Network routing based on reinforcement learning in dynamically changing networks

In this paper we propose a reinforcement learning (RL) algorithm for packet routing in computer networks with emphasis on different traffic conditions. It is shown that routing with an RL approach, considering the traffic, can result in shorter delivery time and less congestion. A simple, but rational simulation of a computer network has also been tested and the suggested algorithm has been compared with other conventional ones. At the end, it is concluded that the suggested algorithm can perform packet routing efficiently with advantage of considering the dynamics in a real network.