Revisiting Machine Learning from Crowds a Mixture Model for Grouping Annotations

Francisco Mena, Ricardo Ñanculef

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

© Springer Nature Switzerland AG 2019. Today, supervised learning is widely used for pattern recognition, computer vision and other tasks. In this setting, data need to be explicitly annotated. Unfortunately, obtaining accurate labels can be difficult, expensive and time-consuming. As a result, many machine learning projects rely on labelling processes that involve crowds, i.e. multiple subjective and inexpert annotators. Handling this noise in a principled way is an important challenge for machine learning, called learning from crowds. In this paper, we present a model that learns patterns of label noise by grouping annotations. In contrast to previous art, we do not model specific labeling patterns for each annotator but explain the data using a fixed-size mixture model. This approach allows to handle a sparse distribution of labels among annotators and obtain a model with less parameters that can scale better to large-scale scenarios. Experiments on real and simulated data illustrate the advantages of our approach.
Original languageEnglish
Title of host publicationRevisiting Machine Learning from Crowds a Mixture Model for Grouping Annotations
Pages493-503
Number of pages11
ISBN (Electronic)9783030339036
DOIs
Publication statusPublished - 1 Jan 2019
EventLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) -
Duration: 1 Jan 2019 → …

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11896 LNCS
ISSN (Print)0302-9743

Conference

ConferenceLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Period1/01/19 → …

Fingerprint Dive into the research topics of 'Revisiting Machine Learning from Crowds a Mixture Model for Grouping Annotations'. Together they form a unique fingerprint.

  • Cite this

    Mena, F., & Ñanculef, R. (2019). Revisiting Machine Learning from Crowds a Mixture Model for Grouping Annotations. In Revisiting Machine Learning from Crowds a Mixture Model for Grouping Annotations (pp. 493-503). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11896 LNCS). https://doi.org/10.1007/978-3-030-33904-3_46