I am currently working with differential privacy and have a question regarding the composition of two DP mechanisms. In particular I am considering the following problem: consider having obtained a result $S$ of a DP mechanism $M_1(D)$ which satisfies $(\epsilon_1, \delta_1)$ differential privacy (for example via a Gaussian Mechanism). Now lets say I want to perform some mechanism $M_2$ on $S$, i.e. $M_2(S)$. Is there some bound I can give on what $M_2$ does with the privacy parameters of $M_1$? In other words can I amplify the privacy of a already private mechanism post-hoc?

Essentially the question boils down to the post-processing property of DP. Specifically, we know that we cannot worsen the privacy guarantee in post, but is there a way to amplify it after DP protection has already been applied?

New contributor
ABCCHEM is a new contributor to this site. Take care in asking for clarification, commenting, and answering. Check out our Code of Conduct.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.