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

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.