Datum
2021-05-04Metadata
Zur Langanzeige
Aufsatz
Toward optimal probabilistic active learning using a Bayesian approach
Zusammenfassung
Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active learning aims at reducing the labeling costs by an efficient and effective allocation of costly labeling resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes the gain in misclassification error, and (2) uses a Bayesian approach by introducing a conjugate prior distribution to determine the class posterior to deal with uncertainties. By reformulating existing selection strategies within our proposed model, we can explain which aspects are not covered in current state-of-the-art and why this leads to the superior performance of our approach. Extensive experiments on a large variety of datasets and different kernels validate our claims.
Zitierform
In: Machine Learning Volume 110 / Issue 6 (2021-05-04) , S. 1199-1231 ; eissn:1573-0565Förderhinweis
Gefördert im Rahmen des Projekts DEALZitieren
@article{doi:10.17170/kobra-202107054244,
author={Kottke, Daniel and Herde, Marek and Sandrock, Christoph and Huseljic, Denis and Krempl, Georg and Sick, Bernhard},
title={Toward optimal probabilistic active learning using a Bayesian approach},
journal={Machine Learning},
year={2021}
}
0500 Oax 0501 Text $btxt$2rdacontent 0502 Computermedien $bc$2rdacarrier 1100 2021$n2021 1500 1/eng 2050 ##0##http://hdl.handle.net/123456789/13198 3000 Kottke, Daniel 3010 Herde, Marek 3010 Sandrock, Christoph 3010 Huseljic, Denis 3010 Krempl, Georg 3010 Sick, Bernhard 4000 Toward optimal probabilistic active learning using a Bayesian approach / Kottke, Daniel 4030 4060 Online-Ressource 4085 ##0##=u http://nbn-resolving.de/http://hdl.handle.net/123456789/13198=x R 4204 \$dAufsatz 4170 5550 {{Maschinelles Lernen}} 5550 {{Klassifikation}} 5550 {{Bayes-Verfahren}} 5550 {{Daten}} 7136 ##0##http://hdl.handle.net/123456789/13198
2021-09-01T11:55:30Z 2021-09-01T11:55:30Z 2021-05-04 doi:10.17170/kobra-202107054244 http://hdl.handle.net/123456789/13198 Gefördert im Rahmen des Projekts DEAL eng Namensnennung 4.0 International http://creativecommons.org/licenses/by/4.0/ active learning classification probabilistic active learning 004 Toward optimal probabilistic active learning using a Bayesian approach Aufsatz Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active learning aims at reducing the labeling costs by an efficient and effective allocation of costly labeling resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes the gain in misclassification error, and (2) uses a Bayesian approach by introducing a conjugate prior distribution to determine the class posterior to deal with uncertainties. By reformulating existing selection strategies within our proposed model, we can explain which aspects are not covered in current state-of-the-art and why this leads to the superior performance of our approach. Extensive experiments on a large variety of datasets and different kernels validate our claims. open access Kottke, Daniel Herde, Marek Sandrock, Christoph Huseljic, Denis Krempl, Georg Sick, Bernhard doi:10.1007/s10994-021-05986-9 Maschinelles Lernen Klassifikation Bayes-Verfahren Daten publishedVersion eissn:1573-0565 Issue 6 Machine Learning 1199-1231 Volume 110 false
Die folgenden Lizenzbestimmungen sind mit dieser Ressource verbunden: