Privacy preserving hierarchical graph federated learning for intelligent task scheduling in edge cloud networks

The increasing reliance on latency-sensitive and privacy-critical applications in modern computing has highlighted the limitations of conventional task offloading methods in hierarchical Local-Edge-Cloud (LEC) environments. Key challenges include managing dynamic resource availability, minimizing execution latency, and ensuring strong data privacy without centralized data sharing. To address these issues, this research proposes a novel framework named Multi-agent Graph-based Reinforcement Federated learning with Improved Builder Optimization Algorithm (MGRF-IBOA). The framework starts by defining task offloading as a constrained optimization problem that reduces execution time, increases energy efficiency, and ensures privacy. MGRF-IBOA combines several intelligent mechanisms, including graph-based task-resource mapping for spatial allocation, multi-agent reinforcement learning for adaptive scheduling, federated learning for decentralized model updates, and heterogeneous differential privacy to meet varying user privacy needs. The optimization process is further enhanced using the Improved Builder Optimization Algorithm (IBOA) for faster and more accurate convergence. Experimental results demonstrate that MGRF-HDP is superior to previous benchmark solutions, reaching a Dice Similarity Coefficient of 98.21%, a Jaccard index of 93.41%, a sensitivity rate of 98.50% and a specificity rate of 96.10%, confirming MGRF-HDP as a practical means of scaling, securing and managing latency in a distributed intelligent system.