Distributed Machine Learning for Internet-of-Things in Smart Cities

Building an intelligent and resilient city is a global mission for both research communities and the general public. Recent advancement on Internet-of-Things (IoT) and machine learning enable new solutions and applications, such as smart energy systems and autonomous vehicles. In the past, most of the IoT systems report sensor data to the cloud and perform machine learning. However, this centralized approach causes high communication overheads and delay. In this paper, we present a distributed machine learning framework for IoT in smart cities. It enables machine learning algorithms to be run on heterogeneous nodes with different capabilities, including end devices, edge devices, and the cloud. We first describe a distributed machine learning-assisted framework for network management in IoT for smart cities. We then provide a case study of urban monitoring using vehicular sensor network. A secure and distributed information fusion scheme based on statistical machine learning is proposed. Finally, we discuss potential extensions on scalable computation and other distributed machine learning methods for large scale networks.