Differential Privacy at Risk: Bridging Randomness and Privacy Budget

• DOI: 10.2478/popets-2021-0005
• Publication Date: 2020-12-22T11:46:59Z
• Authors: Ashish Dandekar; Debabrota Basu; Stéphane Bressan
• Keywords: Differential Privacy; Information sensitivity

Abstract The calibration of noise for a privacy-preserving mechanism depends on the sensitivity query and prescribed privacy level. A data steward must make non-trivial choice level that balances requirements users monetary constraints business entity. Firstly, we analyse roles sources randomness, namely explicit randomness induced by distribution implicit data-generation distribution, are involved in design mechanism. finer analysis enables us to provide stronger guarantees with quantifiable risks. Thus, propose at risk is probabilistic mechanisms. We composition theorem leverages risk. instantiate Laplace providing analytical results. Secondly, cost model bridges gap between compensation budget estimated GDPR compliant convexity proposed leads unique fine-tuning minimises budget. show its effectiveness illustrating realistic scenario avoids overestimation using quantitatively optimal provides guarantee than classical advanced composition. Although illustration specific chosen model, it naturally extends any convex model. also illustrations how uses balance trade-off utility privacy.