Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
139 tokens/sec
GPT-4o
47 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Uncertainty in Graph Neural Networks: A Survey (2403.07185v1)

Published 11 Mar 2024 in cs.LG and stat.ML

Abstract: Graph Neural Networks (GNNs) have been extensively used in various real-world applications. However, the predictive uncertainty of GNNs stemming from diverse sources such as inherent randomness in data and model training errors can lead to unstable and erroneous predictions. Therefore, identifying, quantifying, and utilizing uncertainty are essential to enhance the performance of the model for the downstream tasks as well as the reliability of the GNN predictions. This survey aims to provide a comprehensive overview of the GNNs from the perspective of uncertainty with an emphasis on its integration in graph learning. We compare and summarize existing graph uncertainty theory and methods, alongside the corresponding downstream tasks. Thereby, we bridge the gap between theory and practice, meanwhile connecting different GNN communities. Moreover, our work provides valuable insights into promising directions in this field.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (90)
  1. Bond: Benchmarking unsupervised outlier node detection on static attributed graphs. Advances in Neural Information Processing Systems, 35:27021–27035, 2022.
  2. Uncertainty estimation for molecules: Desiderata and methods. In International conference on machine learning, pages 37133–37156. PMLR, 2023.
  3. Uncertainty quantification of sparse travel demand prediction with spatial-temporal graph neural networks. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 4639–4647, 2022.
  4. Active learning for graph embedding. arXiv preprint arXiv:1705.05085, 2017.
  5. Uncertainty aware semi-supervised learning on graph data. Advances in Neural Information Processing Systems, 33:12827–12836, 2020.
  6. Confidence-based graph convolutional networks for semi-supervised learning. In The 22nd International Conference on Artificial Intelligence and Statistics, pages 1792–1801. PMLR, 2019.
  7. Uag: Uncertainty-aware attention graph neural network for defending adversarial attacks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 7404–7412, 2021.
  8. Gnnexplainer: Generating explanations for graph neural networks. Advances in neural information processing systems, 32, 2019.
  9. A survey of uncertainty in deep neural networks. Artificial Intelligence Review, pages 1–77, 2023.
  10. Good: A graph out-of-distribution benchmark. Advances in Neural Information Processing Systems, 35:2059–2073, 2022.
  11. Alg: Fast and accurate active learning framework for graph convolutional networks. In Proceedings of the 2021 International Conference on Management of Data, pages 2366–2374, 2021.
  12. A review of uncertainty quantification in deep learning: Techniques, applications and challenges. Information fusion, 76:243–297, 2021.
  13. Aleatoric and epistemic uncertainty in machine learning: An introduction to concepts and methods. Machine Learning, 110:457–506, 2021.
  14. A survey of trustworthy graph learning: Reliability, explainability, and privacy protection. arXiv preprint arXiv:2205.10014, 2022.
  15. Quantifying uncertainty in deep spatiotemporal forecasting. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 1841–1851, 2021.
  16. Uncertainty quantification of spatiotemporal travel demand with probabilistic graph neural networks. arXiv preprint arXiv:2303.04040, 2023.
  17. A general framework for quantifying aleatoric and epistemic uncertainty in graph neural networks. Neurocomputing, 521:1–10, 2023.
  18. Gcl: Graph calibration loss for trustworthy graph neural network. In Proceedings of the 30th ACM International Conference on Multimedia, pages 988–996, 2022.
  19. Predictive uncertainty estimation via prior networks. Advances in Neural Information Processing Systems, 31, 2018.
  20. Be confident! towards trustworthy graph neural networks via confidence calibration. Advances in Neural Information Processing Systems, 34:23768–23779, 2021.
  21. On calibration of modern neural networks. In International conference on machine learning, pages 1321–1330. PMLR, 2017.
  22. On calibration of graph neural networks for node classification. In International Joint Conference on Neural Networks, pages 1–8. IEEE, 2022.
  23. What makes graph neural networks miscalibrated? Advances in Neural Information Processing Systems, 35:13775–13786, 2022.
  24. Open-world graph active learning for node classification. ACM Transactions on Knowledge Discovery from Data, 18(2):1–20, 2023.
  25. Graph posterior network: Bayesian predictive uncertainty for node classification. Advances in Neural Information Processing Systems, 34:18033–18048, 2021.
