Dependable Protection of Privacy for Analytics

Purpose and goal: ** Denna text är maskinöversatt **

This project builds a reliable software prototype to protect the integrity of personal data analytics by injecting accurately calibrated noise. To that end, we achieved two main objectives: to understand the risks of injecting noise to protect privacy and to track the privacy budget (ie, a quantitative view of the level of privacy protection offered).

Expected results and effects: ** Denna text är maskinöversatt **

The results project helped to build a robust and reliable system for creating integrity-preserving analysis. Our software can now track the privacy budget and our team understands the risks of an incorrect injection of noise to protect privacy. We have managed to complete three of four tasks. We learned that a task requires more time and resources to succeed.

Approach and implementation: Task 1: Ensuring a correct injection of noise to protect privacy (October, November 2021) - Evaluation of current threats on using floating-point numbers and methods to sample from discrete distributions. - Secure implementation of sampling and adaptation of noisy value representations

Task 2: Tracking privacy budget (June, August, September) - Implementation of an initialization procedure for datasets and the assignment of privacy budget to data analysts. - Evaluation of budget tracking mechanisms and implementation about how analytics consume the privacy budget.