direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Analysis of Neural Data

All Publications

Target Selection: A New Learning Paradigm and its Application to Genetic Association Studies
Citation key Mohr2008b
Author Mohr, J. and Seo, S. and Puls, I. and Heinz, A. and Obermayer, K.
Title of Book Proceedings of the ICMLA '08: The Seventh International Conference on Machine Learning and Applications
Pages 182 – 187
Year 2008
ISBN 978-0-7695-3495-4
DOI 10.1109/ICMLA.2008.58
Publisher IEEE
Abstract In classification problems, the task of a learning machine is usually to select a mapping between several input variables and one or more target variables (each taking values from a set of class labels), such that the expected loss (the risk) is minimized. Often, the technique of feature selection is employed to chose a subset of the input variables which are used in the classification function, based on the assumption that some of the input variables are irrelevant for predicting the target. In this paper a new learning paradigm called target selection is introduced, where the task of a learning machine is to select both one out of several target variables and a function which maps the input variables to the selected target. We propose a cost function for target selection based on mutual information and suggest an algorithm for its optimization. Here, the new paradigm and the proposed algorithm for target selection are applied to genotype-phenotype association analysis.
Link to original publication Download Bibtex entry

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions