6G-LEADER

5G networks face several challenges in providing Wireless for AI services, including: 1) limitations of 5G in supporting real-time AI-driven interactions; 2) increasing demand for intelligent and context-aware applications across verticals with heterogeneous latency and reliability requirements; 3) complexity in processing and interpreting heterogeneous data streams in dynamic environments; 4) need for real-time resource allocation and optimisation to meet QoS requirements; and 5) high control plane overheads of 5G/6G networks, degrading network performance. Moreover, current edge AI systems are not optimised, in terms of the number of end-user devices and their allocated computing power to locally train ML models.

In this PoC, the feasibility and potential of Wireless for AI services through
semantically-aware dApps/xApps will be demonstrated. By leveraging semantics, the “Wireless for AI” PoC will evaluate the performance of different (near) real-time resource and task allocation strategies and control plane overheads reduction for enhanced in-network AI applications. Also, the maximum number of end-user devices will be determined to satisfy the maximum acceptable Mean-Squared Error (MSE) for the aggregated model and the energy and time overheads at the end-user device side resulting from both communication and computing.