A Federated Learning Framework for Multi-Parameter Optimization in Edge Computing
Rajesh Soma, Subham Kumar Sahoo, Fahad Amin, Sambit Kumar Mishra
Abstract
Federated learning (FL) enables distributed model training across edge devices without sharing raw data to preserve privacy and reduce communication overhead. However, FL can sometimes hinder edge settings in terms of energy efficiency and communication. The present research looks into an energy-aware device-to-device (D2D)-assisted FL paradigm in edge computing. With data quality, device reliability, and throughput as multiple parameters, we begin by implementing two baseline approaches from the original D2D-assisted model and optimizing them. We then present an algorithm for adaptive data sampling by modifying a heuristic algorithm that chooses participants dynamically based on energy and performance criteria, thereby enhancing training efficiency. The proposed approach achieves 95.13% accuracy in fewer rounds compared to 94.68 % (optimal) and 71.07% (heuristic), with 30% lower communication overhead, as shown by simulation results, verifying the effectiveness of multi-parameter optimization in FL for edge computing.