Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data

Georgios Balikas; Ioannis Partalas; Eric Gaussier; Rohit Babbar; Massih-Reza Amini

Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data

Georgios Balikas ¹ Ioannis Partalas ² Eric Gaussier ¹ Rohit Babbar Massih-Reza Amini ¹

1 AMA - Analyse de données, Modélisation et Apprentissage automatique [Grenoble]

LIG - Laboratoire d'Informatique de Grenoble

2 VISEO

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.

Keywords : Multi-class classification Quantification

Type de document :

Communication dans un congrès

Fourteenth International Symposium on Intelligent Data Analysis (IDA 2015), Oct 2015, Saint-Etienne, France. 〈https://ida2015.univ-st-etienne.fr/〉

Domaine :

Informatique [cs] / Apprentissage [cs.LG]

Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-01236587
Contributeur : Massih-Reza Amini <>
Soumis le : mardi 1 décembre 2015 - 21:48:35
Dernière modification le : jeudi 11 octobre 2018 - 08:48:04

Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data

Identifiants

Collections

Citation

Exporter

Métriques

110

Contact

Efficient Model Selection for Regularized Classification by Exploiting Unlabeled Data

Identifiants

Collections

Citation

Exporter

Partager

Métriques

110

Contact