On the Robustness of Controlled Deep Reinforcement Learning for Slice Placement

The evaluation of the impact using Machine Learning in management softwarized networks is considered multiple research works. In this paper, we propose to evaluate robustness online learning for optimal network slice placement. A major assumption study consider that request arrivals are non-stationary. We precisely simulate unpredictable load variations and compare two Deep Reinforcement (DRL) algorithms: a pure DRL-based algorithm heuristically controlled DRL as hybrid DRL-heuristic algorithm, order assess these changes traffic on algorithms performance. conduct extensive simulations large-scale operator infrastructure. results show proposed approach more robust reliable than real scenarios.