Abstract
Federated Learning (FL) has emerged as a transformative paradigm in the fieldof distributed machine learning, enabling multiple clients such as mobiledevices, edge nodes, or organizations to collaboratively train a shared globalmodel without the need to centralize sensitive data. This decentralizedapproach addresses growing concerns around data privacy, security, andregulatory compliance, making it particularly attractive in domains such ashealthcare, finance, and smart IoT systems. This survey provides a concise yetcomprehensive overview of Federated Learning, beginning with its corearchitecture and communication protocol. We discuss the standard FL lifecycle,including local training, model aggregation, and global updates. A particularemphasis is placed on key technical challenges such as handling non-IID(non-independent and identically distributed) data, mitigating system andhardware heterogeneity, reducing communication overhead, and ensuring privacythrough mechanisms like differential privacy and secure aggregation.Furthermore, we examine emerging trends in FL research, including personalizedFL, cross-device versus cross-silo settings, and integration with otherparadigms such as reinforcement learning and quantum computing. We alsohighlight real-world applications and summarize benchmark datasets andevaluation metrics commonly used in FL research. Finally, we outline openresearch problems and future directions to guide the development of scalable,efficient, and trustworthy FL systems.