Research map / Graph learning / Graph self-supervised learning
Graph Neural Networks with Adaptive Readouts
David Buterez, Jon Paul Janet, Steven J Kiddle, Dino Oglic, Pietro Liò
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 |
|---|---|---|---|
| replace sum/mean/max readouts with neural readouts | GNN · regression datasets (R2) | R2 ratio: 36 of 45 configurations improved | Improved |
| replace sum/mean/max readouts with neural readouts | GNN · classification datasets (MCC) | MCC: 93 of 147 improved; min ratio 0.79 | Mixed |
| increase set transformer heads from 1 to 8 (ablation) | ST minimal GCN · QM9 | R2: 0.20 -> 0.35 | Improved |
| increase heads beyond 4 or add self-attention block (ablation) | Set Transformer readout · QM9 | R2: 0.35 -> 0.38 (complex GCN, 4 -> 8 heads) | No clear effect |
| use MLP readout instead of standard readouts on 1M+ graphs | VGAE GCN · proprietary datasets | R2: 0.33 -> 0.78 | Improved |
| use neural readouts on 1.5M graph dataset | VGAE GCN · proprietary | R2: 0.07 -> 0.64 | Improved |
| use neural readouts on 2M graph dataset | VGAE GCN · proprietary | R2: 0.06 -> 0.52 | Improved |
| add more GNN layers (2,3,4) | GNN/VGAE · proprietary 1M+ datasets | train loss | No clear effect |
| increase parameter count by switching from MLP to ST | readouts · QM9 | performance | No clear effect |
| use neural readouts for convergence | GNN · QM9 | epochs to converge: 500-1000 -> under 100 epochs | Improved |
| train MLP/GRU readouts and evaluate on node permutations | MLP/GRU readouts · QM9 permutations | error | Hurt |
| use Janossy readouts instead of standard | Janossy readout · QM9 | MAE and R2 | Improved |
Closest papers by meaning
- Affinity-Aware Graph Networks (NeurIPS 2023 · similarity 0.88)
- Bridging Input Feature Spaces Towards Graph Foundation Models (ICLR 2026 · similarity 0.88)
- Higher-Order Graphon Neural Networks: Approximation and Cut Distance (ICLR 2025 · similarity 0.86)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.86)
- Equivariance Everywhere All At Once: A Recipe for Graph Foundation Models (NeurIPS 2025 · similarity 0.86)
- On the Expressivity and Sample Complexity of Node-Individualized Graph Neural Networks (NeurIPS 2024 · similarity 0.86)
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.
