Fast computation of $L^p$ norm-based specialization distances between bodies of evidence - LAGIS-SI
Communication Dans Un Congrès Année : 2014

Fast computation of $L^p$ norm-based specialization distances between bodies of evidence

Mehena Loudahi
  • Fonction : Auteur
John Klein
  • Fonction : Auteur
  • PersonId : 753846
  • IdHAL : john-klein
Jean-Marc Vannobel
Olivier Colot

Résumé

In a recent paper [1], we introduced a new family of evidential distances in the framework of belief functions. Using specialization matrices as a representation of bodies of evidence, an evidential distance can be obtained by computing the norm of the difference of these matrices. Any matrix norm can be thus used to define a full metric. In particular, it has been shown that the $L^1$ norm-based specialization distance has nice properties. This distance takes into account the structure of focal elements and has a consistent behavior with respect to the conjunctive combination rule. However, if the frame of discernment on which the problem is defined has $n$ elements, then a specialization matrix size is $2^n \times 2^n$. The straightforward formula for computing a specialization distance involves a matrix product which can be consequently highly time consuming. In this article, several faster computation methods are provided for $L^p$ norm-based specialization distances. These methods are proposed for special kinds of mass functions as well as for the general case.
Fichier principal
Vignette du fichier
art_belief2014_HAL.pdf (76.38 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01015330 , version 1 (26-06-2014)

Identifiants

Citer

Mehena Loudahi, John Klein, Jean-Marc Vannobel, Olivier Colot. Fast computation of $L^p$ norm-based specialization distances between bodies of evidence. thrid international conference on belief functions, Sep 2014, Oxford, United Kingdom. pp.422-431, ⟨10.1007/978-3-319-11191-9_46⟩. ⟨hal-01015330⟩
193 Consultations
266 Téléchargements

Altmetric

Partager

More