Research map / Graph learning / GNN architectures
Breaking the Limits of Message Passing Graph Neural Networks
Muhammet Balcilar, Pierre Heroux, Benoit Gauzere, Pascal Vasseur, Sebastien Adam, Paul Honeine
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 spectral-domain GNNML3 instead of 1-WL MPNNs | graph classification on graph8c · graph8c | undistinguished graph pairs: 386 -> 0 | Improved |
| Use spectral-domain GNNML3 instead of Chebnet | graph discrimination · EXP | undistinguished graph pairs: 71 -> 0 | Improved |
| Use GNNML3 instead of 1-WL MPNNs | EXP classification · EXP-classify | accuracy: 50\% -> 100\% | Improved |
| Use GNNML3 instead of 1-WL MPNNs | graphlet counting · triangle | test MSE: 2.06E-1 -> 4.44E-4 | Improved |
| Use GNNML3 instead of 1-WL MPNNs | graphlet counting · tailed-tri | test MSE: 1.18E-1 -> 3.18E-4 | Improved |
| Use GNNML3 instead of 1-WL MPNNs | graphlet counting · 4-cycles | test MSE: 1.14E-1 -> 6.62E-4 | Improved |
| Use GNNML3 instead of Chebnet | spectral graph classification · planar graph frequency classes | accuracy: 98.2\% -> 97.8\% | Hurt |
| Use GNNML3 instead of PPGN | Zinc12K graph regression · Zinc12K | MSE: 0.1589 $\pm$ 0.007 -> 0.1612 $\pm$ 0.006 | Hurt |
| Use GNNML3 instead of Chebnet | MNIST-75 image graph classification · MNIST-75 | accuracy: 92.08\% $\pm$ 0.22 -> 91.98\% $\pm$ 0.18 | No clear effect |
| Use spectral-domain filters rather than standard MPNN propagation | GNNML3 · Band-Pass node regression | R^2: 0.0629 -> 0.8189 | Improved |
| Use GNNML1 rather than GNNML3 | graphlet counting · triangle | test MSE: 4.44E-4 -> 2.45E-1 | Hurt |
| Set Chebnet maximum eigenvalue to a fixed value | Chebnet on EXP · EXP | undistinguished graph pairs | Hurt |
Closest papers by meaning
- Spatio-Spectral Graph Neural Networks (NeurIPS 2024 · similarity 0.90)
- Improving the Effective Receptive Field of Message-Passing Neural Networks (ICML 2025 · similarity 0.90)
- Accurate and Scalable Graph Neural Networks via Message Invariance (ICLR 2025 · similarity 0.89)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.89)
- Higher-Order Graphon Neural Networks: Approximation and Cut Distance (ICLR 2025 · similarity 0.88)
- Probabilistically Rewired Message-Passing Neural Networks (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.
