Language-Based Approaches for Ensuring Differential Privacy 

Differential privacy has become the "gold standard" for identifying trends in sensitive data while protecting individual privacy. However, writing programs that satisfy differential privacy turns out to be challenging - instead of crashing, buggy algorithms appear to work, but fail to protect your privacy! 

In this talk, I'll describe two language-based approaches for ensuring that a program satisfies differential privacy - one static, and one dynamic. The static approach, implemented in the Duet language, uses two mutually-embedded languages and two linear type systems to ensure that programs satisfy differential privacy. Duet supports (ε,δ)-differential privacy and recent variants like Rényi differential privacy and zero-concentrated differential privacy.

The dynamic approach, implemented in a Python library called DDuo, tracks sensitivity and privacy relative to explicitly-declared "sources" of sensitive data rather than program variables. This approach allows the analysis to be implemented in a library, with no changes to the host language. 
 

Bio:

Joseph is an assistant professor of computer science at the University of Vermont. His research interests include data privacy, computer security, and programming languages. Joseph received his BS in computer science from Indiana University, and his MS and PhD in computer science from MIT.