The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Practical Differential Privacy at LinkedIn with Ryan Rogers - #346
8 snips
Feb 7, 2020 
Ryan Rogers, a Senior Software Engineer at LinkedIn specializing in differential privacy, shares insights on user data privacy in analytics. He delves into his innovative paper on differential privacy and top-k selection, highlighting how LinkedIn balances user anonymity while providing aggregate insights. The discussion covers challenges in real-world applications, the role of Gumbel noise in algorithm performance, and the significant collaboration in advancing differential privacy in the tech industry.
AI Snips
Chapters
Transcript
Episode notes
Apple's DP Adoption
- Ryan Rogers became interested in differential privacy during his PhD work with Aaron Roth.
- Apple's adoption of differential privacy in iOS 10 further fueled his interest and led him to work there.
Impracticality of Existing Algorithms
- Existing differential privacy algorithms for top-k queries require knowing all possible items beforehand.
- This is impractical for exploratory data analysis where the goal is to discover unknown items.
Practical DP Algorithms
- Practical differential privacy algorithms can be layered on existing infrastructure.
- This allows leveraging existing systems while adding privacy guarantees.