Fairness Guarantees in a Neural Network Adaptive Congestion Control Framework

Abstract

The recently proposed neural network rate control (NNRC) framework that achieves queueing delay and queue length regulation, is expanded to further guarantee fair allocation of network resources among competing sources. This is possible by introducing a novel algorithm that controls in a stable and adaptive manner the number of communication channels in each source. Simulation studies performed on a heterogeneous delay, long-distance high-speed network, illustrate all aspects of the developed methodology.