Seminar Wed. Nov. 2

[Andreas Argyriou]

The seminar talk by Ali Jalali will be on the following topic. It will
take place this Wednesday Nov. 2 at 11-12 am in room 530 at TTIC.

Title:
Advanced in High-Dimensional Statistical Learning

Abstract:
In many areas across science and technology, we increasingly face the
situation that the number of observations are much less than the
number of
variables we want to learn/estimate. In such situations, the only hope
for
consistency is by leveraging underlying structure in the data, such as
sparsity, low-rank, block-sparsity, etc; structure models typically
chosen
for their computational as well as statistical efficiency. However, in
several applications, data may not fit neatly into one structure.

In this talk, we build statistical models and algorithms to capture
data via
the simultaneous use of more than one structural model. This allows
for
robustness, and lower sample complexity, while retaining the same
order of
computational tractability. We present algorithmic and theoretical
results
for methods based on convex optimization, and also faster greedy non-
convex
methods.