by Karin Anna Hummel (Johannes Kepler University Linz), Paul Smith (Austrian Institute of Technology), Markus Tauber (Research Studios Austria), Peter Dorfinger (Salzburg Research) and Olga Saukh (Graz University of Technology)
The NET-IT initiative at the Austrian Computer Society is making networks intelligent to ease adjustment to the digital evolution and societal needs: Let the network learn and act driven by goals!
Europe is experiencing an ongoing digital evolution, yet also faces several challenges for society such as the energy crisis, climate change and supply chain issues in a global economy. Supporting new digitised applications by low latency, high reliability, ubiquitous availability and high throughput networking technologies, while also being sustainable, is a challenge for future networks. Network intelligence is a high-level research topic that is expected to provide solutions to the pressing heterogeneous demands, which we investigate in the research and innovation initiative NET-IT [L1], a working group of the Austrian Computer Society. In this article, we discuss novel networked applications, network intelligence methods and major challenges of a future research agenda on network intelligence.
Intelligently Networked Applications
Innovation in networks and edge and cloud systems is driven by the emergence of novel application fields, which in turn reflect societal changes and needs. Selected examples are:
Digitisation of food production. Environmentally friendly food production in Europe relies on novel approaches such as precision farming, enabled through digital technologies. Autonomous networked control systems [1,2] allow for large-scale wireless networked systems that sense and collect data, and reason about the status of food production such as the health of crops and irrigation. Figure 1 shows a sample system employing drones, see also [L2]. Easy set-up, secure communication, real-time control, and providing the system “as a service” according to farming goals is demanded.
Digitised missions. Missions – such as search and rescue operations or inventories in warehouses – may involve humans, sensors, cloud/edge computing, and ground and aerial robots. The wireless network set-up should be fast and adaptive. Energy supply for the network, and best use and placement of vehicles and robots are key factors that result in a complex solution space for automated mission-based networks. Besides, missions are time-critical and often require low latency and timing guarantees.
Digital health and digital sports. The digital evolution also changes the medical and sports sector. Shifting from quantitative metrics to qualitative metrics allows a new user experience. Providing, e.g. real-time feedback on the quality of the user’s steps while recovering from an injury, or on the turns while skiing may improve therapy results and skills. The supporting system has to place different services autonomously at different locations (device, edge, cloud) to fulfil the application’s specific timing, precision and privacy needs.
Figure 1: Precision farming system – (left) sensed micro-climate conditions visualised by different colours in a vineyard and (right) scanning drone .
These examples demonstrate that addressing system, societal, and ecological demands altogether, results in a manifold of influential factors which challenge rule-based networking mechanisms and demand novel approaches.
Intelligence in Networks and Networked Systems
The vast amount of data available in today’s networks makes it possible to establish a novel portfolio of intelligent networking technologies that may employ the following methods:
Detection and forecasting. Digital applications need to be aware of current and future connectivity. Hereby, network intelligence may help to predict the future communication network quality, e.g. at locations not visited before. Network monitoring and measurement tools are enablers of accurate classification and forecasting of network connection quality .
Autonomy and adaptation. Changes in application demands lead to the need for network adaptation, which may be determined by learning from past observations. Different machine learning approaches may be applied such as automated reinforcement learning based on evaluating the network performance to generate a reward for the networking decisions taken, or, in case manifold features are available, deep learning. In an ideal setting, the network may then adapt to a current situation autonomously.
Selection. Current networked applications may often choose between available networking solutions. In the case of wireless networks, e.g. low-power wireless communication technologies could be preferred to reach a low environmental footprint. An intelligent network solution is expected to select the best technology among available ones, e.g. 5G and future 6G, Wi-Fi, BLE, ZigBee, ANT+, NRF24 or UWB. The placement of service provisioning (edge/cloud) is another autonomous selection task.
Context. Not only data collected by network monitoring, but also data describing the wider context of networked devices, sensors, and vehicles can support networks to improve their performance. For example, location and mobility contexts are of interest for mobile networks, as are the energy demands of networking decisions and the current situation contexts.
Challenges for Network Intelligence
Network intelligence methods have to address traditional networking challenges such as scalability, reliability, interoperability, security, dynamic reconfiguration and programmability. In addition, novel embedded control systems have stringent real-time, dependability and safety requirements. Intelligent networking solutions thus need to provide guarantees from a holistic system perspective and to fulfil certification demands, which is an open issue.
Network intelligence also faces challenges inherent to AI and machine learning, which originate from the dependency on the data. Data bias and small datasets limit the generalisation of methods. Other limitations are due to energy costs and the computation time for (re-)training. Finally, network, edge and cloud providers need to understand the autonomous decisions taken by intelligent networking technologies, which calls for explainable network intelligence.
Network intelligence is a promising paradigm to address the heterogeneous demands on future networks that are, on the one hand, originating from employment in non-traditional sectors due to ongoing digitisation and automation, and on the growing concerns about energy resources and the demand to rethink the impact of technology on society. NET-IT [L1] is an open collaboration initiative that focuses on the particular challenges of network intelligence.
 M. Tauber, et al., “Passive precision farming reshapes the agricultural sector”, IEEE Computer, Vol. 56, No. 01, 2023. doi: 10.1109/MC.2022.3218388.
 F. Papst, et al.: “Embracing opportunities of livestock big data integration with privacy constraints”, in Proc. of 9th Int. Conference on the Internet of Things, ACM, 2019. doi:10.1145/3365871.3365900
 S. Farthofer, et al. “An open mobile communications drive test data set and its use for machine learning”, IEEE Open Journal of the Communications Society, Vol. 3, 2022. doi: 10.1109/OJCOMS.2022.3210289
Karin Anna Hummel, Johannes Kepler University Linz, Austria
Paul Smith, Austrian Institute of Technology, Austria