**2. Background**

EI has witnessed a significant surge in growth lately, being intrinsically tied to the progression of edge computing. However, at the beginning of the ubiquitous and pervasive computing era, there was a significant deficiency of intelligence at the network's edge in favor of a "remote" intelligence. For example, the primary function of Wireless Sensor Networks (WSNs) [9] was to gather and transmit data according to the "sense-and-forward" paradigm, by means of dumb sensors that were relatively simplistic and resource limited. Therefore, cloud computing emerged as a prevalent enabler for supporting WSNs due to its ability to provide scalable data storage and management, remote monitoring and control, and ease of use. Both WSNs and cloud computing have started, hence, to take advantage of one another: the former found the necessary computing power for implementing intelligent solutions to be fed with a huge amount of real-time data provided by the latter.

As the IoT has gained traction [10], a plethora of more sophisticated devices, known as "smart objects", emerged at the periphery of the network. These devices, such as smart thermostats, security cameras, and connected appliances, promised to make our lives easier by automating daily tasks and providing valuable insights. However, despite their improved capabilities, these devices continued to be heavily reliant on cloud-based infrastructure [11]: indeed, devices still remained unable to take actions without the assistance of the cloud since both the volume and the heterogeneity of the data increased dramatically, and consequently, only simple processing tasks and limited storage operations could be performed locally.

Only in the last decade, a number of issues about the IoT–cloud duo has emerged and pushed for a paradigm shift: indeed, while cloud-based infrastructure does provide the necessary scalability and flexibility for IoT devices, it also introduces a number of

challenges. One major concern is the potential for data breaches and privacy violations, as sensitive information is transmitted to and stored on remote servers. Additionally, the reliance on cloud-based infrastructure can also result in unacceptably high latency in field-to-cloud and back transmissions, as well as exorbitant energy and bandwidth consumption. To address these issues, new computing paradigms such as edge and fog computing have emerged, with the intent to carry not only data processing, but more broadly, intelligence (intended as "Interacting, Interoperate, Cooperation, Communicating and Perception" capabilities [12]) as close as possible to the data sources, in the place of remote servers accessed over the Internet.

In this direction, more recently, EI has emerged in order to support the new and ambitious services and scenarios of the IoT. It incorporates both emerging and well-established approaches from edge and cloud computing, AI, data science, and networking to bring intelligence outside the boundaries of the cloud. This approach, also referred to as "Edge AI", pushes as much as possible for local operations, even better if directly on edge devices, such as smartphones, IoT devices, and industrial equipment, rather than in the cloud or a centralized data center. As illustrated in Figure 1, this strategy effectively reduces the amount of data transmitted from the device monitoring a target phenomenon to a remote server, thus reducing the time and the communication that pass between the operations of sensing and actuation. However, the challenge of EI involves not only embedding the characteristics of massive computing systems into tiny, restricted devices, as if trying "to fit an elephant into a small box" [13], but indeed, a full-fledged infrastructure to allow EI, at its extreme, to transform Big Data into intelligent data and to enable real-time device management, system performance, decision-making, and operation monitoring, thus improving overall responsiveness, privacy, effectiveness, and productivity [12,14–16]. Although EI holds the potential to rival the cloud, it cannot maximize its capabilities when employed alone. EI and cloud computing can complement each other seamlessly, with the optimum approach being a synergistic relationship, exemplified by the IoT–edge–cloud continuum [17]. This harmonious connection is crucial in today's technological arena, resulting in enhanced resource utilization, heightened efficiency, and the best outcomes.

**Figure 1.** From remote (e.g., cloud-based) to local (e.g., EI) data processing.
