Federated Learning Privacy: Comprehensive Analysis of Attacks, Defenses, Applications, and Policy Landscape

Federated learning (FL) has emerged as a privacy-preserving technique for collaborative machine learning, but recent studies have shown that the fundamental premise of privacy preservation does not always hold. This survey provides a comprehensive analysis of the different privacy attacks against FL, including data reconstruction, membership inference, and property inference attacks, as well as the various defense mechanisms proposed to mitigate these threats. It also examines the real-world applications of FL across industries and the evolving policy landscape governing data privacy, highlighting the need for robust privacy-preserving techniques to enable the widespread adoption of FL.