Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
167 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
42 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

GNNInterpreter: A Probabilistic Generative Model-Level Explanation for Graph Neural Networks (2209.07924v4)

Published 15 Sep 2022 in cs.LG and cs.AI

Abstract: Recently, Graph Neural Networks (GNNs) have significantly advanced the performance of machine learning tasks on graphs. However, this technological breakthrough makes people wonder: how does a GNN make such decisions, and can we trust its prediction with high confidence? When it comes to some critical fields, such as biomedicine, where making wrong decisions can have severe consequences, it is crucial to interpret the inner working mechanisms of GNNs before applying them. In this paper, we propose a model-agnostic model-level explanation method for different GNNs that follow the message passing scheme, GNNInterpreter, to explain the high-level decision-making process of the GNN model. More specifically, GNNInterpreter learns a probabilistic generative graph distribution that produces the most discriminative graph pattern the GNN tries to detect when making a certain prediction by optimizing a novel objective function specifically designed for the model-level explanation for GNNs. Compared to existing works, GNNInterpreter is more flexible and computationally efficient in generating explanation graphs with different types of node and edge features, without introducing another blackbox or requiring manually specified domain-specific rules. In addition, the experimental studies conducted on four different datasets demonstrate that the explanation graphs generated by GNNInterpreter match the desired graph pattern if the model is ideal; otherwise, potential model pitfalls can be revealed by the explanation. The official implementation can be found at https://github.com/yolandalalala/GNNInterpreter.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (32)
  1. Explainability techniques for graph convolutional networks. ArXiv, abs/1905.13686, 2019.
  2. Deep neural networks for learning graph representations. AAAI’16, pp.  1145–1152. AAAI Press, 2016.
  3. Structure-activity relationship of mutagenic aromatic and heteroaromatic nitro compounds. correlation with molecular orbital energies and hydrophobicity. Journal of Medicinal Chemistry, 34(2):786–797, 1991. doi: 10.1021/jm00106a046. URL https://doi.org/10.1021/jm00106a046.
  4. Graphsvx: Shapley value explanations for graph neural networks. In European Conference on Machine Learning and Knowledge Discovery in Databases (ECML PKDD), 2021.
  5. P ERDdS and A R&WI. On random graphs i. 1960.
  6. When comparing to ground truth is wrong: On evaluating gnn explanation methods. KDD ’21, pp.  332–341, New York, NY, USA, 2021. Association for Computing Machinery. ISBN 9781450383325. doi: 10.1145/3447548.3467283. URL https://doi.org/10.1145/3447548.3467283.
  7. E. N. Gilbert. Random graphs. The Annals of Mathematical Statistics, 30(4):1141–1144, 1959. ISSN 00034851. URL http://www.jstor.org/stable/2237458.
  8. Neural message passing for quantum chemistry. 04 2017.
  9. Inductive representation learning on large graphs. 06 2017.
  10. Benchmark data sets for graph kernels, 2016. URL http://graphkernels.cs.tu-dortmund.de.
  11. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations (ICLR), 2017.
  12. Graph classification using structural attention. pp.  1666–1674, 07 2018. doi: 10.1145/3219819.3219980.
  13. Generative causal explanations for graph neural networks. In International Conference on Machine Learning, pp. 6666–6679. PMLR, 2021.
  14. Orphicx: A causality-inspired latent variable model for interpreting graph neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.  13729–13738, 2022.
  15. Parameterized explainer for graph neural network. Advances in Neural Information Processing Systems, 33, 2020.
  16. The concrete distribution: A continuous relaxation of discrete random variables. 11 2016.
  17. Deep neural networks are easily fooled: High confidence predictions for unrecognizable images. pp.  427–436, 06 2015. doi: 10.1109/CVPR.2015.7298640.
  18. Richard Rado. Universal graphs and universal functions. Acta Arithmetica, 9(4):331–340, 1964.
  19. Layerwise relevance visualization in convolutional text graph classifiers. ArXiv, abs/1909.10911, 2019.
  20. Grad-cam: Visual explanations from deep networks via gradient-based localization. International Journal of Computer Vision, 128:336–359, 2019.
  21. Reinforcement learning enhanced explainer for graph neural networks. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan (eds.), Advances in Neural Information Processing Systems, volume 34, pp.  22523–22533. Curran Associates, Inc., 2021. URL https://proceedings.neurips.cc/paper/2021/file/be26abe76fb5c8a4921cf9d3e865b454-Paper.pdf.
  22. Deep inside convolutional networks: Visualising image classification models and saliency maps. CoRR, abs/1312.6034, 2014.
  23. Graph Attention Networks. International Conference on Learning Representations, 2018. URL https://openreview.net/forum?id=rJXMpikCZ. accepted as poster.
  24. On gnn explanability with activation patterns. 2021.
  25. Pgm-explainer: Probabilistic graphical model explanations for graph neural networks. ArXiv, abs/2010.05788, 2020.
  26. Trustworthy graph learning: Reliability, explainability, and privacy protection. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, pp.  4838–4839, New York, NY, USA, 2022. Association for Computing Machinery. ISBN 9781450393850. doi: 10.1145/3534678.3542597. URL https://doi.org/10.1145/3534678.3542597.
  27. Spatio-temporal attentive rnn for node classification in temporal attributed graphs. pp.  3947–3953, 08 2019. doi: 10.24963/ijcai.2019/548.
  28. Gnnexplainer: Generating explanations for graph neural networks. Advances in neural information processing systems, 32:9240–9251, 12 2019.
  29. Xgnn: Towards model-level explanations of graph neural networks. pp.  430–438, 08 2020a. doi: 10.1145/3394486.3403085.
  30. Explainability in graph neural networks: A taxonomic survey. ArXiv, abs/2012.15445, 2020b.
  31. On explainability of graph neural networks via subgraph explorations. In Marina Meila and Tong Zhang (eds.), Proceedings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of Machine Learning Research, pp.  12241–12252. PMLR, 18–24 Jul 2021. URL https://proceedings.mlr.press/v139/yuan21c.html.
  32. Link prediction based on graph neural networks. 02 2018.
Citations (33)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now