Self-Adaptive Bandwidth Allocation to Address Dynamic Traffic Concept Drift in Optical Access Networks

In this invited talk, we will review our recent work on online reinforcement learning-based dynamic bandwidth allocation (ORL-DBA) scheme that addresses concept drift in ML-enhanced optical access networks for supporting low-latency applications against network traffic dynamicity. The scheme improves bandwidth decisions via exploration or exploits existing decision knowledge to keep a low latency, with extensive simulations showing rapid and self-adaptive bandwidth decisions that yield up to 60% latency improvement in dynamic traffic environments.