Johannes Kaiser

Johannes Kaiser

Blog posts

2025

On the Hypothesis Testing Perspective of DP

3 minute read

Published:

Differential privacy in essence tries to protect an adversary of distinguishing weather a datapoint was present in an the output of the randomized mechanism underlying data. Therefore it naturally alligns with an Hypothesis testing interpretation with given the outputs of the either $M(S)$ or $M(S’)$: <p align="center"> $H_0$: the underlying data set is $S$
$H_1$: the underlying data set is $S’$ </p> Differential privacy can be measured in terms of the advantage in predicitive performance in this hypothesis testing problem.

Bayesian view on DP

4 minute read

Published:

Often, it is assumed, that the adversary has a balanced prior between these two hypothesis i.e., the prior porbability of a datapoint being and not being in the dataset is assumed to be the same. However, this potentially diverges from the real setting in which an adversary querries large amounts of data knowing, that only few will be contained in the data set the attacked mechanism bases its output on.

2012

Blog Post number 1

less than 1 minute read

Published:

This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.