Dynamic Machine Learning Algorithm Selection For Network Slicing Resource Management
22 June 2022
As Advanced 5G/6G networks bring more enhanced performance in terms of latency, reliability, and connectivity allowing the connection of a big number of devices with differentiated and stringent requirements. The Network Slicing technique allows efficient resource allocation and sharing to customized network services through the creation of multiple virtual network instances. Also known as Virtual Network Embedding (VNE), it is an NP-hard problem with different existing solutions. In this demo paper, we present a robust Deep Reinforcement Learning (DRL) solution based on the Algorithm Selection paradigm. The proposed solution uses an Offline or Online Algorithm Selection solution to select a DRL algorithm among a portfolio of agents based on their performance history. The simulation platform we have developed is a containerized testing bed that uses Omnet++ as a simulation environment to show the outperformance of our algorithm selection-based approach compared with standalone algorithms.
