Talk: Leonhard Grosse (March 19, 2025 at 2:00 PM, Seminar room N2409)

Rethinking Privacy: Challenges and Lessons from Developing the Pointwise Maximal Leakage Framework

Leonhard Grosse

KTH Royal Institute of Technology

 

Abstract:

Differential privacy has become a widely adopted standard for ensuring data privacy, but its limitations—such as its impact model performance in learning applications and missing interpretability of its privacy parameter—spark interest into the exploration of alternative privacy frameworks. To this end, information-theoretic privacy measures often offer advantages with respect to utility and interpretability of guarantees. In particular, the pointwise maximal leakage (PML) framework has recently been shown to combine many of the advantages of differential privacy and the framework of quantitative information flow. In this talk, we will discuss some key goals, learnings and challenges encountered when developing the PML framework. Besides introducing the measure and the main results on its convenient properties, we will also focus on open challenges, and what is needed to enable its use in practical applications. These challenges offer insights into the fundamental issues and tradeoffs faced by privacy researchers and practitioners. Further, the overall discussions show how alternative privacy frameworks might be able to address some of the limiting factors that so far have prevented differential privacy to be widely adopted in contemporary applications—and what might limit this potential.

Biography:

Leonhard Grosse received the B.S. degree in electrical engineering and information technology from the University of Stuttgart, Germany, in 2021, and the M.S. degree in information and network engineering from the KTH Royal Institute of Technology, Sweden, in 2023, where he is currently pursuing the Ph.D. degree with the Division of Information Science and Engineering. His research interests include information theory with applications to privacy and learning.