The 5G network slicing feature allows the creation of a multiple of logical networks, known as "network slices", on top of a shared physical infrastructure. Each network slice is customized to specific service requirements and operates in an separate manner from other slices. This technology also is difficult to apply in Radio Access Networks (RANs), since it demands efficient radio resource management while meeting specific needs for each created RAN slices. The first part of this thesis provides an overview of the DRL technology and its application to RAN Management challenges. This will help to develop advanced solutions for capacity sharing in 5G networks. Then, a DRL-based solution is evaluated for multi-tenant scenarios within 5G RAN. The solution involves a Multi-Agent Reinforcement Learning (MARL) approach with Deep Q-Network (DQN) agents. DQN's MARL-based solution is designed to address capacity-sharing challenges within multi-cell networks, allowing users to adjust to changing traffic conditions while maintaining Service Level Agreements (SLA) for each network segment. To this end, the solution is evaluated in four different scenarios and its strengths, weaknesses and opportunities for improvement are examined. Collab platform is used as a tool to test the potential of this tool as an education tool.

Capacity sharing in RAN slicing