1 Graph-Deep-Learning-Based Inference of Fine-Grained Air Quality from Mobile IoT Sensors Tien Huu Do, Evaggelia Tsiligianni, Member, IEEE, Xuening Qin, Jelle Hofman, Valerio Panzica La Manna, Wilfried Philips, and Nikos Deligiannis, Member, IEEE Abstract—Internet of Things (IoT) technologies incorporate a large number of different sensing devices and communication technologies to collect a large amount of data for various applications. Smart cities employ IoT infrastructures to build services useful for the administration of the city and the citizens. In this paper, we present an IoT pipeline for acquisition, processing and visualization of air pollution data over the city of Antwerp, Belgium. Our system employs IoT devices mounted on vehicles as well as static reference stations to measure a variety of city parameters such as humidity, temperature and air pollution. Mobile measurements cover a larger area compared to static stations; however, there is a trade-off between temporal and spatial resolution. We address this problem as a matrix completion on graphs problem, and rely on variational graph autoencoders to propose a deep learning solution for the estima- tion of the unknown air pollution values. Our model is extended to capture the correlation among different air pollutants, leading to improved estimation. We conduct experiments at different spatial and temporal resolution and compare with state-of-the-art methods to show the efficiency of our approach. The observed and estimated air pollution values can be accessed by the interested users through a web visualization tool designed to provide an air pollution map of the city of Antwerp. Index Terms—Internet of Things, smart cities, deep learning, variational graph autoencoder. I. I NTRODUCTION The Internet of Things (IoT) is a modern communication paradigm that is based on the interaction and cooperation of a variety of things or objects for a common goal [2]. IoT systems find applications to a wide range of fields, from smart buildings and healthcare to energy management and agriculture [2], [3]. Urban IoT systems can support information and communi- cations technology (ICT) solutions at a city level, realizing the concept of smart cities and establishing a quality-of-life- improving technology that can come at a reduced operational cost [3]. Smart city technologies are expected to have a high impact in everyday life by enabling the development of new services for citizens, companies and public administration organizations. Smart city services involve different IoT applications such as air quality, transportation and healthcare, to name a few. A part of this work has been presented at the 2019 IEEE International Conference on Accoustics, Speech and Signal Processing (ICASSP’19) [1]. T. H. Do, E. Tsiligianni and N. Deligiannis are with Vrije Universiteit Brus- sel, Pleinlaan 2, B-1050 Brussels, Belgium and also with imec, Kapeldreef 75, B-3001 Leuven, Belgium. (e-mail: {thdo,etsiligi,ndeligia}@etrovub.be). X. Qin and W. Philips are with Ghent University, Sint-Pietersnieuwstraat 25, B-9000 Ghent, Belgium and also with imec, Kapeldreef 75, B-3001 Leuven, Belgium. J. Hofman and V. Panzica La Manna are with imec the Netherlands, High Tech Campus 31, 5656 AE Eindhoven, The Netherlands. Among them, air quality monitoring is of significant interest due to the considerable effect of air pollution on human health and ecosystems [4]. The majority of air quality monitoring systems deploy static measurement stations operated by regu- latory agencies. While static stations can provide reliable mea- surements, their high maintenance cost allows only a limited number of installations. On the other hand, an IoT system enables the deployment of a large amount of heterogeneous sensing devices that collect various data over time not only from fixed stations but from low-cost mobile sensors as well, providing measurements at a wider spatial range. Designing an IoT platform is a complex task. The main challenges include: (i) the design of the IoT infrastructure; (ii) the development of data processing and analysis tools; and (iii) the data accessibility from users and decision makers. We summarize the main reasons raising these challenges. (i) The design of an IoT infrastructure includes the inter- connection of sensing and actuating devices providing the ability to share information across platforms through a unified framework [5]. There are two main reasons making the design of an IoT infrastructure challenging. First, IoT devices are normally constrained when it comes to memory space and processing capacity. Sec- ond, the communication standards and protocols need to support the interoperability and sharing of data across different devices [6]. As a consequence, power-efficient communication protocols with low complexity have been developed to support IoT devices [6], [7]. (ii) IoT devices result in big, real-time data streams [8], [9]. Applying analytics over IoT data streams aims to dis- cover new information, predict future insights, and make control decisions. The huge data volume calls for data- reduction methods such as sampling and aggregation [6]. Sampling is applied by either activating only a subset of the IoT sensors following a control policy [10] or adapt- ing the sampling rate of the sensors to communication or power resources [11], [12]. The aggregation approach aims at gathering and combining data from different sources, by leveraging clustering, multi-path, aggregate trees and various data representation techniques [13], [14]. IoT data are scalar numbers coming from real- time data streams, characterized by a huge volume, heterogeneity, spatio-temporal correlation and noise [15]. This is different from other types of big data (e.g., social media data) which may contain text, images, videos or structured records where spatio-temporal correlation is typically not observed. The analysis of air quality