Authors: Jonathan Lebensold, Doina Precup, Borja Balle

Published on: February 09, 2024

Impact Score: 8.12

Arxiv code: Arxiv:2402.06137

Summary

• What is new: The analysis demonstrates that with bounded queries, it’s possible to achieve pure DP guarantees using Gaussian noise in Report Noisy Max and Above Threshold mechanisms, which was previously unattainable.
• Why this is important: Previous analyses of Report Noisy Max and Above Threshold with Gaussian noise only provided approximate DP guarantees, not pure DP guarantees.
• What the research proposes: By assuming that the underlying queries are bounded, the research provides a method to obtain pure DP bounds for these mechanisms with Gaussian noise.
• Results: The research achieved tighter privacy accounting in high privacy, low data scenarios and introduced a privacy filter that allows for the composition of pure ex-post DP guarantees, resulting in a practically competitive Sparse Vector Technique.

Technical Details

Technological frameworks used: Differential Privacy

Models used: Report Noisy Max, Above Threshold, Gaussian Sparse Vector Technique

Data used: Mobility and energy consumption datasets

Potential Impact

Data privacy and security industries, consumer data analytics, and any firm requiring stringent data privacy measures in analytics.

Want to implement this idea in a business?

We have generated a startup concept here: PrivyAnalytics.