Resilient and Context-Aware Edge AI: Bridging Intelligence, Privacy, and Autonomy

Abstract

Federated learning (FL) is an emerging distributed machine learning paradigm that enables collaborative model training across decentralized data sources while preserving data privacy and reducing communication overhead. This survey presents a comprehensive review of the current state of federated learning from both algorithmic and system perspectives. It begins by examining the core architectures of FL, including client-server and peer-to-peer designs, followed by detailed discussions of aggregation mechanisms, optimization strategies under data heterogeneity, and essential privacy-preserving techniques such as differential privacy and secure aggregation. Real-world applications are analyzed across industries such as healthcare, finance, Internet of Things, and education, demonstrating FL’s practical viability. The paper also outlines major open research challenges, including personalization, scalability, communication efficiency, and regulatory constraints. By unifying advances from both academic and industry settings, this work provides a foundational resource for researchers and practitioners aiming to deploy federated learning in production-grade AI systems.