Federated Learning has emerged as a transformative paradigm which enables many devices or organizations work together to train an AI model without sharing their private data. This survey provides a comprehensive examination of Federated Learning, where it systematically reviews advances in architectures, aggregation algorithms, privacy-preserving techniques, and security mechanisms. The survey also explains how Federated Learning has developed from basic methods like Federated Averaging (FedAvg) to more advanced sophisticated frameworks like addressing data heterogeneity, communication efficiency and personalization that can handle problems such as Different types of data on different devices (statistical heterogeneity), Devices with different processing power and internet speed (system heterogeneity), High communication costs and the need to create customized models for different users. Emerging paradigms including decentralized Federated Learning, clustered Federated Learning, personalized Federated Learning, and federated prompt learning are explored. Finally, the survey explores the remaining challenges and future possibilities. It highlights the need for Federated Learning systems that are more adaptive, efficient, secure, and trustworthy for real-world applications such as healthcare, IoT devices, edge computing, and smart systems.

Shreya Regish, Riya K Prakash, Dr. L.C Manikandan, "Federated Learning: A Survey of Recent Advances, Challenges and Applications" International Journal of Scientific Research and Engineering Development, V9(5): Page(468-477) September - October 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.