.png){kind=icon}
.png){kind=icon}
Return
Adaptive Computation Offloading Decision Optimization in MEC-Assisted FL
.png){kind=icon}
PRE
AI
DOI：10.1007/s10922-026-10086-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The tremendous amount of data generated by heterogeneous devices continues to explosively increase. Data processing can be performed with powerful computing resources, such as at remote cloud servers, but problems with communication and service processing delays can occur, causing unprecedented performance interruption. As for prevention, several learning paradigms have been proposed, such as federated learning (FL) and mobile edge computing (MEC), to shift the computational dependency away from the central cloud. However, both FL and MEC still encounter challenges in resources limitation. In this paper, the integration of MEC and FL, the MEC-assisted FL framework, is conducted to find solutions by leveraging advantages of both paradigms. With our proposed adaptive computation offloading decision approach, namely ADORA, utilizing double deep Q-network (DDQN) algorithm to optimize offloading decision-making, this paper aims to minimize system delay and energy consumption considered data privacy and efficient resource utilization. Following simulation experiments, the evaluation on delay, task drop, task completion, and energy consumption have been conducted. The comparison with two other optimization approaches demonstrates our proposed approach’s capability in selecting the optimal offloading decisions mitigating resource burden on resource-constrained devices.
Keywords:
Deep reinforcement learning
Federated learning
Mobile edge computing
Offloading decision
Double deep Q-network (DDQN)
Journal
IF:
3.9
Papers:
1.0K
Citations:
1.3K