[Solomonov Seminar] 219. Solomonov seminar

[Marko Grobelnik]

V petek 26. marca bo ob 13:00h v Oranzni predavalnici IJS
(drugo nadstropje glavne zgradbe), 219. Solomonov seminar.
Posnetki preteklih seminarjev so na http://videolectures.net/solomon/

Predaval bo Primoz Skraba (trenutno na INRIA, Pariz) o
razvrscanju podatkov (angl. clustering).

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Primoz Skraba:
       Persistence-based Clustering

Clustering is a classical problem which looks for important segments
in an unstructured data set. In general, this is an ill-posed
problem. A common approach is to consider the data set as a sample of
an unknown probability distribution function on some underlying
space. Clustering then becomes a problem of understanding the
behaviour of the distribution function.

In this talk, I will introduce persistence-based clustering. Under
some mild assumptions, the algorithm comes with a variety of strong
theoretical guarantees. In particular, it provably approximates the
structure of the underlying distribution function even when underlying
space is only approximately known. The approach is based heavily on
persistent homology (also refered to as topological persistence), a
relatively recent development in the area of computational
topology. It is precisely this framework which makes many of the
proofs possible. The talk will include a general introduction to
persistence so no prior knowledge is expected. On the practical side,
the algorithm is efficient, both in memory size and running time, so
it can handle large, high dimensional data sets quickly. Finally, it
provides visual feedback in addition to the clusters, something which
is particularly useful when the data sets cannot be visualized