Title: Energy-Optimal Sampling for Edge Computing Feedback Systems: Aperiodic Case

Abstract: We study the problem of optimal sampling in an edge-based video analytics system (VAS), where sensor samples collected at a terminal device are offloaded to a back-end server that processes them and generates feedback for a user. Sampling the system with the maximum allowed frequency results in the timely detection of relevant events with minimum delay. However, it incurs high energy costs and causes unnecessary usage of network and compute resources via communication and processing of redundant samples. On the other hand, an infrequent sampling result in a higher delay in detecting the relevant event, thus increasing the idle energy usage and degrading the quality of experience in terms of responsiveness of the system. We quantify this sampling frequency trade-off as a weighted function between the number of samples and the responsiveness. We propose an energy-optimal aperiodic sampling policy that improves over the state-of-the-art optimal periodic sampling policy. Numerically, we show the proposed policy provides a consistent improvement of more than 10$\mathbf{\%}$ over the state-of-the-art.