Search papers, labs, and topics across Lattice.
This paper introduces a novel quantitative notion of privacy, termed "Bayesian Adversarial Privacy," that is contextual and specific. It leverages Bayesian decision theory, but advocates for disclosure decisions based on the prior viewpoint rather than conditioning on observed data. The authors present toy examples and computational methods to illustrate the specificities of this new privacy definition, arguing it's more meaningful than differential privacy and more rigorous than statistical disclosure theory.
A new privacy metric based on Bayesian decision theory offers a more contextual and rigorous alternative to differential privacy and statistical disclosure theory.
Theoretical and applied research into privacy encompasses an incredibly broad swathe of differing approaches, emphasis and aims. This work introduces a new quantitative notion of privacy that is both contextual and specific. We argue that it provides a more meaningful notion of privacy than the widely utilised framework of differential privacy and a more explicit and rigorous formulation than what is commonly used in statistical disclosure theory. Our definition relies on concepts inherent to standard Bayesian decision theory, while departing from it in several important respects. In particular, the party controlling the release of sensitive information should make disclosure decisions from the prior viewpoint, rather than conditional on the data, even when the data is itself observed. Illuminating toy examples and computational methods are discussed in high detail in order to highlight the specificities of the method.
