Reproduce “Persuasive Privacy” 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
Joshua J Bon, James Bailie, Judith Rousseau, Christian P Robert
The authors introduce persuasive privacy, a framework that evaluates how a release changes a recipient’s beliefs using a chosen privacy score and class of prior beliefs. They prove that releasing the exact empirical average deterministically has threshold κ = r₁ + log r₂ and failure probability δ = 0, for specified multivariate Gaussian priors and marginal Dawid–Sebastiani scores, not all recipients.
Key insights
- A game-theoretic, semantics-first approach allows for the creation of purpose-driven privacy definitions that are interpretable and explicitly link to agent-based interactions.
- The framework provides a unifying theoretical lens to assess and understand existing privacy definitions, including various Differential Privacy variants, by mapping them to specific components of the Persuasive Privacy game.
- Deterministic algorithms, which traditional DP struggles to evaluate, can receive non-vacuous privacy guarantees under defined adversarial assumptions within the Persuasive Privacy framework.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.