| Popis: |
Differential privacy is nowadays considered the "gold standard" when releasing information, e.g., statistics, on sensitive data. To avoid leaking too much sensitive data noise-adding mechanisms may be used. ϵ-differential privacy measures the amount of privacy ϵ that such a mechanism ensures. Of course, adding too much noise results into useless, random information, while adding not enough may lead to privacy violations. This problem is known as the privacy-utility trade-off and raises a natural optimality question: how can we maximise utility for a given amount of privacy ϵ. |
