Spring 2019

Optimality and large-scale learning under local privacy

Apr 17, 2019 1:00 pm – 3:00 pm 

Add to Calendar


John Duchi (Stanford)


Room 116


In this talk, I will explore some recent work of my own and others on optimal estimation under different local privacy constraints. As part of this, we will explore definitions of optimality for statistical estimators—including notions of instance optimality arising out of classical Fisher information—and their connections with privacy. I will also describe recent work developing optimal algorithms for large-scale model fitting and federated learning under local privacy.