Research map / Graph learning / GNN architectures
GNNInterpreter: A Probabilistic Generative Model-Level Explanation for 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 GNNInterpreter instead of XGNN for MUTAG explanations | GCN trained on MUTAG · MUTAG Mutagen and Nonmutagen | class probability: 0.986 ± 0.057 -> 1.000 ± 0.000; 0.991 ± 0.083 -> 1.000 ± 0.000 | Helped |
| Use GNNInterpreter instead of XGNN | GCN trained on MUTAG · MUTAG | training time per class: 128 s -> 12 s | Helped |
| Use GNNInterpreter instead of XGNN for cyclic-class explanation | GCN trained on Is_Acyclic · Is_Acyclic Cyclic | class probability: 0.076 ± 0.000 -> 0.999 ± 0.001 | Helped |
| Use GNNInterpreter instead of XGNN for acyclic-class explanation | GCN trained on Is_Acyclic · Is_Acyclic Acyclic | class probability: 0.927 ± 0.000 -> 1.000 ± 0.000 | Helped |
| Use GNNInterpreter instead of XGNN | GCN trained on Is_Acyclic · Is_Acyclic | training time per class: 45 s -> 20 s | Helped |
| Remove the embedding-similarity objective term (ablation) | GNNInterpreter on MUTAG Mutagen · MUTAG Mutagen | explanation in-distribution faithfulness: null | Hurt |
| Remove the embedding-similarity objective term (ablation) | GNNInterpreter on MUTAG Mutagen · MUTAG Mutagen | logit: 43.45 -> 310.14; 34.98 -> 286.42 | Helped |
| Add the embedding-similarity objective term (ablation) | GNNInterpreter · MUTAG Mutagen | explanation in-distribution faithfulness: null | Helped |
| Generate explanations with GNNInterpreter | GCN trained on Motif · Motif House-X misleading motifs | misclassification of non-House-X graphs as House-X: null | Mixed |
| Generate explanations with GNNInterpreter | GCN trained on Motif · Motif Complete-4 misleading motifs | predicted Complete-4 class probability: 0.5875; 0.593; 0.5874; 0.596; 0.5881; 0.5978 | Mixed |
| Generate explanations with GNNInterpreter | GAT trained on ColorConsistency · ColorConsistency Consistent explanation | semantic validity of explanation: null | Hurt |
Closest papers by meaning
- [Re] GNNInterpreter: A probabilistic generative model-level explanation for Graph Neural Networks (NeurIPS 2024 · similarity 0.92)
- GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules (NeurIPS 2024 · similarity 0.90)
- Stratified GNN Explanations through Sufficient Expansion (AAAI 2024 · similarity 0.90)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.89)
- COMRECGC: Global Graph Counterfactual Explainer through Common Recourse (ICML 2025 · similarity 0.89)
- EiG-Search: Generating Edge-Induced Subgraphs for GNN Explanation in Linear Time (ICML 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.