Adaptive Congestion Control Protocol based on Meta-Reinforcement Learning for Data Communication Networks

Congestion control protocols aim to optimize network capacity utilization and maximize throughput by selecting appropriate transmission rates for the sender.When approached as a Reinforcement Learning (RL) problem, congestion control policies derived from this framework exhibit superior performance compared to manually designed protocols.However, the practical implementation of RL algorithms encounters a significant challenge due to the dynamic nature of network conditions, which hampers their ability to generalize to new scenarios.This study presents a solution to this issue by considering the vast range of network conditions as an unknown task, treating it as a concealed variable that can be inferred from observed network history.By acquiring the capability to estimate this task as an underlying state and conditioning the protocol to respond accordingly, the proposed approach achieves continuous adaptation to evolving network conditions.The results demonstrate that this method not only enhances the utilization of network capacity in congestion control algorithms but also ensures protocol consistency across diverse network characteristics.