Back
"Preserving Statistical Validity in Adaptive Data Analysis" published in 2015 is co-authored by Cynthia Dwork, Vitaly Feldman, Moritz Hardt, Toniann Pitassi, Omer Reingold, Aaron Roth.
From the STOC ACM 2025 Homepage: The paper on Preserving Statistical Validity in Adaptive Data Analysis studies the fundamental problem in statistics of how to perform adaptive analysis on a dataset without overfitting. This work provided the first principled framework to study this problem, establishing that differentially private computations can be used towards ensuring statistical validity. The deep connection between differential privacy and statistics demonstrated that privacy is not only important as an end goal but can also improve seemingly unrelated aspects of statistical analysis. This paper inspired new lines of work on differential privacy and generalization and influenced how practitioners think about data reuse. More information in our news.
More information