Research map / Graph learning / GNN architectures
Self-Consistency Improves the Trustworthiness of Self-Interpretable GNNs
Wenxin Tai, Ting Zhong, Goce Trajcevski, Fan Zhou
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 |
|---|---|---|---|
| add self-consistency fine-tuning | SMGNN · BA-2Motifs | SHD (%): 10.44±4.41 -> 3.48±2.20 | Improved |
| add self-consistency fine-tuning | GSAT · BA-2Motifs | AUC (%): 98.44±0.61 -> 99.30±0.12 | Improved |
| add self-consistency fine-tuning | SMGNN · Benzene | AUC (%): 84.38±2.71 -> 90.07±1.56 | Improved |
| add self-consistency fine-tuning | GSAT · Benzene | SHD (%): 6.93±2.75 -> 2.32±0.62 | Improved |
| add explanation ensemble on top of SC | SMGNN+SC · BA-2Motifs | SHD (%): 3.48±2.20 -> 1.52±1.03 | Improved |
| add self-consistency fine-tuning | SMGNN · 3MR | FID- (%): 1.90±1.01 -> 0.00±0.00 | Improved |
| add SC loss without conciseness regularization (ablation) | GAT · Mutagenicity | AUC (%): 91.71±6.00 -> 81.79±12.31 | Hurt |
| add SC loss without conciseness regularization (ablation) | GAT · BA-2Motifs | SHD (%): 10.41±4.37 -> 16.63±15.21 | Hurt |
| add SC loss without conciseness regularization (ablation) | CAL · BA-2Motifs | SHD (%): 11.40±4.04 -> 17.54±15.31 | Hurt |
| add conciseness regularization to GAT+SC (ablation) | GAT+SC · Mutagenicity | AUC (%): 81.79±12.31 -> 98.30±0.54 | Improved |
| add self-consistency fine-tuning | SMGNN · 3MR | L1 distance between successive explanations (%): 51.01±32.81 -> 2.83±8.84 | Improved |
| add self-consistency fine-tuning | GSAT · Benzene | FID+ (%): 42.90±10.18 -> 28.58±12.25 | Hurt |
Closest papers by meaning
- Redundancy Undermines the Trustworthiness of Self-Interpretable GNNs (ICML 2025 · similarity 0.95)
- Reconsidering Faithfulness in Regular, Self-Explainable and Domain Invariant GNNs (ICLR 2025 · similarity 0.94)
- Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 2024 · similarity 0.93)
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.92)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.91)
- GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking (ICLR 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.
