Sequence-Based Deep Reinforcement Learning for Task Offloading in Mobile Edge Computing: A Comparison Study

Mobile edge computing aims to extend cloud services to the network edge, thereby reducing the computational burden on mobile devices and enabling even simple devices to perform computationally intensive tasks within a reasonable time frame. An important issue in mobile edge computing is whether to offload interdependent and heterogeneous tasks on user equipment to edge servers. In recent years, with the emergence of neural combinatorial optimization, numerous learning-based approaches have been applied to address this problem. One significant approach is to model the task offloading problem as a sequential decision problem and implement sequence-based networks to derive offloading decisions. However, the selection of appropriate sequence-based networks remains insufficiently explored. To this end, this paper compares the performance of three mainstream sequence-based networks (i.e., recurrent neural network, long short-term memory network, and transformers), aiming to offer a concise guideline for future researchers in selecting sequence-based networks for task offloading. The experimental results suggest that the long short-term memory network outperforms other sequence-based networks.