Reproduce “Reliable Estimation of Differential Privacy under Selection Bias” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Valentin Dorseuil, Meenatchi Sundaram Muthu Selva Annamalai, Jamal Atif, Olivier Cappé, Nicolas Papernot
The authors correct privacy audits that select the best membership-inference threshold or number of guesses from the same data used to compute a lower bound. They use established multiple-comparison methods to preserve coverage. In a Gaussian-mechanism simulation, uncorrected selection had 19.36% miscoverage versus 0.11% for five-fold correction at 95% confidence; real-model analyses compare bounds, not coverage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.