Assessment explanation approaches for black box models / Khaled Bouabdallah  

Public ISBD Titre : Assessment explanation approaches for black box models Type de document : texte imprimé Auteurs : Khaled Bouabdallah, Auteur ; Seddik Boudissa, Auteur ; Ahlem Drif, Directeur de thèse Année de publication : 2022 Importance : 1 vol (79 f .) Format : 29cm Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Artificial Intelligence Explainable Artificial Intelligence Index. décimale : 004 Informatique Résumé : Machine learning models, particularly deep learning models, are frequently referred to as ”black box models” due to their enormous complexity. Because of their lack of open- ness, they are difficult to understand and trust. To address this issue and create models with results that humans can understand, a subfield of artificial intelligence (AI) known as Ex- plainable Artificial Intelligence (XAI) was created. This thesis aims to provide an in-depth examination of XAI techniques as well as propose new approaches. We propose the Dis- criminitve Attention Guided Convolutional Neural Network (DAG-CNN), which focuses on lesion regions when making diagnoses and compares similarities between cases for thorax disease classification. This approach showed promising results with competitive accuracy and more explanation in comparison to other state-of-the-art solutions. We also proposed a grey-box model based on semi-supervised methodology utilizing a self-labeling framework that combines the advantages of both black-box and white-box models. The proposed self- labeling Grey-Box shows an accurate black-box classifier for labeling the unlabeled data and a white-box surrogate classifier for building an interpretable model. Côte titre : MAI/0575 En ligne : https://drive.google.com/file/d/1Qf6_zEg10-4tItJ2LU3ek8BrGRtkZE6M/view?usp=share [...] Format de la ressource électronique : pdf Assessment explanation approaches for black box models [texte imprimé] / Khaled Bouabdallah, Auteur ; Seddik Boudissa, Auteur ; Ahlem Drif, Directeur de thèse . - 2022 . - 1 vol (79 f .) ; 29cm. Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Artificial Intelligence Explainable Artificial Intelligence Index. décimale : 004 Informatique Résumé : Machine learning models, particularly deep learning models, are frequently referred to as ”black box models” due to their enormous complexity. Because of their lack of open- ness, they are difficult to understand and trust. To address this issue and create models with results that humans can understand, a subfield of artificial intelligence (AI) known as Ex- plainable Artificial Intelligence (XAI) was created. This thesis aims to provide an in-depth examination of XAI techniques as well as propose new approaches. We propose the Dis- criminitve Attention Guided Convolutional Neural Network (DAG-CNN), which focuses on lesion regions when making diagnoses and compares similarities between cases for thorax disease classification. This approach showed promising results with competitive accuracy and more explanation in comparison to other state-of-the-art solutions. We also proposed a grey-box model based on semi-supervised methodology utilizing a self-labeling framework that combines the advantages of both black-box and white-box models. The proposed self- labeling Grey-Box shows an accurate black-box classifier for labeling the unlabeled data and a white-box surrogate classifier for building an interpretable model. Côte titre : MAI/0575 En ligne : https://drive.google.com/file/d/1Qf6_zEg10-4tItJ2LU3ek8BrGRtkZE6M/view?usp=share [...] Format de la ressource électronique : pdf

Exemplaires (1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
MAI/0575	MAI/0575	Mémoire	Bibliothéque des sciences	Anglais	Disponible Disponible

---

1 (1 - 1 / 1)