Privacy preserving techniques to make shared analyses

Data is often spread over multiple institutes. Analyzing this data is difficult due to privacy concerns. Simply sharing the data between the institutes may not be possible.

Federated learning is a field that has recently risen in prominence where this problem is solved by removing the need to directly share data. Instead of bringing the data to a central server for analysis, the analysis is brought to the data. Local calculations are executed at each party, the results of which are aggregated in a secure way. In this manner, only the final result of the analysis is ever shared, and individual data-owners can be sure their data remains secure.

In this thesis various techniques for federated learning are investigated. The focus lays on vertically partitioned data, that is to say, on scenarios where the different institutes collect different types of data on the same individuals. For example, combining socio-economic data collected by Statistics Netherlands with medical data collected by a hospital.

This research has been funded by the NWO and was part of the CARRIER project.

Daalen, F. van (2025). Privacy preserving vertically partitioned federated learning: New techniques and considerations. Dissertation, Maastricht University, doi:10.26481/dis.20250328fd.