Intelligent Offloading in Multi-Access Edge Computing: A State-of-the-Art Review and Framework

Abstract

Multi-access edge computing (MEC), which is deployed in the proximity area of the mobile user side as a supplement to the traditional remote cloud center, has been regarded as a promising technique for 5G heterogeneous networks. With the assistance of MEC, mobile users can access computing resource effectively. Also, congestion in the core network can be alleviated by offloading. To adapt in stochastic and constantly varying environments, augmented intelligence (AI) is introduced in MEC for intelligent decision making. For this reason, several recent works have focused on intelligent offloading in MEC to harvest its potential benefits. Therefore, machine learning (ML)-based approaches, including reinforcement learning, supervised/unsupervised learning, deep learning, as well as deep reinforcement learning for AI in MEC have become hot topics. However, many technical challenges still remain to be addressed for AI in MEC. In this article, the basic concept of MEC and main applications are introduced, and existing fundamental works using various ML-based approaches are reviewed. Furthermore, some potential issues of AI in MEC for future work are discussed.