mlsanityData to 14 Oct 2026

GNN Explanations that do not Explain and How to find Them

arXiv 2601.20815 ↗First posted 28 Jan 2026cs.LG cs.AIcited by 5

Not on the map. Why?

Abstract (arXiv)

Explanations provided by Self-explainable Graph Neural Networks (SE-GNNs) are fundamental for understanding the model's inner workings and for identifying potential misuse of sensitive attributes. Although recent works have highlighted that these explanations can be suboptimal and potentially misleading, a characterization of their failure cases is unavailable. In this work, we identify a critical failure of SE-GNN explanations: explanations can be unambiguously unrelated to how the SE-GNNs infer labels. We show that, on the one hand, many SE-GNNs can achieve optimal true risk while producing these degenerate explanations, and on the other, most faithfulness metrics can fail to identify these failure modes. Our empirical analysis reveals that degenerate explanations can be maliciously planted (allowing an attacker to hide the use of sensitive attributes) and can also emerge naturally, highlighting the need for reliable auditing. To address this, we introduce a novel faithfulness metric that reliably marks degenerate explanations as unfaithful, in both malicious and natural settings. Our code is available at https://github.com/steveazzolin/gnn_deg_expl.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 69 of its 86 references that are in the map. First posted on arXiv 28 Jan 2026.

Builds on

69 of 86 references.

References · 69

  1. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 7,005
  2. 2019GNNExplainer: Generating Explanations for Graph Neural Networkscited by 737
  3. 2018Which Training Methods for GANs do actually Converge?cited by 679
  4. 2019Fooling LIME and SHAP: Adversarial Attacks on Post hoc Explanation Methodscited by 443
  5. 2020Parameterized Explainer for Graph Neural Networkcited by 426
  6. 2020Explainability in Graph Neural Networks: A Taxonomic Surveycited by 384
  7. 2021On Explainability of Graph Neural Networks via Subgraph Explorationscited by 281
  8. 2019Explanations can be manipulated and geometry is to blamecited by 216
  9. 2022Discovering Invariant Rationales for Graph Neural Networkscited by 188
  10. 2022Interpretable and Generalizable Graph Learning via Stochastic Attention Mechanismcited by 171
  11. 2019Understanding Attention and Generalization in Graph Neural Networkscited by 169
  12. 2022Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphscited by 144
38 more
  1. 2022GOOD: A Graph Out-of-Distribution Benchmarkcited by 141
  2. 2020Graph Information Bottleneck for Subgraph Recognitioncited by 119
  3. 2021Counterfactual Explanations Can Be Manipulatedcited by 113
  4. 2019Fooling Neural Network Interpretations via Adversarial Model Manipulationcited by 112
  5. 2019A Gentle Introduction to Deep Learning for Graphscited by 108
  6. 2021Causal Attention for Interpretable and Generalizable Graph Classificationcited by 108
  7. 2019Rethinking Cooperative Rationalization: Introspective Extraction and Complement Controlcited by 96
  8. 2021ProtGNN: Towards Self-Explaining Graph Neural Networkscited by 96
  9. 2021The Out-of-Distribution Problem in Explainability and Search Methods for Feature Importance Explanationscited by 87
  10. 2022Evaluating Explainability for Graph Neural Networkscited by 87
  11. 2022Learning and Evaluating Graph Neural Network Explanations based on Counterfactual and Factual Reasoningcited by 79
  12. 2021Towards Self-Explainable Graph Neural Networkcited by 75
  13. 2022Global Explainability of GNNs via Logic Combination of Learned Conceptscited by 65
  14. 2023Not All Neuro-Symbolic Concepts Are Created Equal: Analysis and Mitigation of Reasoning Shortcutscited by 63
  15. 2021Have We Learned to Explain?: How Interpretability Methods Can Learn to Encode Predictions in their Interpretationscited by 59
  16. 2022GraphFramEx: Towards Systematic Evaluation of Explainability Methods for Graph Neural Networkscited by 58
  17. 2022Framework for Evaluating Faithfulness of Local Explanationscited by 55
  18. 2022Leveraging Explanations in Interactive Machine Learning: An Overviewcited by 53
  19. 2025Position: Graph Learning Will Lose Relevance Due To Poor Benchmarkscited by 53
  20. 2021Improving Subgraph Recognition with Variational Graph Information Bottleneckcited by 45
  21. 2021Understanding Interlocking Dynamics of Cooperative Rationalizationcited by 36
  22. 2022Explaining the Explainers in Graph Neural Networks: a Comparative Studycited by 35
  23. 2019A Game Theoretic Approach to Class-wise Selective Rationalizationcited by 32
  24. 2022Interpretable Geometric Deep Learning via Learnable Randomness Injectioncited by 27
  25. 2024The Interpretable and Effective Graph Neural Additive Networkscited by 20
  26. 2023Towards Trustworthy Explanation: On Causal Rationalizationcited by 19
  27. 2024PowerGraph: A power grid benchmark dataset for graph neural networkscited by 19
  28. 2024Graph Neural Network Explanations are Fragilecited by 19
  29. 2024A Neuro-Symbolic Benchmark Suite for Concept Quality and Reasoning Shortcutscited by 19
  30. 2022FR: Folded Rationalization with a Unified Encodercited by 18
  31. 2024F-Fidelity: A Robust Framework for Faithfulness Evaluation of Explainable AIcited by 18
  32. 2023Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networkscited by 17
  33. 2024Reconsidering Faithfulness in Regular, Self-Explainable and Domain Invariant GNNscited by 16
  34. 2023Enhancing the Rationale-Input Alignment for Self-explaining Rationalizationcited by 15
  35. 2020Graph Neural Networks Including Sparse Interpretabilitycited by 13
  36. 2024How Interpretable Are Interpretable Graph Neural Networks?cited by 13
  37. 2025Shortcuts and Identifiability in Concept-based Models from a Neuro-Symbolic Lenscited by 13
  38. 2025Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanationscited by 13

50 most cited of 69 All 69 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026A Generalized Tikhonov Layer for Interpretable-by-design Graph Neural Networkscited by 1
  2. 2026Why Self-Inconsistency Arises in GNN Explanations and How to Exploit Itcited by 0
  3. 2026Overcoming Shortcut Learning in Graph Neural Networks through Active Explanation Guidancecited by 0
  4. 2026Representable but Unlearned: Encoding Rank and the Interaction-Prediction Floorcited by 0
  5. 2026Beyond Trained Models: Compiling GNNs for a Sound Explainer Benchmarkcited by 0