IDEAS home Printed from https://ideas.repec.org/p/cir/cirwor/2004s-28.html
   My bibliography  Save this paper

Learning from Partial Labels with Minimum Entropy

Author

Listed:
  • Yves Grandvalet
  • Yoshua Bengio

Abstract

This paper introduces the minimum entropy regularizer for learning from partial labels. This learning problem encompasses the semi-supervised setting, where a decision rule is to be learned from labeled and unlabeled examples. The minimum entropy regularizer applies to diagnosis models, i.e. models of the posterior probabilities of classes. It is shown to include other approaches to the semi-supervised problem as particular or limiting cases. A series of experiments illustrates that the proposed criterion provides solutions taking advantage of unlabeled examples when the latter convey information. Even when the data are sampled from the distribution class spanned by a generative model, the proposed approach improves over the estimated generative model when the number of features is of the order of sample size. The performances are definitely in favor of minimum entropy when the generative model is slightly misspecified. Finally, the robustness of the learning scheme is demonstrated: in situations where unlabeled examples do not convey information, minimum entropy returns a solution discarding unlabeled examples and performs as well as supervised learning. Cet article introduit le régularisateur à entropie minimum pour l'apprentissage d'étiquettes partielles. Ce problème d'apprentissage incorpore le cadre non supervisé, où une règle de décision doit être apprise à partir d'exemples étiquetés et non étiquetés. Le régularisateur à entropie minimum s'applique aux modèles de diagnostics, c'est-à-dire aux modèles des probabilités postérieures de classes. Nous montrons comment inclure d'autres approches comme un cas particulier ou limité du problème semi-supervisé. Une série d'expériences montrent que le critère proposé fournit des solutions utilisant les exemples non étiquetés lorsque ces dernières sont instructives. Même lorsque les données sont échantillonnées à partir de la classe de distribution balayée par un modèle génératif, l'approche mentionnée améliore le modèle génératif estimé lorsque le nombre de caractéristiques est de l'ordre de la taille de l'échantillon. Les performances avantagent certainement l'entropie minimum lorsque le modèle génératif est légèrement mal spécifié. Finalement, la robustesse de ce cadre d'apprentissage est démontré : lors de situations où les exemples non étiquetés n'apportent aucune information, l'entropie minimum retourne une solution rejetant les exemples non étiquetés et est aussi performante que l'apprentissage supervisé.

Suggested Citation

  • Yves Grandvalet & Yoshua Bengio, 2004. "Learning from Partial Labels with Minimum Entropy," CIRANO Working Papers 2004s-28, CIRANO.
  • Handle: RePEc:cir:cirwor:2004s-28
    as

    Download full text from publisher

    File URL: https://cirano.qc.ca/files/publications/2004s-28.pdf
    Download Restriction: no
    ---><---

    References listed on IDEAS

    as
    1. Celeux, Gilles & Govaert, Gerard, 1992. "A classification EM algorithm for clustering and two stochastic versions," Computational Statistics & Data Analysis, Elsevier, vol. 14(3), pages 315-332, October.
    Full references (including those not matched with items on IDEAS)

