A Formal Verification Model for Performance Analysis of Reinforcement Learning Algorithms Applied t o Dynamic Networks

Routing data packets in a dynamic network is a difficult and important problem in computer networks. As the network is dynamic, it is subject to frequent topology changes and is subject to variable link costs due to congestion and bandwidth. Existing shortest path algorithms fail to converge to better solutions under dynamic network conditions. Reinforcement learning algorithms posses better adaptation techniques in dynamic environments. In this paper we apply model based Q-Routing technique for routing in dynamic network. To analyze the correctness of Q-Routing algorithms mathematically, we provide a proof and also implement a SPIN based verification model. We also perform simulation based analysis of Q-Routing for given metrics.