Federated Learning Based Mobile Edge Computing for Augmented Reality Applications
Dawei Chen, Linda Xie, BaekGyu Kim, Li Wang, Choong Seon Hong, Li‐Chun Wang, Zhu Han
Abstract
The past decade has witnessed the prosperous growth of augmented reality (AR) devices, as they provide immersive and interactive experience for customers. AR applications have the properties of high data rate and latency sensitivity. Currently, the available bandwidth is relatively limited to transmit and process enormous generated data. Meanwhile, it is challenging for AR to accurately detect and classify the object in order to perfectly combine the corresponding virtual contents with the real world. In this work, we focus on how to solve the computation efficiency, low-latency object detection and classification problems of AR applications. Firstly, we introduce and analyze the practical mathematical model of AR, and connect the AR operating principles with the object detection and classification problem. To address this problem and reduce the executing latency simultaneously, we propose a framework collaborating mobile edge computing paradigm with federated learning, both of which are decentralized configurations. To evaluate our method, numerical results are calculated based on the open source data CIFAR-10. Compared to centralized learning, our proposed framework requires significantly fewer training iterations.