当前位置:
X-MOL 学术
›
arXiv.cs.LG
›
论文详情
Our official English website, www.x-mol.net, welcomes your
feedback! (Note: you will need to create a separate account there.)
Uncertainty-aware Attention Graph Neural Network for Defending Adversarial Attacks
arXiv - CS - Machine Learning Pub Date : 2020-09-22 , DOI: arxiv-2009.10235 Boyuan Feng, Yuke Wang, Zheng Wang, and Yufei Ding
arXiv - CS - Machine Learning Pub Date : 2020-09-22 , DOI: arxiv-2009.10235 Boyuan Feng, Yuke Wang, Zheng Wang, and Yufei Ding
With the increasing popularity of graph-based learning, graph neural networks
(GNNs) emerge as the essential tool for gaining insights from graphs. However,
unlike the conventional CNNs that have been extensively explored and
exhaustively tested, people are still worrying about the GNNs' robustness under
the critical settings, such as financial services. The main reason is that
existing GNNs usually serve as a black-box in predicting and do not provide the
uncertainty on the predictions. On the other side, the recent advancement of
Bayesian deep learning on CNNs has demonstrated its success of quantifying and
explaining such uncertainties to fortify CNN models. Motivated by these
observations, we propose UAG, the first systematic solution to defend
adversarial attacks on GNNs through identifying and exploiting hierarchical
uncertainties in GNNs. UAG develops a Bayesian Uncertainty Technique (BUT) to
explicitly capture uncertainties in GNNs and further employs an
Uncertainty-aware Attention Technique (UAT) to defend adversarial attacks on
GNNs. Intensive experiments show that our proposed defense approach outperforms
the state-of-the-art solutions by a significant margin.
中文翻译:
用于防御对抗性攻击的不确定性注意图神经网络
随着基于图的学习越来越流行,图神经网络 (GNN) 成为从图获得洞察力的重要工具。然而,与经过广泛探索和详尽测试的传统 CNN 不同,人们仍然担心 GNN 在金融服务等关键环境下的鲁棒性。主要原因是现有的 GNN 通常在预测中充当黑盒,并且不提供预测的不确定性。另一方面,贝叶斯深度学习在 CNN 上的最新进展证明了它成功地量化和解释了这些不确定性以强化 CNN 模型。受这些观察的启发,我们建议 UAG,通过识别和利用 GNN 中的层次不确定性来防御对 GNN 的对抗性攻击的第一个系统解决方案。UAG 开发了贝叶斯不确定性技术 (BUT) 来明确捕获 GNN 中的不确定性,并进一步采用不确定性感知注意力技术 (UAT) 来防御对 GNN 的对抗性攻击。大量实验表明,我们提出的防御方法明显优于最先进的解决方案。
更新日期:2020-09-23
中文翻译:
用于防御对抗性攻击的不确定性注意图神经网络
随着基于图的学习越来越流行,图神经网络 (GNN) 成为从图获得洞察力的重要工具。然而,与经过广泛探索和详尽测试的传统 CNN 不同,人们仍然担心 GNN 在金融服务等关键环境下的鲁棒性。主要原因是现有的 GNN 通常在预测中充当黑盒,并且不提供预测的不确定性。另一方面,贝叶斯深度学习在 CNN 上的最新进展证明了它成功地量化和解释了这些不确定性以强化 CNN 模型。受这些观察的启发,我们建议 UAG,通过识别和利用 GNN 中的层次不确定性来防御对 GNN 的对抗性攻击的第一个系统解决方案。UAG 开发了贝叶斯不确定性技术 (BUT) 来明确捕获 GNN 中的不确定性,并进一步采用不确定性感知注意力技术 (UAT) 来防御对 GNN 的对抗性攻击。大量实验表明,我们提出的防御方法明显优于最先进的解决方案。