    Most related items

    These are the items that most often cite the same works as this one and are cited by the same works as this one.
    1. Adrian O’Hagan & Arthur White, 2019. "Improved model-based clustering performance using Bayesian initialization averaging," Computational Statistics, Springer, vol. 34(1), pages 201-231, March.
    2. François Bavaud, 2009. "Aggregation invariance in general clustering approaches," Advances in Data Analysis and Classification, Springer;German Classification Society - Gesellschaft für Klassifikation (GfKl);Japanese Classification Society (JCS);Classification and Data Analysis Group of the Italian Statistical Society (CLADAG);International Federation of Classification Societies (IFCS), vol. 3(3), pages 205-225, December.
    3. Rasmus Lentz & Jean Marc Robin & Suphanit Piyapromdee, 2018. "On Worker and Firm Heterogeneity in Wages and Employment Mobility: Evidence from Danish Register Data," 2018 Meeting Papers 469, Society for Economic Dynamics.
    4. Faicel Chamroukhi, 2016. "Piecewise Regression Mixture for Simultaneous Functional Data Clustering and Optimal Segmentation," Journal of Classification, Springer;The Classification Society, vol. 33(3), pages 374-411, October.
    5. Mukhopadhyay, Subhadeep & Ghosh, Anil K., 2011. "Bayesian multiscale smoothing in supervised and semi-supervised kernel discriminant analysis," Computational Statistics & Data Analysis, Elsevier, vol. 55(7), pages 2344-2353, July.
    6. Grün, Bettina & Leisch, Friedrich, 2008. "FlexMix Version 2: Finite Mixtures with Concomitant Variables and Varying and Constant Parameters," Journal of Statistical Software, Foundation for Open Access Statistics, vol. 28(i04).
    7. Hornik, Kurt & Grün, Bettina, 2014. "movMF: An R Package for Fitting Mixtures of von Mises-Fisher Distributions," Journal of Statistical Software, Foundation for Open Access Statistics, vol. 58(i10).
    8. M. Vrac & L. Billard & E. Diday & A. Chédin, 2012. "Copula analysis of mixture models," Computational Statistics, Springer, vol. 27(3), pages 427-457, September.
    9. Bouveyron, C. & Girard, S. & Schmid, C., 2007. "High-dimensional data clustering," Computational Statistics & Data Analysis, Elsevier, vol. 52(1), pages 502-519, September.
    10. repec:jss:jstsof:28:i04 is not listed on IDEAS
    11. Chehade, Abdallah & Savargaonkar, Mayuresh & Krivtsov, Vasiliy, 2022. "Conditional Gaussian mixture model for warranty claims forecasting," Reliability Engineering and System Safety, Elsevier, vol. 218(PB).
    12. Nicolas Jouvin & Pierre Latouche & Charles Bouveyron & Guillaume Bataillon & Alain Livartowski, 2021. "Greedy clustering of count data through a mixture of multinomial PCA," Computational Statistics, Springer, vol. 36(1), pages 1-33, March.
    13. García-Escudero, L.A. & Gordaliza, A. & Mayo-Iscar, A. & San Martín, R., 2010. "Robust clusterwise linear regression through trimming," Computational Statistics & Data Analysis, Elsevier, vol. 54(12), pages 3057-3069, December.
    14. Same, Allou & Ambroise, Christophe & Govaert, Gerard, 2006. "A classification EM algorithm for binned data," Computational Statistics & Data Analysis, Elsevier, vol. 51(2), pages 466-480, November.
    15. Frei, Marco & Künsch, Hans R., 2013. "Mixture ensemble Kalman filters," Computational Statistics & Data Analysis, Elsevier, vol. 58(C), pages 127-138.
    16. Guan-Hua Huang & Su-Mei Wang & Chung-Chu Hsu, 2011. "Optimization-Based Model Fitting for Latent Class and Latent Profile Analyses," Psychometrika, Springer;The Psychometric Society, vol. 76(4), pages 584-611, October.
    17. Murphy, Thomas Brendan & Martin, Donal, 2003. "Mixtures of distance-based models for ranking data," Computational Statistics & Data Analysis, Elsevier, vol. 41(3-4), pages 645-655, January.
    18. Rasmus Lentz & Suphanit Piyapromdee & Jean-Marc Robin, 2022. "The Anatomy of Sorting - Evidence from Danish Data," Working Papers hal-03869383, HAL.
    19. Keiji Takai, 2012. "Constrained EM algorithm with projection method," Computational Statistics, Springer, vol. 27(4), pages 701-714, December.
    20. Volodymyr Melnykov & Xuwen Zhu, 2019. "An extension of the K-means algorithm to clustering skewed data," Computational Statistics, Springer, vol. 34(1), pages 373-394, March.
    21. Bouveyron, Charles & Brunet, Camille, 2012. "Theoretical and practical considerations on the convergence properties of the Fisher-EM algorithm," Journal of Multivariate Analysis, Elsevier, vol. 109(C), pages 29-41.

    Corrections

    All material on this site has been provided by the respective publishers and authors. You can help correct errors and omissions. When requesting a correction, please mention this item's handle: RePEc:cir:cirwor:2004s-28. See general information about how to correct material in RePEc.

    If you have authored this item and are not yet registered with RePEc, we encourage you to do it here. This allows to link your profile to this item. It also allows you to accept potential citations to this item that we are uncertain about.

    If CitEc recognized a bibliographic reference but did not link an item in RePEc to it, you can help with this form .

    If you know of missing items citing this one, you can help us creating those links by adding the relevant references in the same way as above, for each refering item. If you are a registered author of this item, you may also want to check the "citations" tab in your RePEc Author Service profile, as there may be some citations waiting for confirmation.

    For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: Webmaster (email available below). General contact details of provider: https://edirc.repec.org/data/ciranca.html .

    Please note that corrections may take a couple of weeks to filter through the various RePEc services.

    IDEAS is a RePEc service. RePEc uses bibliographic data supplied by the respective publishers.