Andres Molina-Markham
Computer Science Building, Room 150
A concerning issue is the increasing extent to which individuals are encouraged to share potentially private information so that service providers can obtain knowledge about a given population of consumers. While a priori, users assume that their data will be mined by a trusted party (often a particular service provider), the reality is that in many cases individuals are not aware of the types of conclusions that can be drawn from the shared information. Furthermore, individuals do not know who else will be empowered to form these conclusions. In my dissertation, I address the problem of obtaining useful information about a population using data that is gathered by resource-constrained devices and that is combined in a privacy-aware manner via untrusted data miners. More explicitly, I argue that a model for performing distributed privacy-preserving computations without relying on trusted data miners can be scaled to be practical on resource constrained systems. A careful combination of cryptographic techniques and distributed system techniques may enable ubiquitous data collection of private information, such that individuals achieve adequate privacy guarantees and analysts obtain information with adequate utility.
Advisor: Kevin Fu