Prof. Emo Welzl and Prof. Bernd Gärtner
|Mittagsseminar Talk Information|
Date and Time: Thursday, January 28, 2010, 12:15 pm
Duration: This information is not available in the database
Location: CAB G51
Speaker: Johannes Blömer (Paderborn University)
We consider the k-median problem, i.e. the problem of partitioning a set of n points into k disjoint sets called clusters, such that the sum of the distances of each point to the center of gravity of its cluster is minimized. To measure the distance between points we use so-called Bregman divergences which include among other distance measures the squared Euclidean distance and the Kulback-Leibler divergence from information theory. We show that a non-uniform sampling strategy analogous to the k-means++ method from Arthur and Vassilvitskii computes a constant factor approximation for well-separated instances of the k-median clustering for Bregman divergences.
Automatic MiSe System Software Version 1.4803M | admin login