Learning-Based Task Offloading for Mobile Edge Computing

Mobile edge computing (MEC) is an important technology for latency-sensitive applications. One of the biggest challenges in MEC is efficiently allocating resources under strict QoS requirements and resource constraints. The purpose of this paper is to study the joint problem of computation offloading and resource allocation in such networks. The problem is formulated as a mixed-integer non-convex optimization problem and is proved to be NP-hard. In order to solve it efficiently, we propose a multi-agent deep reinforcement learning solution based on actor-critic method. To reduce system latency, each agent aims to learn interactively the best offloading policy independently of other agents. The simulation results illustrate the performance and advantages of the proposed solution compared to benchmark solutions.