Cold-Start Heterogeneous-Device Wireless Localization

In this paper, we study a cold-start heterogeneous-device localization problem. This problem is challenging, because it results in an extreme inductive transfer learning setting, where there is only source domain data but no target domain data. This problem is also underexplored. As there is no target domain data for calibration, we aim to learn a robust feature representation only from the source domain. There is little previous work on such a robust feature learning task; besides, the existing robust feature representation proposals are both heuristic and inexpressive. As our contribution, we for the first time provide a principled and expressive robust feature representation to solve the challenging cold-start heterogeneous-device localization problem. We evaluate our model on two public real-world data sets, and show that it significantly outperforms the best baseline by 23.1%–91.3% across four pairs of heterogeneous devices. Introduction Indoor localization using wireless signal strength has attracted increasing interests from both research and industrial communities (Haeberlen et al. 2004; Lim et al. 2006; Xu et al. 2014). The state of the art in wireless localization is the learning-based approach (Haeberlen et al. 2004; Zheng et al. 2008). In an environment with d1 ∈ Z access points (APs), a mobile device receives wireless signals from these APs. The received signal strength (RSS) values at one location are used as a feature vector x ∈ R1 , and the device’s location is a label y ∈ Y , where Y is the set of possible locations in the environment. In an offline training stage, given sufficient labeled data {(xi, yi)}, we learn a mapping function f : R1 → Y . In an online testing stage, we use f to predict location for a new x. Most of the existing models work under the data homogeneity assumption, which is impractical given the prevalent device heterogeneity. Specifically, the assumption requires the data used in training f to have the same distribution as that used in testing. However, in practice, users carry a variety of heterogeneous mobile devices, which are different from the device used to collect data in training f . Due to different sensing chips, these heterogeneous devices easily receive different RSS values even in the same location, thus Copyright c © 2016, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. 0 0.2 0.4 0.6 0.8 1 S=T43, T=T60 S=N810, T=D901C S=N810, T=N95 S=N810, T=X61 A cc ur ac y