Toward optimal probabilistic active learning using a Bayesian approach

Simple item page
dc.date.accessioned2021-09-01T11:55:30Z
dc.date.available2021-09-01T11:55:30Z
dc.date.issued2021-05-04
dc.description.sponsorshipGefördert im Rahmen des Projekts DEALger
dc.identifierdoi:10.17170/kobra-202107054244
dc.identifier.urihttp://hdl.handle.net/123456789/13198
dc.language.isoengeng
dc.relation.doidoi:10.1007/s10994-021-05986-9
dc.rightsNamensnennung 4.0 International*
dc.rights.urihttp://creativecommons.org/licenses/by/4.0/*
dc.subjectactive learningeng
dc.subjectclassificationeng
dc.subjectprobabilistic active learningeng
dc.subject.ddc004
dc.subject.swdMaschinelles Lernenger
dc.subject.swdKlassifikationger
dc.subject.swdBayes-Verfahrenger
dc.subject.swdDatenger
dc.titleToward optimal probabilistic active learning using a Bayesian approacheng
dc.typeAufsatz
dc.type.versionpublishedVersion
dcterms.abstractGathering 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.eng
dcterms.accessRightsopen access
dcterms.creatorKottke, Daniel
dcterms.creatorHerde, Marek
dcterms.creatorSandrock, Christoph
dcterms.creatorHuseljic, Denis
dcterms.creatorKrempl, Georg
dcterms.creatorSick, Bernhard
dcterms.source.identifiereissn:1573-0565
dcterms.source.issueIssue 6
dcterms.source.journalMachine Learningeng
dcterms.source.pageinfo1199-1231
dcterms.source.volumeVolume 110
kup.iskupfalse

Files

Original bundle

Now showing 1 - 1 of 1
Thumbnail Image
Name:
Kottke2021_Article_TowardOptimalProbabilisticActi.pdf
Size:
5.97 MB
Format:
Adobe Portable Document Format
Description:
Download

License bundle

Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
3.03 KB
Format:
Item-specific license agreed upon to submission
Description:
Download

Collections

Artikel