[论文解读] Trustworthy Graph Neural Networks: Aspects, Methods and Trends

He Zhang, Bang Ye Wu+4 more|arXiv (Cornell University)|May 16, 2022

Adversarial Robustness in Machine Learning参考文献 365被引用 39

一句话总结

本次综述定义了可信赖的 GNNs，并系统性地评述 six aspects（robustness、explainability、privacy、fairness、accountability、environmental well-being），概述 open framework、metrics，以及 cross-aspect relations。并讨论研究和工业化的 trending directions。

ABSTRACT

Graph neural networks (GNNs) have emerged as a series of competent graph learning methods for diverse real-world scenarios, ranging from daily applications like recommendation systems and question answering to cutting-edge technologies such as drug discovery in life sciences and n-body simulation in astrophysics. However, task performance is not the only requirement for GNNs. Performance-oriented GNNs have exhibited potential adverse effects like vulnerability to adversarial attacks, unexplainable discrimination against disadvantaged groups, or excessive resource consumption in edge computing environments. To avoid these unintentional harms, it is necessary to build competent GNNs characterised by trustworthiness. To this end, we propose a comprehensive roadmap to build trustworthy GNNs from the view of the various computing technologies involved. In this survey, we introduce basic concepts and comprehensively summarise existing efforts for trustworthy GNNs from six aspects, including robustness, explainability, privacy, fairness, accountability, and environmental well-being. Additionally, we highlight the intricate cross-aspect relations between the above six aspects of trustworthy GNNs. Finally, we present a thorough overview of trending directions for facilitating the research and industrialisation of trustworthy GNNs.

研究动机与目标

• Define trustworthy GNNs and motivate their importance beyond performance.
• Propose an open framework to characterize trust-oriented aspects in GNNs.
• Summarize typical methods, evaluation metrics, and taxonomy for each aspect.
• Analyze cross-aspect relations and trade-offs among trustworthiness dimensions.
• Highlight trends and directions to facilitate research and industrial deployment of trustworthy GNNs.

提出的方法

• Introduce core trustworthiness concepts and six aspects tailored to GNNs.
• Provide fine-grained taxonomy and metrics for each aspect (robustness, explainability, privacy, fairness, accountability, environmental well-being).
• Categorize defence and acquisition methods across phases (pre-training, training, post-training) for robustness.
• Discuss cross-aspect relationships and how improvements in one aspect affect others.
• Compare with related surveys and articulate an open framework extensible to new trust-oriented aspects.

实验结果

研究问题

• RQ1 What constitutes trustworthiness in GNNs and how can it be measured?
• RQ2 What methods exist to improve each trust aspect in GNNs, and how are they classified across development stages?
• RQ3 How do different trust aspects interact or conflict with each other in GNN design and evaluation?
• RQ4 What directions are promising for advancing trustworthy GNNs in research and industry?

主要发现

• An open framework for trustworthy GNNs encompassing robustness, explainability, privacy, fairness, accountability, and environmental well-being.
• A comprehensive, fine-grained taxonomy and metrics for evaluating each trust aspect in GNNs.
• Cross-aspect analysis highlighting how methods in one aspect influence others and where trade-offs occur.
• A comparison showing how trustworthy GNN surveys differ from generic trustworthy AI surveys and from performance-focused GNN surveys.
• An outlook on trending directions to promote research progress and industrial deployment of trustworthy GNNs.