Preventing Sensitive Information Leakage From Mobile Sensor Signals via Integrative Transformation

Abstract

Ubiquitous mobile sensors on human activity recognition pose the threat of leaking personal information that is explicitly contained within the time-series sensor signals and can be extracted by attackers. Existing protective methods only support specific sensitive attributes and require massive relevant sensitive ground truth for training, which is unfavourable to users. To fill this gap, we propose a novel data transformation framework for prohibiting the leakage of sensitive information from sensor data. The proposed framework transforms raw sensor data into a new format, where the sensitive information is messed up and the task information (e.g., human activities) is retained. Training can be conducted without using any personal information as ground truth. Meanwhile, all attributes of sensitive information (e.g., age, gender) can be hidden through a one-time transformation collectively. The experiment results on two multimodal sensor-based human activity datasets manifest the feasibility of the presented framework in concealing users' sensitive information without degrading the usability of the data for activity recognition.

GaitPrivacyON: Privacy-Preserving Mobile Gait Biometrics using Unsupervised Learning

“…proposed a new framework for activity recognition and privacypreserving of sensitive data [26]. The authors wanted to avoid the need for massive collection of sensitive data for model training.…”

“…3.2). For the training of GaitPrivacyON, we adapted the key aspects presented in the image style transformation field [26]. The details are explained in Sec.…”

“…Following this idea, three different loss functions have been considered from the work presented in [26]: task loss (L t ), content loss (L c ), and style loss (L s )…”

“…The style loss (L s ) is responsible for maintaining the transformed data unstyled, thus avoiding the extraction of any sensitive information. For this purpose, we want to modify the style of the data by uniform random noise (N s ) with range [-20, 20] as done by [26]. We consider the Gram matrix (G) to measure the style differences between feature representations.…”

A Method to Improve the Accuracy of Personal Information Detection

“…Another common fraudulent method involves telephone fraud [7]. Based on the information leaked from communication applications on mobile [8], the fraudsters will investigate an individual's net worth, age, graduate school, circle of friends, and other details to determine whether it is worth defrauding them.…”