Development of an adaptive TCP algorithm based on machine learning in telecommunication networks
Афонин Игорь Леонидович, Aleksandr V. Gorelik, Said S. Muratchaev, Alexey Volkov, E. K. Morozov · 2019 Systems of Signal Synchronization, Generating and Processing in Telecommunications (SYNCHROINFO) · 2019
There are many widely used overload control algorithms, such as Reno, Tahoe, CUBIC, and many others. In the article are shown the network overload and packet delay time are determined by a machine learning algorithm with reinforcement. The proposed algorithm uses TCP CUBIC. The result of the algorithm is to study the behavior of the window overload size, using reinforcement learning, as a paradigm, where the algorithm is able to learn and make optimal decisions by using the reward function. Simulation results show performance improvements in the form of increased bandwidth and reduced packet latency compared to various existing TCP algorithms.