Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data

Abstract : Hyper-parameter tuning is a resource-intensive task when optimizing classification models. The commonly used k-fold cross validation can become intractable in large scale settings when a classifier has to learn billions of parameters. At the same time, in real-world, one often encounters multi-class classification scenarios with only a few labeled examples; model selection approaches often offer little improvement in such cases and the default values of learners are used. We propose bounds for classification on accuracy and macro measures (precision, recall, F1) that motivate efficient schemes for model selection and can benefit from the existence of unlabeled data. We demonstrate the advantages of those schemes by comparing them with k-fold cross validation and hold-out estimation in the setting of large scale classification.
Type de document :
Communication dans un congrès
14th International Symposium on Intelligent Data Analysis, IDA, Oct 2015, Saint-Etienne, France. 2015, 〈10.1007/978-3-319-24465-5_3〉
Liste complète des métadonnées

Littérature citée [15 références]  Voir  Masquer  Télécharger

https://hal.archives-ouvertes.fr/hal-01237129
Contributeur : Georgios Balikas <>
Soumis le : mercredi 2 décembre 2015 - 17:11:56
Dernière modification le : jeudi 11 octobre 2018 - 08:48:04
Document(s) archivé(s) le : samedi 29 avril 2017 - 05:11:58

Fichier

Quantification_IDA15.pdf
Fichiers produits par l'(les) auteur(s)

Identifiants

Collections

Citation

Georgios Balikas, Ioannis Partalas, Eric Gaussier, Rohit Babbar, Massih-Reza Amini. Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data. 14th International Symposium on Intelligent Data Analysis, IDA, Oct 2015, Saint-Etienne, France. 2015, 〈10.1007/978-3-319-24465-5_3〉. 〈hal-01237129〉

Partager

Métriques

Consultations de la notice

307

Téléchargements de fichiers

130