Privacy for Data Scientists

This course covers currently available tools to address tensions between utility of data analysis and the privacy of individuals whose data is included. These include frameworks like k-anonymity, differential privacy, and emerging methods for "private machine learning." The course will covers threat-modeling to assess these frameworks and associated algorithms, discuss limitations of the various tools, and study attacks like de-anonymization on real world data sets. The course would include a semester long competition, where student teams would be tasked with designing privacy-preserving methods of data analysis, as well as attempting to compromise the proposed designs. No background would be assumed beyond basic concepts of databases and algorithms.