Unlike 4G networks, 5G mobile networks were designed to serve three categories of devices, identified as 5G use cases: eMBB (Enhanced mobile broadband), URLLC (ultra-reliable low latency communications), and mMTC (massive machine type communications). Thus, there are different types of devices that networks are being designed to serve, with different requirements in terms of rate and latency, in addition to the possibility of a much larger number of devices to be served. Current systems are in a transition process, from a network typically designed to deliver content, without considering the characteristics of each application that requests network resources, to this new, more flexible network. However, existing 5G mobile networks (3GPP Releases 15 and 16) still do not have enough flexibility and intelligence to meet all the requirements for the three main use cases. All the flexibility desired by 5G and beyond 5G (B5G) to meet the large number of different scenarios leads to increased complexity in network management. In this context, artificial intelligence (AI) applied to 5G and B5G mobile networks allows, for example, to achieve the long-awaited characteristics of self-organizing networks (SON) and zero touch. As a result, AI can leverage the end-to-end efficiency of modern and sophisticated networks and provide personalized quality of experience. In this context, an important tool is the AI-based closed loop, as studied by the ETSI Experiential Networked Intelligence Industry Specification Group (ENI ISG). But networks were not properly designed to accommodate AI-based tasks such as data collection, processing, and distribution of results. Similarly, 6G networks are expected to have even greater requirements to meet the needs of even more demanding applications, resulting in even greater complexity to manage these networks in order to meet such requirements. Thus, network management needs to become more autonomous, which is the main point of so-called cognitive networks, which is the concept of a network that observes information to find patterns of resource usage by users, which help in decision-making to optimize the operation of the network without or with minimal human intervention. The rules of actions taken automatically by the network will be passed on by human controllers from the so-called intentions, or intent-based networks. Therefore, this proposal is focused on innovative research using machine learning and artificial intelligence (ML/AI) applied to telecommunications networks to promote the automation of B5G networks. Focusing on processes performed in the physical layer, such as resource allocation to users and beam selection. To assist ML/AI processes, the project will also focus on creating replicas in virtual environments of relevant network scenarios, to allow for the collection of realistic data and experimentation with decisions before applying them in the field.