  26. Revisiting uncertainty estimation for node classification: New benchmark and insights, 2023.
  27. Jurygcn: quantifying jackknife uncertainty on graph convolutional networks. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 742–752, 2022.
  28. Bradley Efron. The jackknife, the bootstrap and other resampling plans. SIAM, 1982.
  29. Uncertainty quantification over graph with conformalized graph neural networks. Advances in Neural Information Processing Systems, 2023.
  30. Conformal prediction sets for graph neural networks. In International Conference on Machine Learning, pages 12292–12318. PMLR, 2023.
  31. Bayesian graph convolutional neural networks for semi-supervised classification. In Proceedings of the AAAI conference on artificial intelligence, volume 33, pages 5829–5836, 2019.
  32. Bayesian graph neural networks with adaptive connection sampling. In International conference on machine learning, pages 4094–4104, 2020.
  33. Dropedge: Towards deep graph convolutional networks on node classification. In International Conference on Learning Representations, 2019.
  34. On the temperature of bayesian graph neural networks for conformal prediction. In Advances in Neural Information Processing Systems 2023 Workshop: New Frontiers in Graph Learning, 2023.
  35. Towards ood detection in graph classification from uncertainty estimation perspective. arXiv preprint arXiv:2206.10691, 2022.
  36. Simple and scalable predictive uncertainty estimation using deep ensembles. Advances in neural information processing systems, 30, 2017.
  37. Test-time data augmentation for estimation of heteroscedastic aleatoric uncertainty in deep neural networks. In Medical Imaging with Deep Learning, 2022.
  38. Empowering graph representation learning with test-time graph transformation. In The Eleventh International Conference on Learning Representations, 2022.
  39. Graphpatcher: Mitigating degree bias for graph neural networks via test-time augmentation. Advances in Neural Information Processing Systems, 36, 2024.
  40. Confidence may cheat: Self-training on graph neural networks under distribution shift. In Proceedings of the ACM Web Conference, 2022.
  41. Understanding measures of uncertainty for adversarial example detection. arXiv preprint arXiv:1803.08533, 2018.
  42. Yao Zhang et al. Bayesian semi-supervised learning for uncertainty-calibrated prediction of molecular properties and active learning. Chemical science, 10(35):8154–8163, 2019.
  43. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In international conference on machine learning, pages 1050–1059. PMLR, 2016.
  44. Uncertainty aware graph gaussian process for semi-supervised learning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 4957–4964, 2020.
  45. Yu Song and Donglin Wang. Learning on graphs with out-of-distribution nodes. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 1635–1645, 2022.
  46. A survey of deep active learning. ACM computing surveys (CSUR), 54(9):1–40, 2021.
  47. Information gain propagation: a new way to graph active learning with soft labels. In International Conference on Learning Representations, 2021.
  48. Multitask active learning for graph anomaly detection. arXiv preprint arXiv:2401.13210, 2024.
  49. Active discriminative network representation learning. In IJCAI International Joint Conference on Artificial Intelligence, 2018.
  50. Graph policy network for transferable active learning on graphs. Advances in Neural Information Processing Systems, 33:10174–10185, 2020.
  51. Deeper insights into graph convolutional networks for semi-supervised learning. In Proceedings of the AAAI conference on artificial intelligence, volume 32, 2018.
  52. Multi-stage self-supervised learning for graph convolutional networks on graphs with few labeled nodes. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 5892–5899, 2020.
  53. Self-enhanced gnn: Improving graph neural networks using model outputs. In International Joint Conference on Neural Networks, pages 1–8. IEEE, 2021.
  54. Entropy-aware self-training for graph convolutional networks. Neurocomputing, 464:394–407, 2021.
  55. Theoretical analysis of self-training with deep networks on unlabeled data. In International Conference on Learning Representations, 2021.
  56. Deep insights into noisy pseudo labeling on graph data. In Thirty-seventh Conference on Neural Information Processing Systems, 2023.
  57. Informative pseudo-labeling for graph neural networks with few labels. Data Mining and Knowledge Discovery, pages 228–254, 2023.
  58. Openwgl: Open-world graph learning. In 2020 IEEE international conference on data mining (icdm), pages 681–690. IEEE, 2020.
