A Survey on Mobile Edge Computing Architectures for Deep Learning Models
Jaehwan Lee, Woongsoo Na · 2022 13th International Conference on Information and Communication Technology Convergence (ICTC) · 2022
With the development of mobile devices and communication technology, the IoT paradigm has recently been gaining attention. Various sensors and high-speed communication technology are equipped by various mobile devices, making it possible to collect a large amount of environmental data. Accordingly, many attempts have been made to process data collected from mobile devices into deep learning models. Since mobile devices have relatively insufficient computing resources and battery constrains to directly execute large-scale deep learning models, these tasks are offloaded to a cloud computing platform instead. In addition, mobile edge computing (MEC) or fog computing architecture has been proposed in order for the heavy tasks to be processed in a nearby computing node, thereby address the latency limits of cloud computing architecture. The deep learning inferencing can be performed with low latency on the surrounding edge computing nodes by applying the MEC paradigm, however, finding the optimal offloading node selection has been challenging problem, which triggered various studies have been conducted. In this paper, communication technologies and edge computing architectures for offloading deep learning models are examined and comparatively analyzed.