Differential Privacy: The Census Algorithm

Series
ACO Student Seminar
Time
Friday, September 6, 2019 - 1:05pm for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Samantha Petti – CS, Georgia Tech – spetti@gatech.eduhttp://people.math.gatech.edu/~spetti3/
Organizer
He Guo

For the first time in 2020, the US Census Bureau will apply a differentially private algorithm before publicly releasing decennial census data. Recently, the Bureau publicly released their code and end-to-end tests on the 1940 census data at various privacylevels. We will outline the DP algorithm (which is still being developed) and discuss the accuracy of these end-to-end tests. In particular, we focus on the bias and variance of the reported population counts. Finally, we discuss the choices the Bureau has yet to make that will affect the balance between privacy and accuracy. This talk is based on joint work with Abraham Flaxman.