Multichannel random access optimization via evolutionary algorithm

Abstract

Recently, researchers have applied multi-agent deep reinforcement learning to improve throughput of multichannel random access systems. However, optimizing hyperparameters of deep neural networks (DNNs) remains challenging. This paper proposes utilizing evolutionary computation to optimize such hyperparameters. Numerical results show that the proposed optimization method improves the convergence speed of DNNs compared to the existing manual tuning methods. Index Terms—Random access, hyperparameter optimization, genetic algorithm, deep learning, reinforcement learning.