Efficient User Localization in Wireless Networks Using Active Deep Learning

Abstract

In this paper, we investigate the problem of user localization in wireless networks using Active Deep Learning. We constructed a deep neural network to learn the channel state information collected by antennas, and implemented active learning mechanism to improve the efficiency of utilizing labeled data. The performance and behavior of different active learning query strategies are compared and analyzed. We propose a location based query strategy that considers both spatial density and model uncertainty when selecting samples to label. Experimental results show that the active learning methodology is capable of achieving a same level of performance as traditional models using far less labeled samples. In addition, the location based strategy outperforms all the other query strategies for the location prediction problem.