Resource Management for Multiuser Opportunistic Edge-Inference by Conformal Lyapunov Optimization

We consider computational and transmission resource management in a wireless networked system where multiple users may opportunistically offload their inference tasks to an edge server, jointly trading inference accuracy with energy consumptions and latency. To this end, we resort to Conformal Lyapunov Optimization (CLO), which has recently been introduced as a resource optimization framework capable of combining classical Lyapunov Optimization with conformal risk control. The effectiveness of the proposed approach is verified by simulation results, which confirm the capability of CLO to handle average long-term constraints, together with deterministic reliability guarantees, when multiple users operate under varying computational and channel conditions.