On the stability, correctness and plausibility of visual explanation methods based on feature importance - CEA - Commissariat à l’énergie atomique et aux énergies alternatives Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2023

On the stability, correctness and plausibility of visual explanation methods based on feature importance

Résumé

In the field of Explainable AI, multiples evaluation metrics have been proposed in order to assess the quality of explanation methods w.r.t. a set of desired properties. In this work, we study the articulation between the stability, correctness and plausibility of explanations based on feature importance for image classifiers. We show that the existing metrics for evaluating these properties do not always agree, raising the issue of what constitutes a good evaluation metric for explanations. Finally, in the particular case of stability and correctness, we show the possible limitations of some evaluation metrics and propose new ones that take into account the local behaviour of the model under test.
Fichier principal
Vignette du fichier
main.pdf (662.14 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

cea-04256974 , version 1 (24-10-2023)

Identifiants

  • HAL Id : cea-04256974 , version 1

Citer

Romain Xu-Darme, Jenny Benois-Pineau, Romain Giot, Georges Quénot, Zakaria Chihani, et al.. On the stability, correctness and plausibility of visual explanation methods based on feature importance. 2023. ⟨cea-04256974⟩
135 Consultations
78 Téléchargements

Partager

Gmail Facebook X LinkedIn More