Full-Decentralized Federated Learning-Based Edge Computing Peer Offloading Towards Industry 5.0

This paper gives a generic architecture and system modeling for full-decentralized on-demand edge computing-based peer offloading. Compared with the conventional centralized peer offloading strategies, the proposed full-decentralized peer offloading, based on federated learning, physically decentralizes peer offloading algorithm into edge computing, fully eliminating the rely on centralized servers (e.g., cloud). Meanwhile, compared with the other previous decentralized offloading schemes (blockchain-based, game theory-based, etc.), edge computing servers in this paper does not require global information to be shared, when they reach consensus of optimal peer offloading. In particular, the adjacent edge computing servers only share property-sensitive data (for the service providers of the edge computing servers) among each other, relying on which the whole edge computing network can reach global optimal peer offloading. In this paper, we consider energy efficiency as a use case to analyze the feasibility and efficiency of the proposed full-decentralized peer offloading architecture.