BinarizedAttack: Structural Poisoning Attacks to Graph-based Anomaly Detection

Zhu, Yulin; Lai, Yuni; Zhao, Kun; Luo, Xiapu; Yuan, Mingquan; Ren, Jing; Zhou, Kai

doi:10.1109/icde53745.2022.00006

Cited by 13 publications

(5 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the gray-box scenario, attackers leverage gradient-based strategies on the adjacency matrix to identify critical modifications. Various approaches exist for selecting edge alterations based on gradients [25][26][27][28]. These adversarial techniques pose a challenge to the robustness of graph-based anomaly detection systems, as they exploit vulnerabilities within the underlying graph structures and classification models.…”

Section: Adversarial Attacks On Graphmentioning

confidence: 99%

See 1 more Smart Citation

Unleashing the Power of Indirect Attacks against Trust Prediction via Preferential Path

Bu,

Zhu,

Geng

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

In the realm of network security, adversarial attacks pose a significant threat, prompting research into innovative strategies for enhancing both attack and defense mechanisms. This paper presents a novel exploration into the enhancement of adversarial attacks on Fairness and Goodness Algorithm (FGA) and Review to Reviewer (REV2), focusing on trust prediction within signed graphs. In contrast to traditional time-based dynamics, the trust propagation in FGA and REV2 is grounded on iterative processes. Through meticulous analysis of network structures, this research uncovers the strong ties and weak ties within FGA. Additionally , it reveals the existence of preferential paths in REV2, which significantly influence the spread of information during algorithm iterations. Leveraging these revelations, we introduce a novel approach known as vicinage-attack designed to bolster adversarial attacks by strategically targeting edges along these influential pathways. The significance of this work lies in identifying adversarial perturbation patterns that impact trust prediction on signed graphs, shedding light on their pervasive influence. Our findings not only contribute to the advancement of adver-sarial attack techniques but also lay the foundation for a deeper understanding of trust propagation patterns. By elucidating the propagation bias within FGA and REV2, this research offers new insights for both network security strategies and adversarial mitigation techniques in trust prediction.

show abstract

Section: Adversarial Attacks On Graphmentioning

confidence: 99%

“…A conventional technique employed for the optimization problem is gradient descent, as thoroughly discussed in previous works [24][25][26][27][28]. The initialization of B commences with a value of 0.5 and is refined through the update mechanism:…”

Section: Projection and Distribution Of Gradientmentioning

confidence: 99%

Unleashing the Power of Indirect Attacks against Trust Prediction via Preferential Path

Bu,

Zhu,

Geng

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Extensive studies have demonstrated that GNNs are highly fragile to adversarial attacks [7,40,51,54,55]. The attackers can greatly degrade the performance of GNNs by limitedly modifying the graph data, namely structure and features.…”

Section: Robust Gnnsmentioning

confidence: 99%

“…Recent studies have shown that GNNs are vulnerable to adversarial attacks [7,40,51,54,55]. In other words, by limitedly rewiring the graph structure or just perturbing a small part of the node features, attackers can easily fool the GNNs to misclassify nodes in the graph.…”

Section: Introductionmentioning

confidence: 99%

Reliable Representations Make A Stronger Defender: Unsupervised Structure Refinement for Robust GNN

Li,

Liu,

et al. 2022

Preprint

View full text Add to dashboard Cite

Benefiting from the message passing mechanism, Graph Neural Networks (GNNs) have been successful on flourish tasks over graph data. However, recent studies have shown that attackers can catastrophically degrade the performance of GNNs by maliciously modifying the graph structure. A straightforward solution to remedy this issue is to model the edge weights by learning a metric function between pairwise representations of two end nodes, which attempts to assign low weights to adversarial edges. The existing methods use either raw features or representations learned by supervised GNNs to model the edge weights. However, both strategies are faced with some immediate problems: raw features cannot represent various properties of nodes (e.g., structure information), and representations learned by supervised GNN may suffer from the poor performance of the classifier on the poisoned graph. We need representations that carry both feature information and as mush correct structure information as possible and are insensitive to structural perturbations. To this end, we propose an unsupervised pipeline, named STABLE, to optimize the graph structure. Finally, we input the well-refined graph into a downstream classifier. For this part, we design an advanced GCN that significantly enhances the robustness of vanilla GCN [24] without increasing the time complexity. Extensive experiments on four real-world graph benchmarks demonstrate that STABLE outperforms the state-of-the-art methods and successfully defends against various attacks.

show abstract

“…However, most studies use one-hot vectors to encode a node's features and rely on a GNN to capture the global features of network snapshots. 2) Taking advantage of its message passing mechanism, GNNs have been developed to deal with various tasks over graph data [20], [21]. However, recent studies have shown that GNNs are vulnerable to adversarial attacks in delivering messages, achieved through malicious modification of their graph structure.…”

Section: Introductionmentioning

confidence: 99%

SAGE-GCN: Graph Convolutional Network Based on Self-adaptive Stable Gates for Link Prediction in Dynamic Complex Networks

Yang¹,

Liu²,

Sun³

et al. 2023

Preprint

View full text Add to dashboard Cite

<p>Link prediction is one of the most important tasks in uncovering evolving mechanisms of dynamic complex networks. Existing dynamic link prediction models suffer from limitations such as vulnerability to adversarial attacks, poor accuracy, and instability. In this paper, we propose a novel dynamic Graph Convolutional Network model incorporating a Self-adaptive Stable Gate (SAGE-GCN) consisting of a state encoding network and a policy network. Firstly, we capture the local topology of the nodes by employing a multi-power adjacency matrix to obtain higher-order topological features, enabling its features to be distinguished at different network snapshots. Then, a stable gate is introduced to ensure multiple spatiotemporal dependency paths within the state encoding network. It is proven that SAGE-GCN with integral Lipschitz graph convolution is stable to relative perturbations in the dynamic networks. Finally, a self-adaptive strategy is proposed to choose different state encoding network instances, with a policy network used to learn the optimal temporal and structural features through corresponding rewards to capture network dynamics. With the aid of extensive experiments on five real-world graph benchmarks, SAGE-GCN is shown to substantially outperform current state-of-the-art approaches in terms of precision and stability of dynamic link prediction and ability to successfully defend against various attacks.</p>

show abstract

BinarizedAttack: Structural Poisoning Attacks to Graph-based Anomaly Detection

Cited by 13 publications

References 28 publications

Unleashing the Power of Indirect Attacks against Trust Prediction via Preferential Path

Unleashing the Power of Indirect Attacks against Trust Prediction via Preferential Path

Reliable Representations Make A Stronger Defender: Unsupervised Structure Refinement for Robust GNN

SAGE-GCN: Graph Convolutional Network Based on Self-adaptive Stable Gates for Link Prediction in Dynamic Complex Networks

Contact Info

Product

Resources

About