Smart Cities with Deep Edges

Abstract

With the advent of deep learning and new embedded devices capable of running these models at the edge of the network there is potential for deep edges in IoT and smart cities. This will enable a considerable increase in the analytics and urban reasoning that can take place at the edge of the network. The end-to-end latency for these models will also be reduced due to the physical proximity of the edge devices, which allows reasoning one hop away from data generation. This will enable a range of urban reasoning applications that require reduced latency and jitter such as vehicle collision detection, network demand prediction and smart grids. The increased accuracy of deep learning models at the edge will reduce traffic flow to the cloud as only a subset of the data will need to be reported after a first pass analysis. This will improve the privacy of users as edge devices can process the reported data to remove identifiable information to keep the user anonymous before sending it to the cloud. This multi-stage analytics allows for initial urban reasoning on a city wide scale for deriving context information with additional analytics in the cloud focusing on certain domain challenges. In this paper we describe the architecture and advantages of deep edges and compare it against alternative IoT urban reasoning architectures such as cloud-based and traditional embedded devices such as raspberry pis.