Deep Reinforcement Learning-Based Task Offloading for Mobile Edge Computing

SDGs: Primary SDG: SDG 9 - Industry, Innovation, and Infrastructure. | Secondary SDGs: SDG 11 - Sustainable Cities and Communities; SDG 12 -Responsible Consumption and Production

Abstract

Mobile Edge computing (MEC) transforms the dynamism of every application that requires high resources and low latency such as Internet of Things (IoT), artificial reality and autonomous vehicles. The mobile devices are also known to have short battery life and processing capabilities coupled with the delays that come along with remote cloud resources. MEC permits tasks to be offloaded to edge servers that are close by. Wireless environments are too complex and varied to make the optimum offloading decisions. The research work addresses the problem of quality conscious task offloading with the help of Reinforcement Learning (RL) framework of Deep Q Network (DQN), which is statistically validated. The proposed solution is a combination of DQN learning and Six Sigma methods to measure the quality of offloading. DQN trained on the Task Offloading Dataset of Kaggle attained a task offloading accuracy of 79.41%, which is higher than other reinforcement learning methods such as Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO), and MultiAgent RL. A comparative research work was able to demonstrate the benefits offered by DQN were statistically significant and of high-quality in-service presentation. This research work contributed to a significant gap in the research since it combines contemporary reinforcement learning techniques with quality assurance standards to provide powerful and pertinent principles in the implementation of MEC.