Hassan Ashtiani - Differentially Private Learning of Distributions

Vector Faculty Affiliate Hassan Ashtiani talks about some recent developments in differentially private distribution learning. These include advances in basic problems of hypothesis selection, learning Gaussians, and learning Gaussian Mixture Models. Through these examples Hassan shows some of the shared challenges in designing private estimators, as well as some generic frameworks (such as private-to-non-private reductions) for addressing them.
Speaker's Bio: Hassan Ashtiani is an Assistant Professor in the Department of Computing and Software at McMaster University and a faculty affiliate at Vector Institute, Toronto. He obtained his Ph.D. in Computer Science in 2018 from University of Waterloo where he was advised by Shai Ben-David. Broadly speaking, a major theme in his research is the design and analysis of sample-efficient learning methods that are robust to (i) privacy-related attacks, (ii) model misspecification, (iii) adversarial attacks, and/or (iv) distribution shift.