Deep Reinforcement Learning for Task Offloading in a Multi-Access Edge Computing Environment

Edge computing is a relatively recent and popular approach that aims to improve the QoS to customers. At the heart of edge computing, lies the decision if a task needs to be offloaded to an edge server or to process the task locally on the edge device. Reinforcement learning is being extensively to make the offloading decision. Single objective tabular and deep reinforcement learning methods are compared to individually optimize the task drop rate, latency and energy. The deep reinforcement method of learning outperforms the table based method in making the offloading decision for all three objectives consistently.