Currently, the execution of machine learning algorithms is typically batch, offline and centralized. Network management and its services require massive execution of distributed and real-time data. In many situations, the real-time validity of the generated data is limited, requiring the reduction of latency in communication and processing. Furthermore, data transmission in a distributed environment is subject to the quality of communication channels, network congestion and available energy in mobile devices. Such restrictions require solutions based on Distributed Artificial Intelligence (DAI), which goes far beyond the traditional execution of machine learning algorithms. An additional strong restriction arises from the adoption of the new General Data Protection Law – LGPD. The data privacy restriction is addressed by the federated learning technique. The large number of devices connected to the Internet of Things requires the handling of a high volume of data generated by thousands of sensors, requiring solutions that meet scalability, geographic distribution, mobility, heterogeneity, security and privacy requirements.
Adaptive resource allocation and orchestration are challenges to be overcome in large-scale IoT networks with thousands of sensors. The integration of IoT and AI enables the construction of diverse intelligent systems, such as smart cities, smart healthcare, and smart energy systems. In addition to typical IoT challenges, solutions must consider the dynamic variation of different demands. Demand prediction is crucial for adaptive systems. AI plays a critical role in the realization of 6G networks and their applications. There are several ways in which AI can be used in 6G, including conventional use of AI for prescriptive, predictive, diagnostic, and descriptive analytics. Prescriptive analytics can be used to make decisions or predictions related to edge AI, such as cache placement, AI model migration, dynamic and adaptive scaling of network slices and their service function chains, as well as automatic resource allocation (e.g., spectrum, cloud, and backhaul).
Predictive analytics helps to predict the future from real-time data acquired for events such as resource availability, user behavior, user location, and traffic patterns, to proactively change the network. Proactive actions can adjust resource allocation, instantiation of security solutions, pre-migration of services at the edge. Diagnostic analytics refers to the detection of network faults and anomalies. This research project aims to investigate intelligent solutions for communication networks and IoT based on DAI, solutions for the allocation and orchestration of distributed resources, for infrastructure management and provision of intelligent services.