Breaking Free from Label Limitations: A Novel Unsupervised Attack Method for Graph Classification

TY - GEN

T1 - Breaking Free from Label Limitations

T2 - 30th International Conference on Database Systems for Advanced Applications, DASFAA 2025

AU - Wang, Yadong

AU - Zhang, Zhiwei

AU - Qiao, Pengpeng

AU - Yuan, Ye

AU - Zhang, Hao

AU - wang, Guoren

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.

PY - 2026

Y1 - 2026

N2 - Graph Neural Networks (GNNs) have proven highly effective for graph classification across diverse fields. However, despite their success, GNNs remain vulnerable to adversarial attacks that can significantly degrade their classification accuracy. Existing adversarial attack strategies mainly focus on supervised approaches, limiting their applicability in scenarios where the label information is scarce or unavailable. This paper introduces an innovative unsupervised attack method for graph classification that operates without relying on label information. Specifically, our method first leverages a graph contrastive learning loss to learn robust graph embeddings by comparing different stochastic augmented views of the graphs. To effectively perturb the graphs, we introduce an implicit estimator and flip edges with the top-k highest scores, determined by the estimator, to maximize the degradation of the model’s performance. Experiments on multiple datasets show the effectiveness of our proposed unsupervised attack strategy in degrading the performance of various graph classification models.

AB - Graph Neural Networks (GNNs) have proven highly effective for graph classification across diverse fields. However, despite their success, GNNs remain vulnerable to adversarial attacks that can significantly degrade their classification accuracy. Existing adversarial attack strategies mainly focus on supervised approaches, limiting their applicability in scenarios where the label information is scarce or unavailable. This paper introduces an innovative unsupervised attack method for graph classification that operates without relying on label information. Specifically, our method first leverages a graph contrastive learning loss to learn robust graph embeddings by comparing different stochastic augmented views of the graphs. To effectively perturb the graphs, we introduce an implicit estimator and flip edges with the top-k highest scores, determined by the estimator, to maximize the degradation of the model’s performance. Experiments on multiple datasets show the effectiveness of our proposed unsupervised attack strategy in degrading the performance of various graph classification models.

KW - Adversarial Attack

KW - GNNs

KW - Graph Classification

UR - https://www.scopus.com/pages/publications/105027112369

U2 - 10.1007/978-981-95-4155-3_40

DO - 10.1007/978-981-95-4155-3_40

M3 - Conference contribution

AN - SCOPUS:105027112369

SN - 9789819541546

T3 - Lecture Notes in Computer Science

SP - 588

EP - 598

BT - Database Systems for Advanced Applications - 30th International Conference, DASFAA 2025, Proceedings

A2 - Zhu, Feida

A2 - Lim, Ee-peng

A2 - Yu, Philip S.

A2 - Nadamoto, Akiyo

A2 - Shim, Kyuseok

A2 - Ding, Wei

A2 - Zhang, Bingxue

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 26 May 2025 through 29 May 2025

ER -