IOL: Statistical Data Privacy for Transparent and Reproducible Research

The need for principled approaches to data privacy and confidentiality has become increasingly urgent, especially in the era of artificial intelligence (AI) as data availability, computational power, and analytical tools continue to expand. Traditional anonymization and statistical disclosure control (SDC) methods, such as suppression and top-coding, are often insufficient in modern settings as demonstrated by re-identification attacks, data linkage risks, and model leakage in complex analytical pipelines. At the same time, growing expectations for transparency and reproducibility through replication packages, shared code, and open models make strict data access restrictions alone an impractical long-term solution.

Statistical data privacy frameworks grounded in formal privacy, including differential privacy, offer a compelling alternative. These approaches provide quantifiable privacy guarantees while enabling the public release of algorithms, parameters, and noise mechanisms, thereby supporting both confidentiality protection and scientific reproducibility. In this introductory overview lecture, we present the core ideas behind differential privacy, discuss how formal privacy methods can be integrated into traditional statistical workflows and AI-driven analyses, and highlight practical tools and resources that researchers can use to begin incorporating formal privacy into their own work.