Ontologies to build a predictive architecture to classify and explain - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Ontologies to build a predictive architecture to classify and explain

Des ontologies pour construire une architecture prédictive, classifier et expliquer

Résumé

Explainable AI is gaining traction because of the widespread use of black box models in the industry. Many explanation methods are proposed to explain models without impacting their design. The literature describes a new architecture where an explainable model interacts with an explanation interface to generate explanations tailored for a user. We propose a novel image classification system that combines an ontology with machine learning models based on this architecture. It uses an ontology to add different labels to the same dataset and generates machine learning models to assess the class of an object and its different properties listed in the ontology. The outputs of these models are added to the ontology to verify that these predictions are consistent, using logical reasoning. The ontology can then be explored to understand the prediction and why it is consistent or not. This system can warn the user when a prediction is uncertain, which will help users to trust it.
Fichier principal
Vignette du fichier
Ontologies_to_build_a_predictive_architecture_to_classify_and_explain (3).pdf (277.41 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03684275 , version 1 (01-06-2022)

Identifiants

  • HAL Id : hal-03684275 , version 1

Citer

Matthieu Bellucci, Nicolas Delestre, Nicolas Malandain, Cecilia Zanni-Merk. Ontologies to build a predictive architecture to classify and explain. DeepOntoNLP Workshop @ESWC 2022, May 2022, Hersonissos, Greece. ⟨hal-03684275⟩
70 Consultations
251 Téléchargements

Partager

Gmail Facebook X LinkedIn More