This website requires JavaScript.

EDoG: Adversarial Edge Detection For Graph Neural Networks

Xiaojun XuYue YuHanzhang WangAlok LalCarl A. GunterBo Li
Dec 2022
摘要
Graph Neural Networks (GNNs) have been widely applied to different tasks suchas bioinformatics, drug design, and social networks. However, recent studieshave shown that GNNs are vulnerable to adversarial attacks which aim to misleadthe node or subgraph classification prediction by adding subtle perturbations.Detecting these attacks is challenging due to the small magnitude ofperturbation and the discrete nature of graph data. In this paper, we propose ageneral adversarial edge detection pipeline EDoG without requiring knowledge ofthe attack strategies based on graph generation. Specifically, we propose anovel graph generation approach combined with link prediction to detectsuspicious adversarial edges. To effectively train the graph generative model,we sample several sub-graphs from the given graph data. We show that since thenumber of adversarial edges is usually low in practice, with low probabilitythe sampled sub-graphs will contain adversarial edges based on the union bound.In addition, considering the strong attacks which perturb a large number ofedges, we propose a set of novel features to perform outlier detection as thepreprocessing for our detection. Extensive experimental results on threereal-world graph datasets including a private transaction rule dataset from amajor company and two types of synthetic graphs with controlled properties showthat EDoG can achieve above 0.8 AUC against four state-of-the-art unseen attackstrategies without requiring any knowledge about the attack type; and around0.85 with knowledge of the attack type. EDoG significantly outperformstraditional malicious edge detection baselines. We also show that an adaptiveattack with full knowledge of our detection pipeline is difficult to bypass it.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?