Maximizing Privacy and Utility: The Case for PML

This project aims to develop a theoretical framework for privacy-preserving data processing with provable privacy guarantees using the pointwise maximal leakage (PML) measure. PML is a novel statistical inference measure that is explainable, flexible, and can incorporate prior domain knowledge.

This will simplify its legal adoption and enhance utility by better adapted privacy mechanisms.In three work packages, this project will develop analytical tools for privacy risk assessment and optimal mechanism design approaches using PML as the privacy measure, considering fundamental building blocks and approaches of privacy-preserving data processing and learning.

The scientific research questions include exploring the analytical properties of PML, assessing the PML privacy guarantee of principal privacy mechanisms, obtaining optimal designs with PML that maximize the privacy-utility tradeoff, and upper-bounding PML in principal learning and optimization algorithms.The ultimate goal of this project is for PML to become a widely accepted and used privacy measure that enables the design of efficient algorithms with sufficiently strong and meaningful provable privacy guarantees that meet the needs of legal compliance assessment.

This framework will promote the responsible use of data sharing while ensuring that individuals´ privacy rights are protected, leading to significant advances in research, technology, and society.