Anticipatory session management and user plane function placement for ai-driven beyond 5g networks

In this paper we aim at facilitating the flexible 5G core network architecture with foresighted session management. To achieve this we propose a 5G-native architecture utilizing a three-stage learning approach, which covers the dimensions of interest to manage sessions in a foresighted manner. In this perspective we exploit the 5G concepts of per-user mobility and activity patterns, apply our CODIPAS RL Framework to anticipate the user behavior and network requirements, and utilize the outcome for learning-based network optimization on the session level and transport network stretch utilizing our Learning of Learning approach. We further contribute with a model that exploits the foresighted session management approach by utilizing the prediction of user behavior for optimized intermediate User Plane Function (iUPF) placement.