| 研究実績の概要 |
This research investigates the problem of task offloading in mobile edge computing systems while considering the diverse delay sensitivities of tasks. The applicant proposed an actor-critic based deep reinforcement learning (ADRL) model to minimize the total penalty that is attributed to the deadline misses of deadline-constrained tasks and the lateness of delay-sensitive tasks. The ADRL model was trained by a real data set, where the diverse delay sensitivities of tasks can be observed. The simulation results show that the ADRL model outperforms several heuristic algorithms in terms of total penalty. Even if the system is heavily loaded or it does not have a lot of computing power at the edge, the ADRL model can still reach low total penalty as it can learn well from the environment.
|
