Wireless Data Acquisition for Edge Learning: Data-Importance Aware Retransmission

By deploying machine-learning algorithms at the network edge, edge learning can leverage enormous real-time data generated by billions of mobile devices to train AI models, which enable intelligent applications. In this emerging research area, one key direction is efficiently utilize radio resources for wireless acquisition minimize latency executing a task an server. Along direction, we consider specific problem retransmission decision in each communication round ensure both reliability and quantity those training accelerating model convergence. To solve problem, new protocol called data-importance aware automatic-repeat-request (importance ARQ) proposed. Unlike classic ARQ focusing merely on reliability, importance selectively retransmits sample based its uncertainty helps be measured using under training. Underpinning proposed derived elegant communication-learning relation between two corresponding metrics, i.e., signal-to-noise ratio (SNR) uncertainty. This facilitates design simple threshold policy ARQ. The first classifier support vector machine (SVM), where distance boundary. then extended more complex convolutional neural networks (CNN) entropy. Extensive experiments have been conducted SVM CNN real datasets with balanced imbalanced distributions. Experimental results demonstrate that effectively copes channel fading noise achieve faster convergence than conventional channel-aware gain significant when dataset imbalanced.