The proliferation of the IoT makes numerous smart devices equipped with rich sensing capabilities part of our everyday life. These sensors enable customized services by measuring a user’s ambient environment such as a fitness tracker recording daily activities (e.g., jogging), allowing users who exercise a lot to get an insurance discount.

However, the ubiquity of sensing raises the problem of oversensing [1], namely inferring user’s sensitive attributes or behaviors (e.g., health conditions, political orientation) from the sensor data that was collected for benign purposes.

In this thesis, we will explore the landscape of oversensing, focusing on the following problem: how to discover the oversensing issues in various sensor data in a scalable (i.e., automated) way?

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] How to Curtail Oversensing in the Home