Building edge intelligence for online activity recognition in service-oriented IoT systems

This paper presents the edge intelligence support for smart Internet of Things (loT) using the service oriented architecture. We propose an edge intelligence framework for building smart loT applications. The proposed edge intelligence framework pushes the streaming processing capability from cloud core to edge devices, in order to better support timely and reliable streaming data analytics in smart loT applications. We have designed annotation based programming primitives for developers to build online learning capabilities on edge devices. We have also implemented a user activity recognition engine, and compared its performances between running on either an edge device or cloud servers. Using our edge intelligence framework can improve the real-time and fault-tolerance performance significantly without degrading the activity recognition accuracy in a smart home. (C) 2018 Elsevier B.V. All rights reserved.