Deep Learning in IoT and Edge/Fog Integrated Environments: A Review

Abstract

Over the recent past, the applicative span of deep learning has encompassed almost every field of human civilization like health care, manufacturing, space exploration, etc. Conventionally, the deep learning applications entail extensive memory and computing requirements, both of which are in dearth in the Internet of Things powered systems. Despite that, deep learning managed to find its way into the IoT world by transferring heavy computational tasks to the cloud. Such systems require substantial data transfer between the IoT sensors and the cloud demanding huge bandwidth requirements. Edge computing facilitates the curtailment of the huge data transfer either by crunching down data from sensor nodes into simpler representations or by providing supplementary computational support to the cloud. In the literature, numerous IoT applications have been implemented using this paradigm. Various researchers have used different deep learning networks like convolutional neural networks, auto-encoders, etc., with the edge computing, and showed how the latency and the bandwidth required can be minimized in IoT applications. In this paper, we present a comprehensive study of deep learning-based IoT applications in edge/fog integrated environments. We prepared an analytical comparison between the related studies in this literature to facilitate the research advances in this field.