Faithful Interpretation for Graph Neural Networks
Not on the map. Why?
Abstract (arXiv)
Currently, attention mechanisms have garnered increasing attention in Graph Neural Networks (GNNs), such as Graph Attention Networks (GATs) and Graph Transformers (GTs). It is not only due to the commendable boost in performance they offer but also its capacity to provide a more lucid rationale for model behaviors, which are often viewed as inscrutable. However, Attention-based GNNs have demonstrated instability in interpretability when subjected to various sources of perturbations during both training and testing phases, including factors like additional edges or nodes. In this paper, we propose a solution to this problem by introducing a novel notion called Faithful Graph Attention-based Interpretation (FGAI). In particular, FGAI has four crucial properties regarding stability and sensitivity to interpretation and final output distribution. Built upon this notion, we propose an efficient methodology for obtaining FGAI, which can be viewed as an ad hoc modification to the canonical Attention-based GNNs. To validate our proposed solution, we introduce two novel metrics tailored for graph interpretation assessment. Experimental results demonstrate that FGAI exhibits superior stability and preserves the interpretability of attention under various forms of perturbations and randomness, which makes FGAI a more faithful and reliable explanation tool.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 47 of its 61 references that are in the map. First posted on arXiv 9 Oct 2024.
Builds on
47 of 61 references.
References · 47
- 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,282
- 2017Graph Attention Networkscited by 8,854
- 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
- 2017Inductive Representation Learning on Large Graphscited by 6,284
- 2018How Powerful are Graph Neural Networks?cited by 4,355
- 2013Distributed Representations of Words and Phrases and their Compositionalitycited by 3,462
- 2017Axiomatic Attribution for Deep Networkscited by 3,350
- 2018Sanity Checks for Saliency Mapscited by 1,149
- 2018Pitfalls of Graph Neural Network Evaluationcited by 994
- 2021How Attentive are Graph Attention Networks?cited by 818
- 2019Heterogeneous Graph Attention Networkcited by 732
- 2015Image-based Recommendations on Styles and Substitutescited by 702
35 more
- 2020A Generalization of Transformer Networks to Graphscited by 653
- 2019Attention is not not Explanationcited by 629
- 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 590
- 2017Interpretation of Neural Networks is Fragilecited by 443
- 2020Parameterized Explainer for Graph Neural Networkcited by 423
- 2018Adversarial Attacks on Neural Networks for Graph Datacited by 395
- 2020Explainability in Graph Neural Networks: A Taxonomic Surveycited by 383
- 2020Graph Structure Learning for Robust Graph Neural Networkscited by 368
- 2019On the (In)fidelity and Sensitivity for Explanationscited by 309
- 2018Adversarial Attack on Graph Structured Datacited by 300
- 2021On Explainability of Graph Neural Networks via Subgraph Explorationscited by 280
- 2019Explanations can be manipulated and geometry is to blamecited by 215
- 2020GNNGuard: Defending Graph Neural Networks against Adversarial Attackscited by 195
- 2022Interpretable and Generalizable Graph Learning via Stochastic Attention Mechanismcited by 171
- 2022Towards Faithful Model Explanation in NLP: A Surveycited by 129
- 2019Stability and Generalization of Graph Convolutional Neural Networkscited by 81
- 2021User Preference-aware Fake News Detectioncited by 76
- 2021Graph-Based Deep Learning for Medical Diagnosis and Analysis: Past, Present and Futurecited by 67
- 2019Certifiable Robustness to Graph Perturbationscited by 62
- 2019Certifiable Robustness and Robust Training for Graph Convolutional Networkscited by 59
- 2017The Promise and Peril of Human Evaluation for Model Interpretabilitycited by 52
- 2021Graph Robustness Benchmark: Benchmarking the Adversarial Robustness of Graph Machine Learningcited by 44
- 2022Reinforced Causal Explainer for Graph Neural Networkscited by 34
- 2021On the Sensitivity and Stability of Model Interpretations in NLPcited by 30
- 2022SEAT: Stable and Explainable Attentioncited by 25
- 2024Editable Concept Bottleneck Modelscited by 22
- 2021Graph-based Deep Learning for Communication Networks: A Surveycited by 21
- 2024Semi-supervised Concept Bottleneck Modelscited by 21
- 2024Graph Neural Network Explanations are Fragilecited by 18
- 2024A Hopfieldian View-based Interpretation for Chain-of-Thought Reasoningcited by 15
- 2024How Interpretable Are Interpretable Graph Neural Networks?cited by 13
- 2021GCN-SE: Attention as Explainability for Node Classification in Dynamic Graphscited by 12
- 2023GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarkingcited by 11
- 2024Faithful and Accurate Self-Attention Attribution for Message Passing Neural Networks via the Computation Tree Viewpointcited by 5
- 2023Semantic Interpretation and Validation of Graph Attention-based Explanations for GNN Modelscited by 3
Built on by
Citing papers per year: 2024: 1; 2025: 2.
3 papers.