HAL : hal-00336147, version 3

Fiche détaillée

Récupérer au format BibTeX - EndNote - TEI - RefWorks -

AISTATS 2009, Clearwater : États-Unis (2007)

Versions disponibles :

v1 (03-11-2008)

v2 (15-12-2008)

v3 (10-09-2009)

Chromatic PAC-Bayes Bounds for Non-IID Data

Liva Ralaivola 1, Marie Szafranski 1, Guillaume Stempfel 1

(20/04/2009)

Pac-Bayes bounds are among the most accurate generalization bounds for classifiers learned with \iid data, and it is particularly so for margin classifiers. However, there are many practical cases where the training data show some dependencies and where the traditional \iid assumption does not apply. Stating generalization bound for such frameworks is therefore of the utmost interest. In this work, we propose the first, to the best of our knowledge, \pac-Bayes generalization bounds for classifiers trained on data exhibiting dependencies. The approach is based on the decomposition of a so-called dependency graph of the data in sets of independent data, through the tool of fractional covers. Our bounds are very general, since being able to find an upper bound on the chromatic number of the dependency graph is sufficient for it get new bounds for specific settings. We show how our results can be used to derive bounds for bipartite ranking and windowed prediction.

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

Domaine

:

Informatique/Apprentissage Mathématiques/Statistiques Statistiques/Théorie

Mots Clés : Generalization bound – PAC Bayes – non IID data

hal-00336147, version 3

http://hal.archives-ouvertes.fr/hal-00336147

oai:hal.archives-ouvertes.fr:hal-00336147

Contributeur : Liva Ralaivola <>

Soumis le : Jeudi 10 Septembre 2009, 12:56:38

Dernière modification le : Jeudi 10 Septembre 2009, 12:57:38