Course Identification

While there are no formal prerequisites, an undergraduate-level familiarity with algorithm design and analysis and probability is expected. An undergraduate-level knowledge of complexity theory and machine learning will also be helpful.

This course provides a foundational perspective on individual privacy in the context of statistical data analysis and machine learning. The focus will be on differential privacy, a rigorous mathematical formulation of individual privacy. We will study privacy concerns and attacks, the framework of differential privacy, state-of-the-art differentially private algorithms for data analysis and machine learning, and (time permitting) connections to adaptive data analysis and fair classification.

A past website for the course can be consulted

SPECIAL NOTE for Fall 2020: I expect that all lectures in this course will be remote (via Zoom)

Upon successful completion of this course students should be able to:

1. Demonstrate understanding of basic privacy attacks and concerns, such as re-identification, reconstruction and differencing attacks, composition.
2. Demonstrate understanding of the definition and guarantees provided by (several variants of) differential privacy.
3. Demonstrate understanding of basic and advanced algorithms for privacy-preserving data analysis, and a "toolbox" for differentially private algorithm design.

The Algorithmic Foundations of Differential Privacy, Cynthia Dwork and Aaron Roth

The Complexity of Differential Privacy, Salil Vadhan