Advanced search
1 file | 433.81 KB
Author
Organization
Abstract
In recent years several novel algorithms have been developed for maximizing the instance-wise Fβ-measure in multi-label classification problems. However, so far, such algorithms have only been tested in tandem with shallow base learners. In the deep learning landscape, usually simple thresholding approaches are implemented, even though it is expected that such approaches are suboptimal. In this article we introduce extensions of utility maximization and decision-theoretic methods that can optimize the Fβ-measure with (convolutional) neural networks. We discuss pros and cons of the different methods and we present experimental results on several image classification datasets. The results illustrate that decision-theoretic inference algorithms are worth the investment. While being more difficult to implement compared to thresholding strategies, they lead to a better predictive performance. Overall, a decision-theoretic inference algorithm based on proportional odds models outperforms the other methods.
Keywords
Fβ-measure, Bayes optimal predictions, multi-label image classification, convolutional neural networks

Downloads

  • (...).pdf
    • full text
    • |
    • UGent only
    • |
    • PDF
    • |
    • 433.81 KB

Citation

Please use this url to cite or link to this publication:

Chicago
Decubber, Stijn, Thomas Mortier, Krzysztof Dembczyński, and Willem Waegeman. 2018. “Deep F-measure Maximization in Multi-label Classification : a Comparative Study.” In Machine Learning and Knowledge Discovery in Databases : European Conference, ECML PKDD 2018 : Proceedings, Part I, ed. Michele Berlingerio, Francesco Bonchi, Thomas Gärtner, Neil Hurley, and Georgiana Ifrim, 11051:290–305. Cham, Switzerland: Springer Nature.
APA
Decubber, S., Mortier, T., Dembczyński, K., & Waegeman, W. (2018). Deep F-measure maximization in multi-label classification : a comparative study. In M. Berlingerio, F. Bonchi, T. Gärtner, N. Hurley, & G. Ifrim (Eds.), Machine learning and knowledge discovery in databases : European conference, ECML PKDD 2018 : proceedings, part I (Vol. 11051, pp. 290–305). Presented at the 2018 European conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2018), Cham, Switzerland: Springer Nature.
Vancouver
1.
Decubber S, Mortier T, Dembczyński K, Waegeman W. Deep F-measure maximization in multi-label classification : a comparative study. In: Berlingerio M, Bonchi F, Gärtner T, Hurley N, Ifrim G, editors. Machine learning and knowledge discovery in databases : European conference, ECML PKDD 2018 : proceedings, part I. Cham, Switzerland: Springer Nature; 2018. p. 290–305.
MLA
Decubber, Stijn, Thomas Mortier, Krzysztof Dembczyński, et al. “Deep F-measure Maximization in Multi-label Classification : a Comparative Study.” Machine Learning and Knowledge Discovery in Databases : European Conference, ECML PKDD 2018 : Proceedings, Part I. Ed. Michele Berlingerio et al. Vol. 11051. Cham, Switzerland: Springer Nature, 2018. 290–305. Print.
@inproceedings{8574368,
  abstract     = {In recent years several novel algorithms have been developed for maximizing the instance-wise F\ensuremath{\beta}-measure in multi-label classification problems. However, so far, such algorithms have only been tested in tandem with shallow base learners. In the deep learning landscape, usually simple thresholding approaches are implemented, even though it is expected that such approaches are suboptimal. In this article we introduce extensions of utility maximization and decision-theoretic methods that can optimize the F\ensuremath{\beta}-measure with (convolutional) neural networks. We discuss pros and cons of the different methods and we present experimental results on several image classification datasets. The results illustrate that decision-theoretic inference algorithms are worth the investment. While being more difficult to implement compared to thresholding strategies, they lead to a better predictive performance. Overall, a decision-theoretic inference algorithm based on proportional odds models outperforms the other methods.},
  author       = {Decubber, Stijn and Mortier, Thomas and Dembczy\'{n}ski, Krzysztof and Waegeman, Willem},
  booktitle    = {Machine learning and knowledge discovery in databases : European conference, ECML PKDD 2018 : proceedings, part I},
  editor       = {Berlingerio, Michele and Bonchi, Francesco and G{\"a}rtner, Thomas and Hurley, Neil and Ifrim, Georgiana},
  isbn         = {9783030109240},
  issn         = {0302-9743},
  language     = {eng},
  location     = {Dublin, Ireland},
  pages        = {290--305},
  publisher    = {Springer Nature},
  title        = {Deep F-measure maximization in multi-label classification : a comparative study},
  url          = {http://dx.doi.org/10.1007/978-3-030-10925-7\_18},
  volume       = {11051},
  year         = {2018},
}

Altmetric
View in Altmetric