Clustering using the minimum description length principle

Talk announcement:
HIIT Seminar Kumpula, Friday Dec 10, 10:15 a.m., Exactum C222

SPEAKER:
Panu Luosto
University of Helsinki

TITLE:
Clustering using the minimum description length principle

ABSTRACT:
Within a model class framework, the best model for data is according to
the minimum description length principle the one that leads to the most
efficient compression of the data in the worst case sense.  However,
many useful model classes have a property called infinite parametric
complexity, and in those cases an optimal solution cannot be defined in
a straightforward way.

This talk introduces one solution to the problem.  The resulting code
length functions are applied to two kinds of clustering applications.
In the first case, an unknown number of Gaussian clusters is searched
for in the presence of uniform background noise.  In the second
application, clustering with a richer variety of model classes is used
for one-dimensional density estimation.

BIO:
Panu Luosto is a PhD student under the supervision of Jyrki Kivinen in
the Department of Computer Science at the University of Helsinki.


Welcome!
--Matti Järvisalo