Selective Edge Computing for Mobile Analytics

An increasing number of mobile applications rely on Machine Learning (ML) routines for analyzing data. Executing such tasks at the user devices saves energy spent transmitting and processing large data volumes distant cloud-deployed servers. However, due to memory computing limitations, often cannot support required resource-intensive fail accurately execute tasks. In this work, we address problem edge-assisted analytics in resource-constrained systems by proposing evaluating a rigorous selective offloading framework. The their locally outsource them cloudlet servers only when they predict significant performance improvement. We consider practical scenario where gains resource costs are time-varying; propose an online optimization algorithm that maximizes service without requiring know information. Our approach relies approximate dual subgradient method combined with primal-averaging scheme, works under minimal assumptions about system stochasticity. fully implement proposed wireless testbed evaluate its using state-of-the-art image recognition application, finding cost savings.