HAL : hal-00674847, version 4

arXiv : 1202.6228

Fiche détaillée

Récupérer au format BibTeX - EndNote - TEI - RefWorks -

International Conference on Machine Learning 2012, Edinburgh : Royaume-Uni (2012)

Versions disponibles :

v1 (28-02-2012)

v2 (22-03-2012)

v3 (24-05-2012)

v4 (30-06-2012)

v5 (04-07-2012)

PAC-Bayesian Generalization Bound on Confusion Matrix for Multi-Class Classification

Emilie Morvant (  Auteur correspondant ) 1, Sokol Koço (  Auteur correspondant ) 1, Liva Ralaivola (  Auteur correspondant ) 1

(2012)

In this work, we propose a PAC-Bayes bound for the generalization risk of the Gibbs classifier in the multi-class classification framework. The novelty of our work is the critical use of the confusion matrix of a classifier as an error measure; this puts our contribution in the line of work aiming at dealing with performance measure that are richer than mere scalar criterion such as the misclassification rate. Thanks to very recent and beautiful results on matrix concentration inequalities, we derive two bounds showing that the true confusion risk of the Gibbs classifier is upper-bounded by its empirical risk plus a term depending on the number of training examples in each class. To the best of our knowledge, this is the first PAC-Bayes bounds based on confusion matrices.

1 :	Laboratoire d'informatique Fondamentale de Marseille (LIF)
	CNRS : UMR6166 – Université de la Méditerranée - Aix-Marseille II – Université de Provence - Aix-Marseille I

Équipe de recherche : QARMA

Domaine

:

Statistiques/Machine Learning

Mots Clés : PAC-Bayes generalization bounds – Confusion Matrix – Concentration Inequality – Multi-Class Classification

hal-00674847, version 4

http://hal.archives-ouvertes.fr/hal-00674847

oai:hal.archives-ouvertes.fr:hal-00674847

Contributeur : Emilie Morvant <>

Soumis le : Samedi 30 Juin 2012, 15:32:13

Dernière modification le : Samedi 30 Juin 2012, 21:23:19