MTO Colloquium By dr. Reza Mohammadi Tuesday, March 8 WZ 104 at 12:45h

MTO Colloquium
Tuesday, March 8
WZ 104 at 12:45h
By dr. Reza Mohammadi
(Department Methodology & Statistics, Tilburg University)
‘Bayesian Structure Learning in Graphical Models’
Discovering complicated patterns among large number of variables
is one of the main challenges in wide variety of applications ranging from
social science to biology. The challenging question is how to extract the
underlying network structure among those variables from information in
data. To dealing with this issue, graphical models provide powerful tools
and for the purpose of performing structure learning, Bayesian approaches
provide a straightforward tool, explicitly incorporating underlying graph
uncertainty. However, Bayesian inference in graphical models suffered
from computational problems. In this regard, the key is to design
computationally efficient search algorithms that are able to quickly move
towards high posterior probability regions. This talk first will review
some of the basic ideas underlying graphical models, including the
algorithmic ideas that allow graphical models to deal with highdimensionally problems. We next introduce some computationally
efficient search algorithms based on our Bayesian framework to extract
underlying graph structure. The methods have been implemented in C++
and interfaced with R as an R package BDgraph which is freely available
online. In addition, possible extensions of the methods and their
applications in social science are discussed.