Biblio

While cloud-based deep learning benefits for high-accuracy inference, it leads to potential privacy risks when exposing sensitive data to untrusted servers. In this paper, we work on exploring the feasibility of steganography in preserving inference privacy. Specifically, we devise GHOST and GHOST+, two private inference solutions employing steganography to make sensitive images invisible in the inference phase. Motivated by the fact that deep neural networks (DNNs) are inherently vulnerable to adversarial attacks, our main idea is turning this vulnerability into the weapon for data privacy, enabling the DNN to misclassify a stego image into the class of the sensitive image hidden in it. The main difference is that GHOST retrains the DNN into a poisoned network to learn the hidden features of sensitive images, but GHOST+ leverages a generative adversarial network (GAN) to produce adversarial perturbations without altering the DNN. For enhanced privacy and a better computation-communication trade-off, both solutions adopt the edge-cloud collaborative framework. Compared with the previous solutions, this is the first work that successfully integrates steganography and the nature of DNNs to achieve private inference while ensuring high accuracy. Extensive experiments validate that steganography has excellent ability in accuracy-aware privacy protection of deep learning.

Device-free localization of people and objects indoors not equipped with radios is playing a critical role in many emerging applications. This paper presents an accurate model-based device-free localization system LiFS, implemented on cheap commercial off-the-shelf (COTS) Wi-Fi devices. Unlike previous COTS device-based work, LiFS is able to localize a target accurately without offline training. The basic idea is simple: channel state information (CSI) is sensitive to a target's location and by modelling the CSI measurements of multiple wireless links as a set of power fading based equations, the target location can be determined. However, due to rich multipath propagation indoors, the received signal strength (RSS) or even the fine-grained CSI can not be easily modelled. We observe that even in a rich multipath environment, not all subcarriers are affected equally by multipath reflections. Our pre-processing scheme tries to identify the subcarriers not affected by multipath. Thus, CSIs on the "clean" subcarriers can be utilized for accurate localization. We design, implement and evaluate LiFS with extensive experiments in three different environments. Without knowing the majority transceivers' locations, LiFS achieves a median accuracy of 0.5 m and 1.1 m in line-of-sight (LoS) and non-line-of-sight (NLoS) scenarios respectively, outperforming the state-of-the-art systems. Besides single target localization, LiFS is able to differentiate two sparsely-located targets and localize each of them at a high accuracy.