A deep reinforcement learning framework for scalable slice orchestration in beyond 5G networks

This Thesis introduces a flexible Reinforcement Learning queuing-based framework for dynamic slice orchestration in Beyond 5G networks, supporting multiple concurrent slices that span different technological domains and are governed by diverse end-to-end Service Level Agreements. Different (Deep) Reinforcement Learning methods (single or multi-agent) are investigated to address the state and action complexity hurdles arising in such combinatorial problems, which render the use of "vanilla" Reinforcement Learning algorithms impractical. The performance of the proposed schemes is validated through simulations under both synthetic Markovian traffic and real traffic scenarios.