Decentralized Edge Intelligence-Driven Network Resource Orchestration Mechanism

With the development of artificial intelligence of things (AIoT), multi-access edge computing (MEC) becomes a key enabler to migrate cloud services to edge clients. In comparison to traditional cloud computing techniques, MEC is characterized with low transmission latency, good flexibility, adaptability and robustness. Nevertheless, traditional resource allocation methods are difficult to meet the requirements of achieving a ubiquitous, pervasive, and intelligent computation offloading strategy in high-dynamic network environments. In this article, we construct an edge intelligence-enabled cloud-edge-client collaborative network structure, and conceive a model-aided multi-agent deep deterministic policy gradient (MA2DDPG) computation offloading framework relying on both centralized training and distributed execution. Simulation results corroborate that our proposed decentralized resource orchestration platform significantly reduces the energy consumption and the transmission latency against state-of-the-art methods. Finally, we highlight open challenges and potential solutions.