An Introduction to Differentially Private Statistics

Abstract: In today's era of massive data, various scientific and technological endeavors have relied on machine learning or statistics models trained on users (e.g., medical results from patient data, better advertisement algorithms from phone data, etc.). Differential Privacy has recently emerged as one of the most popular methods to protect the privacy of users. In this talk, I will be giving an overview of differential privacy and will focus on how we can solve various statistical problems, such as estimating the mean and covariance of Multivariate Gaussian distributions, with differential privacy using few samples. If time permits, I will describe some more recent advanced methods of solving these problems with even fewer samples.

Computer scienceMathematicsPhysics

Audience: researchers in the topic

MIT Simple Person's Applied Mathematics Seminar