Graphical Model is a very useful field in Statistics, Biology, and so
forth these days. People have already found many algorithms to find
top graphical models for real data. However, if we only focus on few
in large number of variables, the usual global graphical model search
will be not efficient. So the first part of this talk will be about
"local graphical model search", which is fairly new and still not
clear very much. Local graphical model search will apply to the
problem if we are only interested in one variable(Y) in thousands of
variables, for example, and wish to know which variables are neighbors
of Y in the  graphical models. We do not care the graphical model
structure elsewhere, but just around Y.

The second part of this talk is a biological project about
cancer. Biology people collected many datasets including the same
7,738 genes with cancer samples and non-cancer samples. Unfortunately,
different datasets have different scales and different missing-data
patterns. Therefore, we cannot combine them. What we did is running
HDBCS(an global graphical model search algorithm by A. Dobra and
M. West) for every dataset, and then try to compare the results among
different datasets(some are cancer, some are non-cancer). The result
will show which genes are related to cancer, based on our work.