Real-Time Privacy Leak Detection in Decentralized Graph Analytics

While Wang et al.'s PrivGED (2022) enables privacy-preserving eigendecomposition on decentralized social graphs, it assumes static privacy guarantees and doesn't account for runtime vulnerabilities. This research proposes a real-time monitoring layer that continuously analyzes intermediate computations (e.g., gradient updates in federated eigendecomposition) for emergent privacy leaks. Using lightweight anomaly detection models trained on synthetic privacy-leak patterns, the system would dynamically adjust differential privacy noise levels when deviations occur. Unlike Jog et al.'s (2025) adaptive framework for federated intrusion detection—which focuses on network threats—this targets analytical privacy breaches. The novelty lies in treating privacy leaks as "adversarial patterns" detectable via ML, bridging the gap between static privacy designs and runtime threats.

References:

1. Privacy-Preserving Analytics on Decentralized Social Graphs: The Case of Eigendecomposition. Songlei Wang, Yifeng Zheng, Xiaohua Jia, X. Yi (2022). IEEE Transactions on Knowledge and Data Engineering.
2. An adaptive framework for privacy-preserving analytics in federated intrusion detection. Shwetha Jog, Damodharan Palaniappan, M.A. Jabbar (2025). Decision Analytics Journal.