Energy-Aware Secure Hierarchical Federated Learning for 5G/6G Edge Networks

In this paper we present the Energy-Aware Secure Hierarchical Federated Learning (EA-SHFL) framework designed to face the main challenges of energy efficiency, security and model accuracy in5G/6Gedge networks. The sudden surge in the number of IoT devices and edge computation could increase the demand for federated learning approaches that aim to reduce energy consumption on client devices and guarantee security against attacks, including poisoning and data integrity violations. The proposed system employs a three-tiers of hierarchy aim to route from the IoT devices through the edge servers accessing to the cloud server in order to enhance resource allocation and scalability. Here, energy-aware device selection at the client allocates priority to those devices with adequate battery life and processing power to extend the operational time of the network. Moreover, model update compression via quantization-aware training minimizes the communication overhead while maintaining performance. Security is enforced through a two-tier aggregation where at the edge, integrity checks based on blockchain technology and anomaly detection will filter out malicious updates; while at the cloud level, dynamically adjusted trust scores will be used to solve a convex optimization problem for robust aggregation of edge models. The experimental results show that EA-SHFL has greater energy efficiency and is less prone to adversarial attacks while maintaining high model accuracy when compared to existing techniques. This framework provides a feasible approach for the implementation of federated learning in resource-constrained edge environments with high security risks, thereby facilitating the pathway towards the sustainable and trusted AI development in the future networks.