The ubiquity of IoT sensors enables customized user services such as smart health or smart home. Recently, the advances in machine learning have been exploited to discover private user attributes (e.g., gender, age) from sensor data collected for different purposes such as activity recognition, violating user’s privacy.

Two recent works [1, 2] utilize state-of-the-art machine learning techniques to suppress private user attributes in sensor data while maintaining the utility of the target application (e.g., target activity recognition remains accurate).

In this thesis, we will critically evaluate the above proposals, with respect to their security (can other private attributes be learned on these data), generalizability (would they still work on a slightly different sensor data?), and deployability (can such approaches run on edge devices?).

The precise topic addressing the above research goal would be tailored depending on your skillset. However, a solid background in machine learning and data mining is required in addition to a thorough understanding of privacy issues stemming from sensor data (also known as inference attacks).

[1] Protecting Sensory Data against Sensitive Inferences

[2] Preventing Sensitive Information Leakage from Mobile Sensor Signals via Integrative Transformation