Geometric and Multiscale Methods for Statistical Learning

Large, high-dimensional datasets appear in a wide variety of applications. Extracting information from these datasets and performing machine-learning tasks on them can be challenging for both fundamental statistical reasons and because of computational barriers.

In this lecture, Mauro Maggioni will discuss a family of ideas, algorithms and results for learning from high-dimensional data. These methods rely on the idea that complex, high-dimensional data has geometric structures that, once discovered, assist in a variety of tasks, including statistical learning and data visualization. He will focus on multiscale decompositions that can be used to solve problems such as dictionary learning, classification and regression. These decompositions lead to the construction of novel probabilistic models for data, new notions of learning and approximation of high-dimensional stochastic systems.

If this lecture is videotaped, it will be posted here after production.