Deep Reinforcement Learning for Mobile Edge Computing Systems

Abstract

In mobile edge computing (MEC) systems, network entities and mobile devices need to make decisions to enable efficient use of network and computational resources. Such decision making can be challenging because the environment in MEC systems can be complex and involve time-varying system dynamics. To address such challenges, deep reinforcement learning (DRL) emerges as a promising method. It enables agents (e.g., network entities, mobile devices) to learn the optimal decision-making policy through interacting with the environment. In this chapter, we describe how DRL can be incorporated into MEC systems for improving the system performance. We first give an overview of DRL techniques. Then, we present a case study on the task offloading problem in MEC systems. In particular, we focus on the unknown and time-varying load level dynamics at the edge nodes and formulate a task offloading problem for minimizing the task delay and the ratio of dropped tasks. We propose a deep Q-learning-based algorithm that enables the mobile devices to make their task offloading decisions in a decentralized fashion with local information. This algorithm incorporates double deep Q-network (DQN) and dueling DQN techniques for enhancing the algorithm performance. Simulation results demonstrate that the proposed algorithm can reduce the task delay and ratio of dropped tasks significantly when compared with the existing methods. Finally, we outline several challenges and future research directions.