Machine Learning Systems for Smart Services in the Internet of Things
Machine learning technologies are rapidly emerging in the Internet of Things (IoT) to provision smart services with intelligence. Current approaches often rely on IoT devices to collect data and send it to the cloud for further processing. While the cloud offers abundant computing resources, communication delays over wireless networks and from traversing the Internet present challenges for the IoT to fully benefit from machine learning technologies in dynamic and distributed settings. This review article moves beyond existing machine learning algorithms and cloud-driven design to investigate the less-explored socio-technical and scaling aspects for consolidating machine learning and the IoT. In particular, we provide an extensive coverage of the latest developments (up to 2020) on scaling and distributing machine learning across cloud, edge, and IoT devices. In addition, we derive a multi-layered framework to classify design considerations and illuminate multi-stakeholder concerns that shape the development of IoT smart services. Our goal is to equip researchers and system engineers with a better understanding of the underlying problems resulting from the intersection of machine learning and the IoT. To inspire follow-up research, we identify open problems and highlight emerging trends with regard to on-device inference, automation, and privacy-preserving machine learning. To the best of our knowledge, this is the first comprehensive review to address the fundamental concerns of developing and deploying machine learning systems in the rising cloud-edge-device continuum in terms of functionality, business alignment and trustworthiness.
READ FULL TEXT