The University of Arizona
Structure in Data
Currently, humans are generating and collecting massive amounts of data from nearly every endeavor we are involved in. One of the features of this data is that it tends to have high complexity and is often high-dimensional, which makes it difficult for practitioners to analyze efficiently. However, it turns out that much of this data either has low-dimensional structures hiding in it, or can be effectively represented in a low-dimensional manner. This observation gives rise to many interesting mathematical tools and questions, and has motivated work in manifold learning, sparse dictionary learning, low-rank approximations, dimensionality reduction, and data clustering. This talk will give an overview of some of these problems and tools with a specific focus on the problem of learning a union of subspaces model from data and some associated low-rank matrix approximation methods. Theoretical solutions which give exact recovery in certain cases will be presented as well as experimental results on benchmark datasets from computer vision tasks. We will also discuss related aspects of feature selection and interpretable representations of data via randomized and deterministic column selection.
Coffee to be served 30 minutes prior to the talk in the alcove outside of FO 2.406.
Viswanath Ramakrishna , 972-883-6873
Questions? Email me.