Advancing Differentially Private Random Forests
Name of applicant
Christian Janos Lebeda
Title
Postdoctoral Fellow
Institution
University of Copenhagen
Amount
DKK 2,029,256
Year
2025
Type of grant
Reintegration Fellowships
What?
Differential privacy is the gold standard technology for privacy-preserving statistics and machine learning. Random forests are general-purpose machine learning tools useful in many domains. I will advance the state of the art for differentially private random forests. These techniques allow researchers and practitioners to analyze sensitive data while rigorously protecting individual privacy.
Why?
Random forest methods are popular in many fields that rely on sensitive data, such as medicine, economics, and social sciences. Privacy concerns often limit access and use. By improving privacy-preserving algorithms, I will enable responsible data-driven discoveries without exposing personal information.
How?
I will design new theoretical tools for differential privacy by adapting modern techniques to random forests algorithms. I will analyze the privacy and accuracy of these algorithms in theory, and then test them in experiments to confirm that the methods work reliably in practice.