An ADMM-based incentive approach for cooperative data analysis in edge computing

Authors: WEIWEI FANG, XUE WANG, QINGLI WANG, YI DING

Abstract: Edge computing is a new paradigm that provides data processing capabilities at the network edge. In view of the uneven data distribution and the constrained onboard resource, an edge device often needs to call for a number of neighboring devices as followers to cooperate on data analysis tasks. However, these followers may be rational and selfish, having their private optimization objectives such as energy efficiency. Therefore, the leader device needs to incentivize the followers to achieve a certain global objective, e.g., maximizing task accomplishment, rather than their own objectives. In this paper, we model the aforementioned challenges in edge computing as a Stackelberg game, and integrate this Stackelberg game with alternating direction method of multipliers (ADMM) to solve the conflict between the global and individual objectives. Through rigorous analysis and extensive experiments, we verify that the proposed approach can quickly converge to the optimum regardless of the number of followers and is very robust to parameter variations.

Keywords: Edge computing, ADMM, game theory, computation offloading

Full Text: PDF