Research map / Graph learning / GNN architectures
Path Neural Networks: Expressive and Accurate Graph Neural Networks
Gaspard Michel, Giannis Nikolentzos, Johannes F. Lutzeyer, Michalis Vazirgiannis
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 PathNN-SP+ instead of 3WLGNN | graph classification · CSL | accuracy: 97.8 ± 10.9 -> 100.0 ± 0.0 | Improved |
| use PathNN-SP instead of 3WLGNN | graph classification · CSL | accuracy: 97.8 ± 10.9 -> 90.0 ± 0.0 | Hurt |
| aggregate all simple paths instead of single shortest paths (ablation) | PathNN · CSL | accuracy: 90.0 ± 0.0 -> 100.0 ± 0.0 | Improved |
| use PathNN-SP+ instead of GIN | graph classification · NCI1 | accuracy: 80.0 ± 1.4 -> 82.3 ± 1.9 | Improved |
| increase path length from K=1 to K=3 (ablation) | PathNN-SP · NCI1 | accuracy: 77.5 ± 1.6 -> 82.2 ± 1.7 | Improved |
| increase path length from K=1 to K=3 (ablation) | PathNN-SP · DD | accuracy: 76.9 ± 3.7 -> 77.0 ± 3.1 | No clear effect |
| use PathNN-P (K=1) instead of the strongest listed baseline | graph classification · ENZYMES | accuracy: 69.4 ± 6.2 -> 73.0 ± 5.2 | Improved |
| use PathNN-P (K=1) instead of GIN | graph classification · IMDB-M | accuracy: 48.5 ± 3.3 -> 50.8 ± 4.5 | Improved |
| use PathNN-SP instead of GCN | graph classification · Peptides-Functional | Average Precision: 59.30 ± 0.23 -> 68.16 ± 0.26 | Improved |
| use PathNN-SP+ instead of GatedGCN | graph regression · Peptides-Structural | Mean Absolute Error: 0.3420 ± 0.0013 -> 0.2540 ± 0.0046 | Improved |
| use PathNN-SP+ instead of CIN | graph classification · ogbg-molhiv | ROC-AUC: 80.94 ± 0.57 -> 79.17 ± 1.09 | Hurt |
| use PathNN-AP instead of CIN | graph regression · ZINC12K | Mean Absolute Error: 0.079 ± 0.006 -> 0.090 ± 0.004 | Hurt |
Closest papers by meaning
- The Expressive Power of Path-Based Graph Neural Networks (ICML 2024 · similarity 0.90)
- Weisfeiler and Lehman Go Paths: Learning Topological Features via Path Complexes (AAAI 2024 · similarity 0.90)
- Neural Bellman-Ford Networks: A General Graph Neural Network Framework for Link Prediction (NeurIPS 2021 · similarity 0.89)
- Path Integral Based Convolution and Pooling for Graph Neural Networks (NeurIPS 2020 · similarity 0.89)
- Random Walk Graph Neural Networks (NeurIPS 2020 · similarity 0.89)
- Commute Graph Neural Networks (ICML 2025 · 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.
