17-880 Algorithms for Private Data Analysis

12 units, Carnegie Mellon University

We study the following question in this course How do we perform useful analysis on a data set that contains sensitive information about individuals without compromising the privacy of those individuals To study this question, we will introduce differential privacy, a framework of designing data analysis algorithms with strong, meaningful, and mathematically provable privacy guarantees. We will survey a set of algorithmic tools that allow us to privately perform a wide range of statistical analyses. Of course, privacy does not come for free, and we will also study some of the fundamental limitations imposed by the requirement of differential privacy. Through the discussion of these results, we will also demonstrate some of the most novel and surprising connections between differential privacy and other areas of theoretical computer science, including machine learning theory, cryptography, convex geometry, and game theory.

Prerequisites: none listed

Courseweave needs JavaScript to show sections, workloads and planning tools.