Demystifying Uneven Vulnerability of Link Stealing Attacks against Graph Neural Networks
Authors: He Zhang, Bang Wu, Shuo Wang, Xiangwen Yang, Minhui Xue, Shirui Pan, Xingliang Yuan
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We first present theoretical evidence of the uneven vulnerability of GNNs to link stealing attacks, which lays the foundation for demystifying such uneven risks among different groups of edges. We further demonstrate a group-based attack paradigm to expose the practical privacy harm to GNN users derived from the uneven vulnerability of edges. Finally, we empirically validate the existence of obvious uneven vulnerability on ten real-world datasets |
| Researcher Affiliation | Collaboration | 1Department of Software Systems and Cybersecurity, Faculty of Information Technology, Monash University, Australia 2CSIRO s Data61, Australia 3School of Information and Communication Technology, Griffith University, Australia. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code or links to a code repository for the described methodology. |
| Open Datasets | Yes | Our evaluations employ ten real-world datasets: Cora (Kipf & Welling, 2017), Citeseer (Kipf & Welling, 2017), Pubmed (Kipf & Welling, 2017), COX2 (Sutherland et al., 2003), DHFR (Sutherland et al., 2003), Enzymes (Dobson & Doig, 2003), Proteins full (Borgwardt et al., 2005), Credit defaulter graph (Yeh & Lien, 2009), German credit graph (Dua et al., 2017) and Ogbn-Arxiv (Hu et al., 2020). |
| Dataset Splits | No | The paper mentions using a 'training graph' for GNNs but does not provide specific details on training, validation, or test dataset splits (e.g., percentages, sample counts, or references to standard splits). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, specific library versions) required to reproduce the experiments. |
| Experiment Setup | Yes | The GCNs have 2 hidden layers with 16 units and employ Re LU and softmax as activation functions; the GATs have 2 hidden layers (16 units) with 1 head of attentions and use ELU and softmax as activation functions; the Graph SAGE models have 1 hidden layer (16 units, the Relu activation function) and use 1 MLP layer as the classifier. Following previous attacks (He et al., 2021a), we use cosine, euclidean, correlation, chebyshev, braycurtis, canberra, cityblock and sqeuclidean distance to measure the similarity of two nodes posteriors. |