Differential Privacy at Scale - Uber and Berkeley Collaboration - USENIX Enigma - 2018

Explore a collaborative effort between Uber and Berkeley researchers to implement differential privacy at scale in this 21-minute conference talk from USENIX Enigma 2018. Delve into the challenges of applying differential privacy techniques to real-world industry requirements and learn about the open-source releases resulting from this partnership. Discover how the team addressed privacy risks, overcame limitations in current research, and developed practical solutions for large-scale data analysis. Examine the Chorus framework, which enforces differential privacy through query rewriting, and understand mechanisms like Elastic Sensitivity and Sample & Aggregate. Gain insights into experimental evaluations and the broad support for analytics queries in privacy-preserving environments.

Differential Privacy at Scale
Examples of Privacy Risks
Outline
Anonymization: Not a Solution
Differential Privacy: a Formal Privacy Guarantee
Real-world Use of Differential Privacy
Challenges for Practical General-purpose Differential Privacy
Broad Support for Analytics Queries
Easy Integration with Existing Data Environments
Chorus: a Framework for Practical Privacy preserving Analytics
Chorus Enforces Differential Privacy by Query Rewriting
Example Mechanism: Elastic Sensitivity
A Rewriter for Elastic Sensitivity: Concrete Example
Example Mechanism: Sample & Aggregate
A Rewriter for Sample & Aggregate
Experimental Evaluation
Mechanism Support for Real-world Queries