Dropout-Robust Mechanisms for Differentially Private and Fully Decentralized Mean Estimation

2025-06-04

AI Summary

To address the high communication overhead and sharp accuracy degradation caused by dynamic node departures in decentralized differential privacy (DP) mean estimation, this paper proposes IncA—a fully decentralized protocol requiring no central coordinator and supporting arbitrary node dropouts. Its core innovation is an incremental correlated noise injection mechanism: nodes locally estimate sensitivity and generate low-variance, statistically correlated noise to jointly preserve privacy and estimation accuracy. Theoretically, IncA achieves mean estimation error convergence at the same rate as the centralized optimal bound—even without assuming permanent node failures. Empirically, under 30% random node dropout, IncA reduces mean squared error by over 42% compared to state-of-the-art decentralized DP methods, demonstrating significantly improved robustness and communication efficiency.

Technology Category

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

Application Category

• Security and Privacy: Data transparency and provenance
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Achieving differentially private computations in decentralized settings poses significant challenges, particularly regarding accuracy, communication cost, and robustness against information leakage. While cryptographic solutions offer promise, they often suffer from high communication overhead or require centralization in the presence of network failures. Conversely, existing fully decentralized approaches typically rely on relaxed adversarial models or pairwise noise cancellation, the latter suffering from substantial accuracy degradation if parties unexpectedly disconnect. In this work, we propose IncA, a new protocol for fully decentralized mean estimation, a widely used primitive in data-intensive processing. Our protocol, which enforces differential privacy, requires no central orchestration and employs low-variance correlated noise, achieved by incrementally injecting sensitive information into the computation. First, we theoretically demonstrate that, when no parties permanently disconnect, our protocol achieves accuracy comparable to that of a centralized setting-already an improvement over most existing decentralized differentially private techniques. Second, we empirically show that our use of low-variance correlated noise significantly mitigates the accuracy loss experienced by existing techniques in the presence of dropouts.

Problem

• Achieving accurate decentralized differentially private mean estimation
• Reducing communication cost in private decentralized computations
• Enhancing robustness against dropouts in decentralized settings

Innovation

• Fully decentralized mean estimation protocol
• Low-variance correlated noise technique
• Incremental sensitive information injection