  59. Learning graph neural networks for multivariate time series anomaly detection. arXiv preprint arXiv:2111.08082, 2021.
  60. Pygod: A python library for graph outlier detection. arXiv preprint arXiv:2204.12095, 2022.
  61. Quantifying the confidence of anomaly detectors in their example-wise predictions. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 227–243. Springer, 2020.
  62. Deep gaussian embedding of graphs: Unsupervised inductive learning via ranking. arXiv preprint arXiv:1707.03815, 2017.
  63. Variational graph recurrent neural networks. Advances in neural information processing systems, 32, 2019.
  64. Variational graph auto-encoders. arXiv preprint arXiv:1611.07308, 2016.
  65. Hyperbolic variational graph neural network for modeling dynamic graphs. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 4375–4383, 2021.
  66. Uncertainty propagation in node classification. In 2022 IEEE International Conference on Data Mining (ICDM), pages 1275–1280. IEEE, 2022.
  67. Ud-gnn: Uncertainty-aware debiased training on semi-homophilous graphs. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 1131–1140, 2022.
  68. A graph is more than its nodes: Towards structured uncertainty-aware learning on graphs. Advances in Neural Information Processing Systems 2023 Workshop: New Frontiers in Graph Learning, 2022.
  69. Unsupervised graph outlier detection: Problem revisit, new insight, and superior method. In 2023 IEEE 39th International Conference on Data Engineering (ICDE), pages 2565–2578. IEEE, 2023.
  70. Calibrate automated graph neural network via hyperparameter uncertainty. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management, pages 4640–4644, 2022.
  71. Robust graph convolutional networks against adversarial attacks. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pages 1399–1407, 2019.
  72. Uncertainty-matching graph neural networks to defend against poisoning attacks. In Proceedings of the Proceedings of the AAAI Conference on Artificial Intelligence Conference on Artificial Intelligence, volume 35, pages 9524–9532, 2021.
  73. A multi-view confidence-calibrated framework for fair and stable graph representation learning. In 2021 IEEE International Conference on Data Mining (ICDM), pages 1493–1498. IEEE, 2021.
  74. A survey on explainability of graph neural networks. arXiv preprint arXiv:2306.01958, 2023.
  75. Explaining predictive uncertainty with information theoretic shapley values. In Thirty-seventh Conference on Neural Information Processing Systems, 2023.
  76. Improving generalization via uncertainty driven perturbations. arXiv preprint arXiv:2202.05737, 2022.
  77. Uncertainty quantification for traffic forecasting: A unified approach. In 2023 IEEE 39th International Conference on Data Engineering (ICDE), pages 992–1004. IEEE, 2023.
  78. A bayesian graph convolutional network for reliable prediction of molecular properties with uncertainty quantification. Chemical science, 10(36):8438–8446, 2019.
  79. Reliable graph neural networks for drug discovery under distributional shift. In NeurIPS 2021 Workshop on Distribution Shifts: Connecting Methods and Applications, 2021.
  80. Uncertainty estimation in deep learning-based property models: Graph neural networks applied to the critical properties. AIChE Journal, 68(6):e17696, 2022.
  81. Uncertainty-aware prediction of chemical reaction yields with graph neural networks. Journal of Cheminformatics, 14:1–10, 2022.
  82. Topology-aware uncertainty for image segmentation. In Thirty-seventh Conference on Neural Information Processing Systems, 2023.
  83. Career mobility analysis with uncertainty-aware graph autoencoders: A job title transition perspective. IEEE Transactions on Computational Social Systems, 2023.
  84. Bayesian hierarchical graph neural networks with uncertainty feedback for trustworthy fault diagnosis of industrial processes. IEEE Transactions on Neural Networks and Learning Systems, 2023.
  85. Uncertainty quantification and aerodynamic robust optimization of turbomachinery based on graph learning methods. Energy, 273:127289, 2023.
  86. Uncertainty-aware correspondence identification for collaborative perception. Autonomous Robots, pages 1–14, 2023.
  87. Modeling user reviews through bayesian graph attention networks for recommendation. ACM Transactions on Information Systems, 41(3):1–29, 2023.
  88. Values: A framework for systematic validation of uncertainty estimation in semantic segmentation. International Conference on Learning Representations, 2024.
  89. Reliable data distillation on graph convolutional network. In Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, pages 1399–1414, 2020.
  90. Distilling the knowledge in a neural network. stat, 1050:9, 2015.
Citations (4)
View on Semantic Scholar

