Enhancing Model Privacy in Federated Learning with Random Masking and Quantization

2025-08-26

AI Summary

To address the vulnerability of model parameters to privacy inference attacks in federated learning, this paper proposes a synergistic privacy-preserving mechanism integrating randomized masking with gradient quantization. Specifically, clients apply controllable random masking locally prior to low-bit gradient quantization, enhancing parameter irreversibility while achieving communication compression. This is further augmented by differential privacy noise injection and secure aggregation, establishing a multi-tiered privacy guarantee. Compared to single-technique baselines, the method achieves significantly improved adversarial robustness under strict privacy budgets (ε ≤ 4) without compromising convergence stability. Extensive experiments across image classification and text prediction tasks demonstrate an accuracy drop of less than 1.2%, alongside approximately 40% reduction in communication overhead. These results validate the method’s effective balance among privacy preservation, model utility, and system efficiency.

Technology Category

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

Application Category

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

Abstract

Experimental results across various models and tasks demonstrate that our approach not only maintains strong model performance in federated learning settings but also achieves enhanced protection of model parameters compared to baseline methods.

Problem

• Enhancing model privacy in federated learning
• Protecting model parameters from unauthorized access
• Maintaining performance while improving security

Innovation

• Random masking for privacy enhancement
• Quantization to protect model parameters
• Maintains performance in federated learning