Research map / Graph learning / Graph self-supervised learning
[Re] On Explainability of Graph Neural Networks via Subgraph Explorations
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 the reduced player set and approximation scheme (compare against Monte Carlo Shapley on all nodes) (ablation) | SubgraphX Shapley scoring · BBBP efficiency | explanation time: 865.4 -> 77.8 | Helped |
| Add the reduced player set (versus MCTS-dagger) (ablation) | SubgraphX · BBBP efficiency | Fidelity: 0.53 -> 0.55 | No clear effect |
| Replace exact Shapley enumeration with Monte Carlo sampling on reduced set (ablation) | SubgraphX · BBBP efficiency | explanation time: >10 hours -> 77.8s | Helped |
| Use SubgraphX instead of GNNExplainer | explanation method · BBBP efficiency | Fidelity: 0.19 -> 0.55 | Helped |
| Use SubgraphX instead of PGExplainer | explanation method · BBBP efficiency | Fidelity: 0.18 -> 0.55 | Helped |
| Use SubgraphX instead of GNNExplainer | explanation method · BBBP efficiency | explanation time: 16.2 -> 77.8 | Hurt |
| Use SubgraphX instead of PGExplainer (which needs training on whole dataset) | explanation method · BBBP efficiency | explanation time: 0.02 -> 77.8 | Hurt |
| Switch pruning from High2low to Low2high (ablation) | SubgraphX · BBBP pruning study | Fidelity: 0.61046 -> 0.66149 | Helped |
| Switch pruning from Low2high to High2low (ablation) | SubgraphX · BBBP pruning study | explanation time: 107.24 -> 21.52 | Helped |
| Use SubgraphX over baselines on Fidelity-versus-Sparsity curves | SubgraphX · MUTAG, BBBP, Graph-SST2, BA-2Motifs, BA-Shape | Fidelity at matched Sparsity: outperforms in 5 of 6 experiments | Mixed |
| Use SubgraphX instead of PGExplainer for node classification | SubgraphX · BA-Shape (GCN) | Fidelity at matched Sparsity | No clear effect |
| Use Shapley scoring instead of raw GNN prediction as MCTS reward (MCTS_GNN baseline) (ablation) | SubgraphX · MUTAG, Graph-SST2 | explanation quality (qualitative) | Helped |
Closest papers by meaning
- Generating In-Distribution Proxy Graphs for Explaining Graph Neural Networks (ICML 2024 · similarity 0.90)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 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)
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.89)
- Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 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.