Summary

  • The paper presents a systematic classification of uncertainty in GNNs, distinguishing between aleatoric and epistemic sources.
  • It evaluates various quantification techniques including Bayesian, Frequentist, and ensemble approaches for precise uncertainty assessment.
  • The survey demonstrates how uncertainty analysis enhances GNN reliability in applications such as traffic regulation and molecular property prediction.

Uncertainty in Graph Neural Networks: A Comprehensive Survey

Introduction to Uncertainty in GNNs

Graph Neural Networks (GNNs) have emerged as pivotal tools in deciphering the complex structures inherent in various real-world datasets. However, the inherent and induced uncertainties within these frameworks often compromise the stability and reliability of their predictions. This survey rigorously addresses the critical facets of predictive uncertainty in GNNs, underscoring its significance in enhancing model performance and reliability across numerous applications, including traffic regulation, molecular property prediction, and outlier detection.

Identifying Sources of Uncertainty

Understanding the genesis of uncertainty in GNNs is quintessential. The survey delineates this uncertainty into aleatoric and epistemic categories. Aleatoric uncertainty, or inherent data randomness, and epistemic uncertainty, stemming from model limitations, collectively impact the predictive confidence. This classification provides a foundational step in aptly addressing and quantifying uncertainty in GNN-focused tasks.

Quantifying Uncertainty

The survey presents a taxonomy of methods to quantify uncertainty in GNNs, spanning direct estimation, Bayesian and Frequentist approaches, alongside ensemble models. Each category boasts its strengths and applicability, contingent on the intended downstream tasks and the granularity of uncertainty one aims to capture. For instance, Bayesian-based methods excel in delineating between different sources of uncertainty but at a higher computational cost.

Application in Downstream Tasks

The utilization of uncertainty quantification in GNN tasks has been extensively explored. From improving the selection process in active learning and self-training to enhancing abnormality detection and ensuring model robustness, the survey elaborates on how different types of uncertainty serve divergent yet crucial roles. It also unearths the imperative need for task-specific uncertainty quantification for optimal performance across varied applications.

Real-world Applications

Diving into real-world scenarios, the survey highlights significant domains where uncertainty in GNNs has been pivotal. In traffic management and molecular property prediction, understanding and addressing uncertainty has led to improved predictions and insights. These instances underscore the immense potential of leveraging uncertainty in enhancing the reliability and efficiency of GNNs in practical settings.

Future Directions

Pointing towards future research avenues, the survey emphasizes the necessity for more granular identification of uncertainty, the construction of ground-truth datasets, and the development of unified evaluation metrics. It also advocates for the formulation of more robust, efficient, and application-specific uncertainty quantification methods that cater to the unique challenges posed by GNNs across varied tasks.

Conclusion

In essence, this survey meticulously curates and analyses the existing methodologies, applications, and challenges associated with uncertainty in GNNs. It sheds light on the paramount importance of accurately identifying, quantifying, and utilizing uncertainty, paving the way for more reliable, efficient, and interpretable GNN models in the future. The exploration of uncertainty in GNNs not only enhances model performance but also broadens the horizon for GNN applications across diverse domains.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
Youtube Logo Streamline Icon: https://streamlinehq.com

YouTube