Tutorials

Towards Fully Mobile Publish/Subscribe

Speakers:

The publish/subscribe communication paradigm provides a mechanism for anonymous and loosely coupled communications between event producers and interested subscribers. This paradigm, initially used in large scale systems (e.g., the Internet), has also been used recently in wireless sensor networks and the Internet of Things.

The event notification service is the most complex part of a publish/subscribe system. It is responsible for correctly delivering the events to the subscribers. The nodes that constitute the event notification service are called brokers. A centralized event notification service (i.e., composed of a single broker) is much simpler to develop and deploy, but has some clear drawbacks, e.g., its lack of scalability and fault tolerance. Indeed, a single broker is both a bottleneck and a single point of failure. On the other hand, in a distributed event notification service, composed of a set of brokers, if one broker fails, a new route can be found through the remaining brokers. Moreover, the load of each broker is reduced, offering a better scalability. These advantages come at the price of a required coordination among brokers in order to efficiently deliver events.

Mobility in a distributed system is quite a difficult topic. A network composed of fully mobile devices, each working independently and sometimes communicating with each other, without a central connection point is difficult to handle.

In this tutorial a survey of the publish/subscribe paradigm will be presented, with an emphasis in mobility and fault tolerance, even inside the event notification service, i.e., the set of brokers. In this regard, we will present two protocols that handle full mobility in a publish/subscribe system. This means that not only publishers and subscribers are able to move, join or leave the system, but brokers are also allowed this ability.

Graph Neural Networks and Applications in Intelligent Environments

Speakers:

Many real-world networks including the World Wide Web and the Internet of Things are graphs in their abstract forms. Composed of nodes and edges, graphs are naturally capable of capturing interactions between entities. This makes graphs a popular choice for modelling IoT sensing data to analyse large amount of data generated by sensing and computing devices in the network. However, a key challenge arise as graphs are in non-Euclidean spaces, which means that existing Convolutional Neural Networks (CNNs) are not suitable for such data type. E Recently, Graph Neural Networks (GNNs) have emerged as the main solution for deep learning on graphs. GNNs show great performance on graph learning tasks including node classification, graph classification, and link prediction.

In this tutorial, we first introduce the background and preliminaries for understanding GNNs. Then we will describe the architectural design of some state-of-the-art models with a focus on the latest advancements in this area. The tutorial will conclude with a review of some applications of GNNs in intelligent environments.

Toward Bridging Business Processes and IoT Big Data

Speaker:

In recent years, we have witnessed huge momentum of the Internet of Things (IoT) as the new infrastructure in various fields such as smart city, smart transportation, smart manufacturing, intelligent healthcare, smart energy and so on. As the IoT evolves, its focus outstretches from connecting and bringing various devices to the Internet, to synergizing human, cyber and physical worlds, and to delivering services to businesses, etc. Meanwhile, the Process-Aware Information Systems (PAIS) and Business Process Management (BPM) are playing an increasingly important role for the modern enterprises. In line with the above trends, there is an increasing need to integrate IoT and BPM.

BPM may sense the physical world faster and more accurately by means of IoT, and the fast-expanding IoT big data may be more effectively leveraged and ‘programmed’ by means of BPM. However, the integration of BPM and IoT is facing some paradigm misalignment challenges, for example, mismatch of programming mechanisms, mismatch of resource management mechanisms and mismatch of adaptation mechanisms.

This tutorial will focus on the integration IoT and BPM to boost the next generation PAIS. We will overview the challenges and survey the relevant methods. We will cover techniques including ‘abstraction and servitization of IoT data’, ‘resource space’, ‘event modelling and transformation’ and ‘process awareness and adaptation’. We will also give case studies and discuss relevant future research directions.

Video Object Tracking: a Deep Learning Perspective

Speakers:

Video object tracking is one of the core open problems in computer vision. In its simplest definition, it consists of the persistent recognition and localization of a generic target object in a video. This task is at the base of intelligent systems that use video cameras to observe environments and localize specific targets. Examples of practical applications include video surveillance, behavior understanding, autonomous driving, and robotics. Several challenges such as object occlusions, pose and scale changes, rotations and shape variations, and the presence of similar objects, must be tackled to accurately keep track of a target’s position. The ultimate goal of video object tracking is to build robust models capable to overcome such challenging factors. In the past, such issues have been addressed by disparate principles formalizing the concepts of appearance model, motion model, and matching operation. In recent years, algorithms based on deep learning tried to learn such conceptual blocks by exploiting the ability of deep neural networks in learning complex functions from visual examples. Thanks to these advancements, today deep learning-based solutions are the way-to-go to implement strong video tracking algorithms.

