Abstract
With the increasing growth of location-based services (LBS), Wi-Fi localization have attracted great interest due to its ubiquitous wireless coverage in indoor environments. Recently, deep neural networks (DNNs) improved the localization performance using Wi-Fi signals. However, DNN models are vulnerable to adversarial examples by adding a subtle perturbation, which will influence wireless localization accuracy. In this paper, we propose adversarial deep learning for an indoor localization system with Wi-Fi received signal strength indicator (RSSI). In particular, we consider adversarial attacks on floor classification and location prediction with Wi-Fi RSSI, where three white-box attacks methods are exploited including fast gradient sign attack (FGSM), projected gradient descent (PGD), and momentum iterative method (MIM). In the experiment, we show that the performances of DNN-based floor classification and location prediction are greatly influenced under these white-box attacks. We have also implemented adversarial training to generate effective robust floor classification model against adversaries.