Towards Trustworthy Federated Learning with Untrusted Participants

2025-05-03

AI Summary

In federated learning without a trusted central server, achieving simultaneous privacy preservation, robust aggregation, and high model utility remains challenging. Method: This paper proposes CafCor, the first algorithm that unifies these objectives relying solely on pairwise private random seeds shared among users—without any trusted third party. CafCor integrates robust gradient aggregation, correlation-aware noise injection, and a pairwise stochastic coordination mechanism. Contribution/Results: Theoretically, CafCor achieves near-central differential privacy (DP) utility and collusion resistance, offering a provably strong privacy–utility trade-off—significantly outperforming local DP. Empirically, on standard benchmarks, CafCor attains high test accuracy, strong Byzantine robustness, and rigorous DP guarantees, with model utility approaching that of central DP.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: Safety and Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Resilience against malicious parties and data privacy are essential for trustworthy distributed learning, yet achieving both with good utility typically requires the strong assumption of a trusted central server. This paper shows that a significantly weaker assumption suffices: each pair of workers shares a randomness seed unknown to others. In a setting where malicious workers may collude with an untrusted server, we propose CafCor, an algorithm that integrates robust gradient aggregation with correlated noise injection, leveraging shared randomness between workers. We prove that CafCor achieves strong privacy-utility trade-offs, significantly outperforming local differential privacy (DP) methods, which do not make any trust assumption, while approaching central DP utility, where the server is fully trusted. Empirical results on standard benchmarks validate CafCor's practicality, showing that privacy and robustness can coexist in distributed systems without sacrificing utility or trusting the server.

Problem

• Achieving trustworthy federated learning with untrusted participants
• Balancing privacy and utility without a trusted central server
• Combining robust gradient aggregation with correlated noise injection

Innovation

• Uses shared randomness seeds between workers
• Integrates robust gradient aggregation with noise
• Achieves strong privacy-utility trade-offs without trust