Robust Algorithms for Machine Learning

Catalog description

As machine learning systems start to make more important decisions in our society, we need learning algorithms that are reliable and robust. In this course, we will (1) cover basic tools in linear algebra, matrix calculus, and statistics that are useful in theoretical machine learning, (2) explore different adversarial models and examine whether existing algorithms are robust in these models, and (3) design and analyze provably robust algorithms for fundamental tasks in machine learning. In particular, we will focus on the research areas of high-dimensional robust statistics, non-convex optimization, learning with strategic agents, and spectral graph theory. This is a research-oriented course where students are asked to read and present sophisticated papers in top machine learning conferences. Knowledge of basic linear algebra, algorithms, data structures, probability and statistics is essential. Prior experience with machine learning is required.