Enabling Ensemble Reinforcement Learning with ICN in Mobile Edge Computing Environments
Yaodong Huang, Biying Kong, Wenliang Feng, Changkang Mo, Han Liu, Yukun Yuan, Laizhong Cui · 2024
Smart devices now own the capability to conduct machine learning tasks. Such capabilities offer the potential to improve the learning efficiency and accuracy for certain difficult learning tasks with information sharing among edge devices. In this paper, we discuss the implementation of ensemble reinforcement learning in mobile edge computing environments in real-time. We propose a novel system using Information-centric Networking to facilitate fast and efficient multicast transmission over edge devices. The system improves the efficiency and accuracy of ensemble reinforcement learning over edge devices by using targeted data transmission. To achieve efficient and instant transmission, we propose a new transport layer protocol as the middleware for upper-layer learning data and lower-layer ICN links and radios. We implement the proposed system using real edge devices and conduct experiments with several training testbeds to evaluate the performances. The results from the experiments show that the system can achieve up to 28.47% better reward within the same iterations compared to traditional multicast networks and 7.26 times reward compared to training without instant information sharing.