Skip to content
Conference

Privacy-Preserving Techniques in Federated Learning: A Comprehensive Survey and Critical Review

Oct 2026 · 2026 6th International Conference on Ubiquitous Computing and Intelligent Information Systems (ICUIS) · pp. 282-286 · 0 citations · 26 references

Abstract

Federated Learning (FL) is a distributed machine learning paradigm in which multiple clients collaboratively train a model while their raw data remains local. Despite this design, FL is exposed to privacy risks such as inference attacks, model inversion, membership inference, and gradient leakage, which threaten sensitive user data. Privacy-preserving techniques have therefore been studied extensively. This paper reviews four main families: differential privacy (DP), secure multi-party computation (SMPC), homomorphic encryption (HE), and trusted execution environments (TEE). DP adds controlled noise to model updates so that individual records cannot be identified, but excessive noise degrades model performance. SMPC lets several parties jointly compute a function over their inputs without revealing them, but its resource demands make large-scale FL deployments difficult. HE allows computation directly on encrypted data and offers strong confidentiality, but its computational cost makes real-time use impractical. TEEs rely on hardware-based isolation to protect sensitive computation from untrusted parts of the system, but they require dedicated hardware, which has limited adoption. We evaluate each technique in terms of privacy protection, effect on model performance, and communication overhead, and we outline emerging trends. Hybrid privacy-preserving approaches are identified as a promising research direction.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.