Research map / Graph learning / GNN architectures
How Powerful are Graph Neural Networks?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 GIN-0 with sum aggregation and MLP updates | graph classification models · REDDIT-BINARY | accuracy (%): 87.9 ± 2.5 -> 92.4 ± 2.5 | Helped |
| Use GIN-0 with sum aggregation and MLP updates | graph classification models · REDDIT-MULTI5K | accuracy (%): 54.7 ± 2.9 -> 57.5 ± 1.5 | Helped |
| Use GIN-0 with sum aggregation and MLP updates | graph classification models · MUTAG | accuracy (%): 92.6 ± 4.2 -> 89.4 ± 5.6 | Hurt |
| Use GIN-0 with sum aggregation and MLP updates | graph classification models · NCI1 | accuracy (%): 86.0 ± 1.8 -> 82.7 ± 1.7 | Hurt |
| Learn ε rather than fixing it to 0 (ablation) | GIN-0 · graph classification benchmarks | test accuracy: GIN-0 slightly but consistently outperforms GIN-ε | Hurt |
| Learn ε rather than fixing it to 0 (ablation) | GIN-0 · training sets | training accuracy | No clear effect |
| Replace sum aggregation with mean aggregation (ablation) | GIN-0 · REDDIT-BINARY | accuracy (%): 92.4 ± 2.5 -> 50.0 ± 0.0 | Hurt |
| Replace sum aggregation with mean aggregation (ablation) | GIN-0 · REDDIT-MULTI5K | accuracy (%): 57.5 ± 1.5 -> 20.0 ± 0.0 | Hurt |
| Replace MLPs with 1-layer perceptrons (ablation) | sum-aggregation GNN · graph classification benchmarks | training accuracy | Hurt |
| Replace sum aggregation with mean aggregation (ablation) | GIN-0 · graph classification benchmarks | training accuracy | Hurt |
| Replace sum aggregation with max pooling (ablation) | GIN-0 · graph classification benchmarks | training accuracy | Hurt |
| Provide node degrees as input features to mean aggregation (ablation) | mean-MLP · REDDIT-BINARY | accuracy (%): 71.2±4.6% vs. sum-based GNNs | Hurt |
Closest papers by meaning
- A New Perspective on "How Graph Neural Networks Go Beyond Weisfeiler-Lehman?" (ICLR 2022 · similarity 0.92)
- A Fair Comparison of Graph Neural Networks for Graph Classification (ICLR 2020 · similarity 0.92)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.91)
- Can Classic GNNs Be Strong Baselines for Graph-level Tasks? Simple Architectures Meet Excellence (ICML 2025 · similarity 0.91)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.90)
- Improving Distinguishability of Class for Graph Neural Networks (AAAI 2024 · similarity 0.90)
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.