36-763 Advanced Theory II

6 units, Carnegie Mellon University

A natural property that we want from an estimation procedure is stability: a small change in the training data should not lead to drastic challenges in the algorithm's output.Over the years, various notions of algorithmic stability have been proposed such as robustness to outliers, robustness to heavy-tailed data, differential privacy, and replicability. While these notions appear to encode different notions of stability prima facie, it has been observed that, surprisingly, these notions are intimately related. In this course, we will study these notions and understand the formal connections between these notions. This course will be theory-oriented and will follow many recent papers in the area.

Prerequisites: none listed

Offered: Spring 2026

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