作者: Wanjiun Liao , Yi-Huai Hsu , Wan-Ni Lin , Kai-Hsiang Liu
DOI: 10.1109/WCNC49053.2021.9417477
关键词:
摘要: In this paper, we study fine-grained offloading for multi-access edge computing (MEC) in 5G. Existing works computation is on a per-task basis and do not take into account the execution order among tasks one application. Fine-grained offloading, other hand, considers task structure of an application upon making decision may only offload computation-hungry to MEC, thus better use system resource. To solve problem, propose online solution based Actor-Critic Federated Learning, called AC-Federate. AC-Federate, consider multi-MEC network which each node trains model-free advantage (AC) model local data. The AC jointly optimizes continuous actions (i.e., radio resource allocations) discrete action decision), with weighted loss function. further improve inference accuracy model, uploads gradients its actor critic neural networks central controller asynchronous manner. then ensembles collected from different nodes updates all integrated parameters. Simulation results show that proposed AC-Federate outperforms DDPG others terms delay, energy consumption, mixed consideration delay consumption performance even when number UEs very large.