Edge-Aware Federated Learning: A Scalable and Fault-Tolerant System Architecture
Bhanuprakash Madupati, Ramesh Somayajula, Rohith Varma Vegesna, Ugandhar Dasi · 2025
Federated Learning (FL) enables decentralized machine learning by allowing edge devices to collaboratively train models without sharing raw data. This paradigm is particularly important in scenarios involving sensitive data and privacy concerns, such as healthcare, smart cities, and industrial IoT systems. However, the deployment of FL in edge environments is far from trivial due to critical challenges such as intermittent device connectivity, heterogeneous computational resources, non-independent and identically distributed (non-IID) data, and vulnerability to adversarial attacks. These constraints can severely impact the stability, scalability, and fault tolerance of FL systems, making robust solutions essential.To address these limitations, this paper proposes a scalable fault-tolerant framework for FL on the edge by combining robust aggregation, blockchain-enhanced trust, and federated continual learning (FCL). We evaluate the framework under simulated edge conditions and demonstrate improved reliability and model convergence.