Research map / Graph learning / GNN architectures
Paradigm Shift of GNN Explainer from Label Space to Prototypical Representation Space
Jun Yin, Senzhang Wang, Ziluowen Luo, Peng Huo, Hao Yan, Hao Miao, Chaozhuo Li, Shirui Pan, Chengqi Zhang
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 |
|---|---|---|---|
| shift explainer optimization from label space to prototypical representation space (IDEA) | PGExplainer · Average over 5 datasets | ROC-AUC: 0.8000 -> 0.8856 | Improved |
| add IDEA over best baseline | IDEA · Average over 5 datasets | ROC-AUC: 0.8479 -> 0.8856 | Improved |
| add IDEA over best baseline | IDEA · Average over 5 datasets | Precision: 0.4050 -> 0.6022 | Improved |
| use prototypical alignment instead of direct alignment in GNN encoded space | IDEA · Average over 5 datasets | ROC-AUC: 0.7116 -> 0.8856 | Improved |
| add IDEA to ReFine | ReFine · Average over 5 datasets | ROC-AUC: 0.6912 -> 0.7538 | Improved |
| add IDEA to ProxyExplainer | ProxyExplainer · Average over 5 datasets | ROC-AUC: 0.8479 -> 0.8690 | Improved |
| add IDEA to V-InfoR | V-InfoR · Mutagenicity | ROC-AUC: 0.6075 -> 0.5734 | Hurt |
| add IDEA to V-InfoR | V-InfoR · Average over 5 datasets | ROC-AUC: 0.6683 -> 0.6696 | No clear effect |
| use Wasserstein distance instead of KL divergence in alignment (ablation) | IDEA · Average over 5 datasets | ROC-AUC: 3.33% average improvement | Improved |
| add structural information disentanglement stage (ablation) | IDEA · Average over 5 datasets | ROC-AUC: gap of 0.0682 | Improved |
| unify representations in prototypical space instead of aligning raw representations (MSE or InfoNCE) (ablation) | IDEA · Average over 5 datasets | ROC-AUC | Improved |
| use dual-stage training instead of joint optimization (ablation) | IDEA · Average over 4 datasets | ROC-AUC: 0.6832 -> 0.8676 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- V-InFoR — V-InFoR: A Robust Graph Neural Networks Explainer for Structurally Corrupted Graphs (NeurIPS 2023) · better on 2 of 2 shared results
- D4Explainer — D4Explainer: In-distribution Explanations of Graph Neural Network via Discrete Denoising Diffusion (NeurIPS 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.91)
- Factorized Explainer for Graph Neural Networks (AAAI 2024 · similarity 0.91)
- GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules (NeurIPS 2024 · similarity 0.91)
- Stratified GNN Explanations through Sufficient Expansion (AAAI 2024 · similarity 0.91)
- Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 2024 · similarity 0.90)
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · 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.
