PreprintCC BY 4.0arXiv
Empirical Auditing of Edge-Private Graph Generators
Abstract
We empirically audit privacy leakage by testing whether outputs from edge-neighbouring inputs remain distinguishable, using statistically valid lower bounds on the privacy loss witnessed by our attacks. Our framework compares direct-edge, local-structural, and GNN-based attacks through the geometry surrounding a target edge. Experiments across two generators and two networks show that privacy leakage is both mechanism- and network-dependent, with learned representations revealing information not captured by conventional local statistics.
Related
- Auditing Bayesian Graph Alignment: Diagnostic Comparisons and Reference Failure
- Weak Private Information Retrieval for Graph-based Storage under ε-privacy
- Polynomial-Time Differentially Private Synthetic Graph Release with Nearly-Linear Additive Error
- Local Private Information Retrieval for Graph-Based Replicated Systems
- Coupling and Maximal Inequalities for Graph-Dependent Empirical Processes
BibTeX
@misc{fatima2026empirical,
title = {{Empirical Auditing of Edge-Private Graph Generators}},
author = {Anum Fatima and Stratis Limnios and James Adams and Lukasz Szpruch and Carsten Maple and Gesine Reinert and Andrew Elliott},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.25155},
eprint = {2609.25155},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.25155},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Empirical Auditing of Edge-Private Graph Generators
AU - Anum Fatima
AU - Stratis Limnios
AU - James Adams
AU - Lukasz Szpruch
AU - Carsten Maple
AU - Gesine Reinert
AU - Andrew Elliott
PY - 2026
DA - 2026/09/21
PB - arXiv
DO - 10.48550/arxiv.2609.25155
UR - https://arxiv.org/abs/2609.25155
AB - We empirically audit privacy leakage by testing whether outputs from edge-neighbouring inputs remain distinguishable, using statistically valid lower bounds on the privacy loss witnessed by our attacks. Our framework compares direct-edge, local-structural, and GNN-based attacks through the geometry surrounding a target edge. Experiments across two generators and two networks show that privacy leakage is both mechanism- and network-dependent, with learned representations revealing information not captured by conventional local statistics.
ER -