Research topic: Privacy-aware transparent machine learning

Umeå University, the Department of Computing Science, is seeking candidates for two PhD student position in Computer Science with focus on data privacy.

The Department of Computing Science is a dynamic environment with around 140 employees from more than 20 countries worldwide. The Privacy-aware transparency decisions research group (led by Prof. Vicenç Torra) conducts research in data privacy for data to be used for machine and statistical learning. It is well known that data can be highly sensitive, and that naive anonymization is not sufficient to avoid disclosure. Models and aggregates can also lead to disclosure as they can contain traces of the data used in their computation. We want to understand the fundamental principles that permit us to build privacy-aware AI systems, and develop algorithms for this purpose. The group collaborates with several national and international research groups, edits one of the major journals on data privacy (Transactions on Data Privacy), and has active links with the private and public sectors. For more information see https://www.umu.se/en/research/groups/nausica-privacy-aware-transparent-decisions-group-/

Project Description

We will develop machine learning algorithms that build data-driven models avoiding disclosure of private information and that are resistant to different types of attacks (t ex. transparency and membership attacks). The objective is to build statistical and machine learning models taking into account different types of privacy models (differential and integral privacy, k-anonymity), as well as different types of scenarios (centralized and decentralized data). Because of that, the project will consider centralized machine learning as well as federated learning approaches. Models are expected to follow trustworthy AI principles, and, in particular, take into account explainability. These models are attractive because they allow people to understand why decisions are made, but at the same time explainability implies additional privacy threats to be tackled.

More Information and Application

View all positions