A Survey on Mobile Edge Computing for Deep Learning

Abstract

Deep learning-based services such as AI assistants and self-driving cars are of great interest in academia and industry because of their unrivaled performance. Because these services require high computing power, providing such services in mobile devices encounters several practical limitations like battery consumption, heat generation and high latency. To overcome this limitation, a mobile edge computing architecture that offloads computation has been proposed. We introduce 1) resource optimization method, 2) deep learning model optimization method, and 3) joint optimization method of resources and deep learning model as studies to support deep learning-based services under the MEC structure. In particular, joint optimization of resource and deep learning model is a promising solution to respond to dynamic environment changes of networks and devices more efficiently. At the end, we suggest further research topics to enable joint optimization of resource and deep learning model.