Federated Graph Analytics with Differential Privacy

TL;DR

This work formulates federated graph analytics (FGA) under differential privacy, addressing utility loss due to limited local views and privacy leakage from overlapping subgraphs. It introduces FEAT, a general framework that privately aggregates subgraphs via a differentially private set union (DPSU) to produce a noisy global graph for queries, and FEAT+, which leverages true local subgraphs through degree-based partitioning to further boost accuracy. The authors provide concrete mechanisms for edge DP, including DPSU with ECC-based cryptography, and derive unbiased post-processing methods for k-star and triangle counting. Experimental results on real-world networks show FEAT reducing error by up to 4x over baselines and FEAT+ delivering at least a 10x improvement over FEAT, with trade-offs in running time and communication overhead. The work offers a scalable, privacy-preserving pathway for cross-institution graph analytics with broad practical impact in finance, social networks, and epidemiology.

Abstract

Collaborative graph analysis across multiple institutions is becoming increasingly popular. Realistic examples include social network analysis across various social platforms, financial transaction analysis across multiple banks, and analyzing the transmission of infectious diseases across multiple hospitals. We define the federated graph analytics, a new problem for collaborative graph analytics under differential privacy. Although differentially private graph analysis has been widely studied, it fails to achieve a good tradeoff between utility and privacy in federated scenarios, due to the limited view of local clients and overlapping information across multiple subgraphs. Motivated by this, we first propose a federated graph analytic framework, named FEAT, which enables arbitrary downstream common graph statistics while preserving individual privacy. Furthermore, we introduce an optimized framework based on our proposed degree-based partition algorithm, called FEAT+, which improves the overall utility by leveraging the true local subgraphs. Finally, extensive experiments demonstrate that our FEAT and FEAT+ significantly outperform the baseline approach by approximately one and four orders of magnitude, respectively.

Figures (12)

• Figure 1: Comparisons among central, local and federated scenarios. (a) In a central scenario jian2021publishingding2021differentiallyday2016publishingprivGraph, one trusted server owns the entire graph. (b) In a local scenario imola2021locallyye2020lfimola2022communicationliu2022collecting, each client owns one node and its 1-hop path information. (c) In a federated scenario, each client owns a subgraph that consists of multiple nodes and edges among them.
• Figure 2: Overview of $\mathsf{FEAT}$.
• Figure 3: Motivation of $\mathsf{FEAT}$+.
• Figure 4: The MSE in 2-star counting.
• Figure 5: The MRE in 2-star counting.

Theorems & Definitions (15)

• Definition 1: Differential Privacy dwork2014algorithmic
• Definition 2: Global Sensitivity dwork2014algorithmic
• Definition 3: Laplace Mechanism dwork2006calibrating
• Definition 4: Edge LDP qin2017generating
• Definition 5: Neighboring Graphs
• Definition 6: Edge Distributed Differential Privacy (Edge DDP)
• Theorem 1
• Theorem 2
• Proposition 1
• Proposition 2