Prof. Emo Welzl and Prof. Bernd Gärtner
|Mittagsseminar Talk Information|
Date and Time: Tuesday, September 20, 2005, 12:15 pm
Duration: This information is not available in the database
Location: This information is not available in the database
Speaker: Volker Roth
In supervised learning scenarios, feature selection has been studied widely in the literature. Selecting features in unsupervised partitioning scenarios, however, is a much harder problem, due to the absence of class labels that would guide the search for relevant information. Most approaches to this combined partitioning and extraction problem suffer both from conceptual shortcomings and ambiguous solutions. In this work I propose to overcome these problems by combining a Gaussian mixture model with a Bayesian relevance determination principle. This interleaved handling of clustering and feature selection makes it possible to formulate both aspects as a joint optimization problem that involves only one single objective function. Applications in the fields of Computer Vision and Bioinformatics show that the proposed approach is capable of simultaneously finding meaningful partitions and extracting relevant features or influence factors.
Automatic MiSe System Software Version 1.4803M | admin login