My research interests is mining information out of rich datasets. In particular, my work is based on the sparse approximation paradigm, where a few patterns are sought that explain a large part of the dataset. In this setting, constraints such as nonnegativity are a way to ensure these patterns are complementary descriptions of the data (e.g. they are not allowed to cancel out each other), making nonnegative approximations ``part-based'' representations of dataset. On the other hand, low-rank approximations (LRA) aim at producing simple patterns that are easily exploitable.
I often consider rather abstract models, sometimes with the goal of applying them to practical problems. Nevertheless I am mostly interested in the theory of constrained low-rank approximation itself!
A few subtopics that I have particular interest in are described below.



