Multi-Dimensional Resource Orchestration Toward Edge Intelligence in 6G Networks

The sixth generation (6G) networks will support a range of intelligent applications such as face recognition and next-word prediction. However, relying on remote central clouds to implement these applications cannot meet the high-quality service (QoS) requirements of 6G users. Multi-access edge computing (MEC) is a powerful approa ch to providing services by leveraging the multi-dimensional resources available at the network edge, including heterogeneous network infrastructures with different computation and communication capacities. MEC makes it promising to achieve Edge Intelligence (EI). This article introduces multi-dimensional resource orchestration toward EI in 6G networks. Specifically, we highlight the key features of 6G from the perspectives of infrastructure, services, and users. Based on these features, we summarize potential technologies that can enable EI over multi-dimensional resources. These include scheduling of mobility-assisted edge networks, Zero-touch Network and Service Management (ZSM), and collaboration among distributed users with private data. To address the most general graph-structured services, we investigate a multi-dimensional resource orchestration method based on Node-Link Policy Network Reinforcement Learning (NL-PNRL). By exploring online resource allocation for services, the proposed method can improve the revenue-cost ratio of the infrastructure provider. Finally, we analyze open challenges for future research in this area.