Mitigating Privacy-Utility Trade-off in Decentralized Federated Learning via $f$-Differential Privacy

2025-10-22

AI Summary

This work addresses the inaccurate privacy budget accounting and the difficulty of balancing privacy protection with model utility in decentralized federated learning (DFL). To this end, we propose the first $f$-differential privacy ($f$-DP) accounting framework tailored to DFL. Methodologically, we introduce two novel mechanisms—Pairwise Network $f$-DP and Secret-based $f$-Local DP—that jointly incorporate random-walk communication, shared-key-driven noise scheduling, sparse local iterations, and structured noise injection, enabling theoretically grounded privacy amplification and precise accounting. Theoretically, by leveraging Markov chain concentration and decentralized topology analysis, we derive tighter $(varepsilon,delta)$-DP bounds. Empirically, our approach significantly outperforms Rényi DP baselines on both synthetic and real-world datasets, improving model utility by 12.7%–23.4%, thereby establishing the first empirical validation of $f$-DP’s theoretical superiority and practical viability in decentralized FL.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Distributed Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Data transparency and provenance
• Systems and Infrastructure for Web, Mobile and WoT: Decentralized Web and Fediverse systems

Abstract

Differentially private (DP) decentralized Federated Learning (FL) allows local users to collaborate without sharing their data with a central server. However, accurately quantifying the privacy budget of private FL algorithms is challenging due to the co-existence of complex algorithmic components such as decentralized communication and local updates. This paper addresses privacy accounting for two decentralized FL algorithms within the $f$-differential privacy ($f$-DP) framework. We develop two new $f$-DP-based accounting methods tailored to decentralized settings: Pairwise Network $f$-DP (PN-$f$-DP), which quantifies privacy leakage between user pairs under random-walk communication, and Secret-based $f$-Local DP (Sec-$f$-LDP), which supports structured noise injection via shared secrets. By combining tools from $f$-DP theory and Markov chain concentration, our accounting framework captures privacy amplification arising from sparse communication, local iterations, and correlated noise. Experiments on synthetic and real datasets demonstrate that our methods yield consistently tighter $(ε,δ)$ bounds and improved utility compared to Rényi DP-based approaches, illustrating the benefits of $f$-DP in decentralized privacy accounting.

Problem

• Quantifying privacy budget in decentralized federated learning algorithms
• Developing f-DP accounting methods for pairwise and secret-based privacy
• Capturing privacy amplification from sparse communication and local iterations

Innovation

• Pairwise Network f-DP for user pair privacy
• Secret-based f-Local DP with shared secrets
• Combining f-DP theory with Markov chain concentration