Research map / Graph learning / GNN architectures
Message Passing on the Edge: Towards Scalable and Expressive GNNs
Pablo Barcelo, Fabian Jogl, Alexander Kozachinskiy, Matthias Lanzinger, Stefan Neumann, Cristobal Rojas
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 |
|---|---|---|---|
| edge-based message passing over edges and triangles instead of node-based MPNN | EB-GNN · CSL | accuracy: 10% -> 20% | Improved |
| add triangle counts as node features to MPNN (ablation) | GIN · BREC | distinguishable pairs: 0 -> 0 on Basic, Regular, Extension, CFI | No clear effect |
| use EB-GNN instead of NC-GNN | edge-based GNN · BREC Extension | distinguishable pairs: 0 -> 60 | Improved |
| use EB-GNN instead of NC-GNN | edge-based GNN · BREC Regular and CFI | distinguishable pairs: Regular 48 vs 48; CFI 0 vs 0 | No clear effect |
| EB-GNN vs D-MPNN on QMD edge-level tasks | EB-GNN · QMD edge-level (4 tasks) | MAE: Bond Index 6.65 -> 4.42 | Improved |
| EB-GNN vs D-MPNN on QMD graph-level tasks | EB-GNN · QMD graph-level (4 tasks) | MAE: IP 4.29 -> 5.49; Dipole Moment 4.59 -> 4.33 | Mixed |
| EB-GNN vs 5-lGIN on QM9 | EB-GNN · QM9 (12 tasks) | normalized MAE: better on 4 tasks, comparable on 7 | Mixed |
| EB-GNN vs MPNN, GPS, HyMN on large graphs | EB-GNN · MalNet-Tiny | accuracy: 91.10 -> 93.30 | Improved |
| EB-GNN instead of 5-lGNN for runtime | EB-GNN · QM9 | runtime: approximately 4 times faster | Improved |
| EB-GNN on CFI graphs | EB-GNN · BREC CFI | distinguishable pairs: 0 | No clear effect |
Closest papers by meaning
- Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness (ICLR 2024 · similarity 0.90)
- What Expressivity Theory Misses: Message Passing Complexity for GNNs (NeurIPS 2025 · similarity 0.90)
- Towards characterizing the value of edge embeddings in Graph Neural Networks (ICML 2025 · similarity 0.89)
- Towards Dynamic Message Passing on Graphs (NeurIPS 2024 · similarity 0.89)
- Recurrent Graph Neural Networks and Their Connections to Bisimulation and Logic (AAAI 2024 · similarity 0.89)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · 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.
