Energy-Efficient Scheduling in RIS-Aided MEC Networks for Collaborative Inference

In this paper, we consider a reconfigurable intelligent surface (RIS) aided mobile edge computing (MEC) network and investigate the minimization of the energy consumption for collaborative inference. Within the collaborative inference framework, the user equipments (UEs) are allowed to offload parts of computation-intensive perception services to the MEC server and thereby reduce the energy consumption subject to their latency constraints. In this setting, considering the collaborative inference task and the transmission model, we aim to minimize the global energy consumption under the UEs' latency constraints. A three-step optimization algorithm is devised to address the considered energy minimization problem in which the RIS phase shift matrix, inference partition decisions and CPU frequency allocations at the MEC server are optimally determined. Simulation results on the collaborative inference task demonstrate the effectiveness of our proposed approach in reducing the energy consumption levels while satisfying the UEs' latency constraints.