Learn to Sense: a Meta-learning Based Distributed Sensing Algorithm in Wireless Sensor Networks
Hui Wu, Zhaoyang Zhang, Chunxu Jiao · 2018
In the era of big data, it has been much more challenging to sense and reconstruct a scalar field for wireless sensor network (WSN). Both academia and industry show an urgent need for cost-efficient field estimation algorithm. To address this issue, this paper appeals to a newly proposed communication computation integrated framework. On the basis of reinforcement learning and meta-learning, a distributed two-layer learning and sensing algorithm, which adaptively determines the most informative sensing location, is presented. It significantly reduces the communication overhead and lays a good foundation for future efficient information acquisition system. Numerical results show that the proposed two-layer algorithm brings a remarkable improvement in acquisition efficiency compared to conventional ones. In the meantime, it shows strong robustness to time-varying field functions and task changes.