Privacy-preserving machine learning: a review of federated learning techniques and applications

Abstract

Federated Learning (FL), which permits decentralized model training without sharing raw data, guarantees ‎adherence to privacy laws like GDPR and HIPAA. This study offers a thorough analysis of FL with an ‎emphasis on its exceptional capacity to strike a balance between data value and privacy in industries ‎including healthcare, the Internet of Things, and finance. In contrast to previous evaluations, this study ‎explores sophisticated privacy-preserving techniques, such as differential privacy and homomorphic ‎encryption, and assesses how well they work to handle issues like adversarial threats, non-IID data ‎distributions, and communication overhead. The study also discusses the practical uses of optimization ‎techniques like Federated Proximal (FedProx) and Federated Averaging (FedAvg). This paper provides ‎practical insights and future approaches to promote the use of FL in privacy-sensitive AI applications by ‎comparing and contrasting current methods and pointing out research gaps. FL is positioned as a ‎revolutionary method for privacy-conscious machine learning because to this fresh viewpoint. This update highlights the paper's distinctive features that set it apart from prior reviews, including the ‎thorough examination of privacy mechanisms, assessment of optimization techniques, and identification ‎of research needs‎.