PAC-Bayesian Theorems for Multiview Learning

Anil Goyal 1, 2 Emilie Morvant 1 Pascal Germain 3 Massih-Reza Amini 2
3 SIERRA - Statistical Machine Learning and Parsimony
DI-ENS - Département d'informatique de l'École normale supérieure, ENS Paris - École normale supérieure - Paris, CNRS - Centre National de la Recherche Scientifique, Inria de Paris
Abstract : We tackle the issue of multiview learning which aims to take advantages of multiple represen-tations/views of the data. In this context, many machine learning algorithms exist. However, the majority of the theoretical studies focus on learning with exactly two representations. In this paper, we propose a general PAC-Bayesian theory for multiview learning with more than two views. We focus our study to binary classification models that take the form of a majority vote. We derive PAC-Bayesian generalization bounds allowing to consider different relations between empirical and true risks by taking into account a notion of diversity of the voters and views, and that can be naturally extended to semi-supervised learning.
Type de document :
Pré-publication, Document de travail
2016
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-01336260
Contributeur : Anil Goyal <>
Soumis le : jeudi 17 novembre 2016 - 17:14:14
Dernière modification le : dimanche 20 novembre 2016 - 01:02:05

Fichiers

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

Identifiants

  • HAL Id : hal-01336260, version 2
  • ARXIV : 1606.07240

Collections

Citation

Anil Goyal, Emilie Morvant, Pascal Germain, Massih-Reza Amini. PAC-Bayesian Theorems for Multiview Learning. 2016. <hal-01336260v2>

Partager

Métriques

Consultations de
la notice

115

Téléchargements du document

49