The goal of this tutorial is to present the latest progress in the exploitation of deep learning for building an accurate visual tracker. The tutorial will introduce the fundamental concepts to reason in the video object tracking domain as well as the challenges to be faced. The session will then describe how the state-of-the-art solutions employ deep learning techniques. This will include the description of tracking-specific deep neural network architectures and learning modalities, as well as the datasets, protocols, and metrics available to evaluate deep learning-based trackers. In the end, the tutorial will present the most popular software tools to develop and test video trackers.

Resilient Predictive Analytics Services in IoT Smart City Environments

Speakers:

In distributed computing environments, the collaboration of IoT/Edge nodes for predictive analytics at the network edge of Smart City infrastructure plays a crucial role in supporting real-time smart city services like Air Quality Management, Traffic Management, Smart Parking, and Smart Waste Management. When a node’s service turns unavailable for various reasons (e.g., service updates, node maintenance, or even node failure), the rest available nodes could not efficiently replace its service due to different data and predictive model (e.g., Machine Learning (ML) models). To build and maintain the systems’ resilience to node’s service unavailability/failure and avoid interruptions to their predictive services, decision-making strategies based on statistical signatures of nodes’ data should be introduced. Such strategies aim to identify surrogate node capable of substituting failing nodes’ services by building enhanced predictive models.

Mastering data for enabling Smart Cities

Data is being generated, collected, analyzed, and distributed at a fast-growing pace. This growth is due to, among others, the vast proliferation of connected devices (such as cameras, smartphones, sensors, and smart household appliances), the widespread and intensive usage of social networks, and the fast-paced digitization of business and organizational processes and services. As a consequence, these developments are transforming our living environment into smart cities. In this tutorial, we discuss the role of data in realizing the vision of smart cities and how we should treat data to make use of its full potential without imposing adverse impacts on individuals, groups and society. Treating data involves applying various techniques for, for instance, data quality enhancement, privacy and fairness protection, and data lifecycle management. The tutorial describes some enabling technologies to enrich data for use in smart cities in a responsible way.

Feedback and prospects in the development of a smart and resilient village. Example of Smart Village of Cozzano (Corsica, France)

In the Smart Village of Cozzano (Corsica, France), we would like to present feedback on deploying an IT infrastructure for a rural area. The characteristics and performances of Lora-LoraWAN network, digital services developed, and the interactions with the population and the decision-makers.

In a second step, in a prospective version, we will address the issue of these systems by integrating the concept of resilience (at different levels) into climate risks.

From IoT to IoV – Towards a Groundbreaking Merger of IoT and the Automotive Industry

The promising notion of the Internet of Things (IoT) has gained considerable momentum over the past two decades or so and is considered as one of the key technological enablers for the successful realization of a number of application domains, including but not limited to, smart homes, smart cities, smart factories, smart agriculture, and smart healthcare. In particular, in the context of the smart cities, the recent technological breakthroughs in IoT and vehicular ad hoc networks, and their intelligent convergence, have transformed vehicles into smart objects, thereby paving the way for the evolution of the promising paradigm of the Internet of Vehicles (IoV). Simply put, IoV attributes to the IoT-on-wheels, wherein vehicles broadcast safety-critical information among one another and their immediate ambiences via Vehicle-to-Everything (V2X) communication for guaranteeing highly reliable and intelligent traffic flows.

This tutorial would, therefore, present a historical overview of IoT, its (underlying) fundamental concepts and emerging applications, and key research directions. It would emphasize on the notable impact of IoT in revolutionizing the automotive industry, particularly, on its key contributions and potential for the evolution of connected and autonomous vehicles, their supporting V2X infrastructure, and an immensely invaluable IoV data generated via the same that could be intelligently harnessed for providing a better and secure information for decision making in safety-critical contexts. This tutorial would further delineate on the need for strengthening the resilience of the IoV networks along with proposing intelligent solutions for the same.