Research map / Graph learning / Graph self-supervised learning
Rotary Position Encodings for Graphs
Isaac Reid, Arijit Sehanobish, Cederik Höfs, Bruno Kacper Mlodozeniec, Leonhard Vulpius, Federico Barbero, Adrian Weller, Krzysztof Marcin Choromanski, Richard E. Turner, Petar Veličković
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 WIRE rotary encodings with m=5 spectral coordinates | Transformer · Shortest path distance, Watts-Strogatz | test RMSE: 0.065(5) -> 0.038(6) | Improved |
| increase spectral coordinates from m=5 to m=10 (ablation) | Transformer + WIRE · Shortest path distance, Watts-Strogatz | test RMSE: 0.038(6) -> 0.045(4) | No clear effect |
| add WIRE (m=10) to the transformer | Transformer · Monochromatic subgraph, 5 edges deleted | normalised test RMSE: 0.087(1) -> 0.068(5) | Improved |
| add WIRE with only m=3 coordinates when graph is far from grid (ablation) | Transformer · Monochromatic subgraph, 15 edges deleted | normalised test RMSE: 0.068(2) -> 0.064(3) | No clear effect |
| increase m from 3 to 10 on more irregular graphs (ablation) | Transformer + WIRE · Monochromatic subgraph, 5 edges deleted | normalised test RMSE: 0.075(2) -> 0.068(5) | Improved |
| add WIRE to Performer | GraphGPS Performer · MNIST | accuracy: 97.56(2) -> 98.10(1) | Improved |
| add WIRE to Performer | GraphGPS Performer · ogbg-molpcba | test metric: 0.238(3) -> 0.264(1) | Improved |
| add WIRE to Performer | GraphGPS Performer · ogbg-code2 | test metric: 0.1731(9) -> 0.1733(9) | No clear effect |
| add WIRE to Performer | GraphGPS Performer · ogbg-ppa | test metric: 0.8009(8) -> 0.804(2) | No clear effect |
| add WIRE to Performer | GraphGPS Performer · MalNet-Tiny | accuracy: 92.81(5) -> 93.46(2) | Improved |
| add WIRE to Performer | GraphGPS Performer · Peptides-struct | test metric (MAE): 0.2616(4) -> 0.2566(4) | Improved |
| add gauge-invariance handling (SignNet etc.) to WIRE | WIRE · synthetic and benchmark tasks | performance | No clear effect |
Closest papers by meaning
- Round and Round We Go! What makes Rotary Positional Encodings useful? (ICLR 2025 · similarity 0.89)
- PaTH Attention: Position Encoding via Accumulating Householder Transformations (NeurIPS 2025 · similarity 0.88)
- Towards Principled Graph Transformers (NeurIPS 2024 · similarity 0.88)
- On the Stability of Expressive Positional Encodings for Graphs (ICLR 2024 · similarity 0.88)
- Positional Encoding meets Persistent Homology on Graphs (ICML 2025 · similarity 0.87)
- What Improves the Generalization of Graph Transformers? A Theoretical Dive into the Self-attention and Positional Encoding (ICML 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.
