Research map / Graph learning / GNN architectures
ProtGNN: Towards Self-Explaining 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 |
|---|---|---|---|
| Add prototype learning to GCN | GCN · BBBP | accuracy (%): 84.6$\pm 3.4$ -> 89.4$\mathbf{\pm 4.1}$ | Helped |
| Add prototype learning to GCN | GCN · Graph-SST2 | accuracy (%): 89.7$\pm 0.5$ -> 89.9$\mathbf{\pm 2.4}$ | No clear effect |
| Add prototype learning to GCN | GCN · Graph-Twitter | accuracy (%): 67.5$\pm 1.9$ -> 68.9$\mathbf{\pm 5.9}$ | No clear effect |
| Add prototype learning to GCN | GCN · BA-Shape | accuracy (%): 91.9$\pm 1.7$ -> 95.7$\mathbf{\pm 1.4}$ | Helped |
| Add conditional subgraph sampling to ProtGNN | GCN ProtGNN · BBBP | accuracy (%): 89.4$\mathbf{\pm 4.1}$ -> 88.0$\pm 4.6$ | No clear effect |
| Use conditional subgraph sampling instead of MCTS for ProtGNN+ sampling | ProtGNN+ · BBBP | training time: $\textgreater$ 2 hrs -> 632.7 s | Helped |
| Add prototype projection with MCTS | GCN · BBBP | training time: 177.9 s -> 506.3 s | Hurt |
| Vary prototypes per class; increasing the number beyond 5 (ablation) | ProtGNN · BBBP and Graph-Twitter | accuracy | Helped |
| Remove diversity loss (ablation) | ProtGNN · BBBP | prototype diversity | Hurt |
| Add diversity loss (ablation) | ProtGNN · BBBP | prototype diversity | Helped |
| Vary cluster and separation loss weights; use $\lambda_1 = 0.10$ and $\lambda_2 = 0.05$ (ablation) | ProtGNN · BBBP | accuracy | Helped |
Closest papers by meaning
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.90)
- Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 2024 · similarity 0.90)
- GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules (NeurIPS 2024 · similarity 0.89)
- EiG-Search: Generating Edge-Induced Subgraphs for GNN Explanation in Linear Time (ICML 2024 · similarity 0.89)
- Graph Segmentation and Contrastive Enhanced Explainer for Graph Neural Networks (AAAI 2025 · similarity 0.89)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · 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.