Machine Learning for Network Slicing in Future Mobile Networks: Design and Implementation

Abstract

Designing mechanisms for orchestration and management is one of the central issues in 5G and Beyond-5G (B5G) network infrastructures. This is due, among other things, to the sheer amount of traffic that they are expected to support, the decentralized nature of the architectures and the services they run. In these future communication networks, orchestration and management mechanisms will be more complicated when compared to previous communication technologies, and centralized legacy solutions are expected to be limited in ensuring revenue for the infrastructure providers, the service providers and a good Quality of Experience (QoE) for the end users.In order to tackle the issues of orchestration and management, both academia and industry have provided state of the art contributions based on Artificial Intelligence (AI) and its integration into 5G/B5G networks. Among these, we can find advanced forms of traffic forecasting, admission control (for network slices and user service requests) solutions based on AI and decentralized Reinforcement Learning (RL). This paper summarizes some of the AI-based solutions developed by the authors as contributions to the problems of orchestration and management in 5G/B5G networks developed within the framework of the MonB5G project. The contributions in this scope are: 1) SCHEMA, a distributed RL mechanism for Service Function Placement (SFC), 2) CATP, a Context-Aware Traffic Predictor, and 3) Pre-BAC, an admission control mechanisms that exploits advanced time-series forecasting.