.png){kind=icon}
.png){kind=icon}
Return
Task Offloading With Differential Privacy in Multi-Access Edge Computing: An A3C-Based Approach
.png){kind=icon}
PRE
AI
DOI：10.1109/TCCN.2026.3657108.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-access Edge Computing (MEC) allows mobile users to transfer computation-heavy tasks to the edge of the network, thereby providing low-latency computation services. However, with growing concerns about user privacy, the offloading behaviors of users can lead to the leakage of location and usage pattern privacy. To overcome this challenge, this paper proposes a privacy-aware task offloading framework incorporating differential privacy (DP), with the purpose of protecting the user’s privacy while reducing computational costs. The technical challenge lies in avoiding excessive perturbation caused by the traditional Laplace mechanism, as its unbounded noise may lead to deviations in offloading decisions. To address this issue, the scheme introduces a truncated Laplace based perturbation mechanism to perturb the user’s offloading ratio, thus ensuring that the perturbed ratio stays within a reasonable range. Meanwhile, we theoretically prove that this perturbation mechanism satisfies differential privacy. Considering the time-varying nature of MEC network systems, we develop an Asynchronous Advantage Actor-Critic (A3C) based privacy-aware task offloading scheme, named A3CS. This approach enables the derivation of the optimal offloading strategy within a continuous policy space and speeds up the learning process using asynchronous multi-threaded training in complex and dynamic MEC systems. Simulation results demonstrate that compared with existing benchmarks, the proposed scheme reduces the total computational cost while ensuring the same level of privacy protection.
Keywords:
Multi-access edge computing
differential privacy
truncated laplace mechanism
reinforcement learning
Journal
I
IF:
7
Papers:
1.5K
Citations:
5.5K
Organization
View more organizations