / GNN architectures
Reducing SO(3) Convolutions to SO(2) for Efficient Equivariant GNNs
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 SO(2)-reduced equivariant convolution in eSCN instead of the SCN message-passing model | OC20 All · OC20 Test, S2EF | Force MAE [meV/Å]: 17.7 -> 17.1 | Helped |
| Use the SO(2)-reduced equivariant convolution in eSCN instead of SCN | OC20 All + MD · OC20 Test, S2EF | Force MAE [meV/Å]: 17.2 -> 15.6 | Helped |
| Use eSCN instead of GemNet-OC | OC22 · OC22 Test, S2EF ID | Force MAE [eV/Å]: 29.4 -> 23.8 | Helped |
| Use eSCN instead of GemNet-OC | OC22 · OC22 Test, S2EF OOD | Force MAE [eV/Å]: 39.6 -> 35.7 | Helped |
| Reduce degree L from 6 to 2 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 26.7 | Hurt |
| Reduce degree L from 6 to 4 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 22.2 | Hurt |
| Increase degree L from 6 to 8 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 21.3 | No clear effect |
| Reduce maximum order M from 2 to 0 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 26.5 | Hurt |
| Reduce maximum order M from 2 to 1 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 23.4 | Hurt |
| Increase maximum order M from 2 to 3 or 4 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 21.2 | No clear effect |
| Reduce layers K from 12 to 4 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 25.6 | Hurt |
| Reduce layers K from 12 to 8 (ablation) | eSCN · OC20 2M Validation | Force MAE [meV/Å]: 21.3 -> 22.4 | Hurt |
Closest papers by meaning
- Are High-Degree Representations Really Unnecessary in Equivariant Graph Neural Networks? (NeurIPS 2024 · similarity 0.90)
- A Characterization Theorem for Equivariant Networks with Point-wise Activations (ICLR 2024 · similarity 0.90)
- On Explaining Equivariant Graph Networks via Improved Relevance Propagation (ICML 2025 · similarity 0.89)
- Universally Invariant Learning in Equivariant GNNs (NeurIPS 2025 · similarity 0.89)
- On the Generalization of Equivariant Graph Neural Networks (ICML 2024 · similarity 0.88)
- Fast, Expressive $\mathrm{SE}(n)$ Equivariant Networks through Weight-Sharing in Position-Orientation Space (ICLR 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.