ADMIRE_SCALE: A Scalable Routing with Hierarchical Reinforcement Learning in Large-Scale IPoDWDM Networks

Abstract:
We propose ADMIRE_SCALE, a distributed collaborative data-driven and model-driven routing engine in large-scale IPoDWDM networks. By problem decomposition and the introduction of hier-archical reinforcement learning, more than 60% training time and decision-making time saving have been achieved in two real-world network topologies compared to ADMIRE.