.png){kind=icon}
.png){kind=icon}
Return
Vehicle Coalition-Based Incentive Algorithm for Model Deployment and Task Offloading
.png){kind=icon}
PRE
AI
DOI：10.1109/TNSM.2026.3674158.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In vehicular edge computing (VEC), efficient strategies for model deployment and task offloading provide tremendous potential to improve quality of services for deep neural network (DNN) inference. However, existing works fail to co-consider selfishness and cooperation of vehicles and characteristic of DNN inference tasks, which results in a bottleneck of performance improvement for DNN inference in VEC. This paper aims to fill this gap by investigating a joint model deployment and task offloading problem for DNN inference in VEC. We formulate a problem with an objective of maximizing social welfare, under constraints of per task accuracy level, per vehicle/roadside unit utility, etc. To solve the problem, an incentive algorithm, called ICA, is proposed based on coalition game and auction mechanism by joint model deployment and task offloading for DNN inference in VEC. Additionally, an incentive algorithm, called IDA, is proposed based on deep reinforcement learning and auction mechanism to maximize the social welfare. Besides, we prove that the proposed algorithms guarantee essential economic properties, i.e., truthfulness and individual rationality. We also prove that the proposed algorithms converge, and that the final coalition structure generated by ICA is Nash-stable. Extensive simulation results show that the proposed algorithms outperform the state-of-the-art methods for all cases, in terms of social welfare.
Keywords:
Inference algorithms
Quality of service
Artificial neural networks
Accuracy
Computational modeling
Games
Integrated circuit modeling
Edge computing
Delays
Energy consumption
Vehicular edge computing
DNN inference
model deployment
task offloading
incentive algorithms
Journal
IF:
5.4
Papers:
520
Citations:
9.2K