Titre :	Application of Adversarial Attacks on Machine Learning Systems
Type de document :	texte imprimé
Auteurs :	Sara Israa Zerroug, Auteur ; Amani Maissoune Mellas ; Yasmine Harbi, Directeur de thèse
Editeur :	Setif:UFA
Année de publication :	2024
Importance :	1 vol (65 f .)
Format :	29 cm
Langues :	Anglais (eng)
Catégories :	Thèses & Mémoires:Informatique
Mots-clés :	Artificial Intelligence (AI) Machine Learning (ML) Adversarial Machine Learning (AML) Fast Gradient Sign Method (FGSM)
Index. décimale :	004 - Informatique
Résumé :	Machine Learning (ML), a subset of Artificial Intelligence (AI), have revolutionized various domains by enabling machines to learn from data and make intelligent decisions autonomously. However, the integration of ML into critical applications, such as healthcare, raises concerns about security and robustness. This study investigates the intersection of ML and security through a deep learning-based brain tumor classification model. We develop a model capable of accurately classifying brain tumor types. Furthermore, we explore the vulnerability of the model to adversarial attacks, employing techniques such as Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD). Our experiments encompass comprehensive evaluations of model performance metrics, including accuracy, precision, recall, FPR and F1-score. Through meticulous analysis, we uncover insights into the impact of perturbation strength and subset size on the model’s susceptibility to attacks. Our findings underscore the importance of robust defense mechanisms in mitigating adversarial threats in ML-driven systems. By elucidating the challenges and opportunities in brain tumor classification using ML techniques, this study contributes to the broader understanding of AI’s role in healthcare and the imperative of ensuring security in ML applications.
Note de contenu :	Sommaire Table of Contents ii List of Figures v List of Tables vii Table of Abbreviations viii General Introduction 1 1 State-of-the-art 3 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2 Overview of Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.1 Types of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.2 Types of Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.2.3 Most Popular Algorithms . . . . . . . . . . . . . . . . . . . . . . . 6 1.2.4 ML Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.2.5 ML Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.3 Adversarial Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.2 Adversarial ML Model . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.3 Adversarial Attacks Types . . . . . . . . . . . . . . . . . . . . . . . 15 1.3.4 Adversarial Attacks Methods . . . . . . . . . . . . . . . . . . . . . 18 1.4 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2 Contribution 22 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2 Problem Statement, Objectives, and Hypothesis . . . . . . . . . . . . . . 22 2.3 Proposed DL-based Healthcare Model . . . . . . . . . . . . . . . . . . . . 23 2.3.1 Dataset Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.3.2 Dataset Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.3 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.4 Medical Image Classification . . . . . . . . . . . . . . . . . . . . . . 26 2.4 Proposed Adversarial DL-based Healthcare Model . . . . . . . . . . . . . 28 2.4.1 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.4.2 Evasion Attack Steps . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3 Experiments and Results 35 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.2 Hardware and Software Environments . . . . . . . . . . . . . . . . . . . . 35 3.2.1 Experiment Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.3 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.4.1 Proposed DL-based Healthcare Model . . . . . . . . . . . . . . . . 39 3.4.2 Proposed Adversarial DL-based Healthcare Model . . . . . . . . . 42 3.5 Comparison to Related Works . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Côte titre :	MAI/0887

Application of Adversarial Attacks on Machine Learning Systems [texte imprimé] / Sara Israa Zerroug, Auteur ; Amani Maissoune Mellas ; Yasmine Harbi, Directeur de thèse . - [S.l.] : Setif:UFA, 2024 . - 1 vol (65 f .) ; 29 cm.
Langues : Anglais (eng)

Catégories :	Thèses & Mémoires:Informatique
Mots-clés :	Artificial Intelligence (AI) Machine Learning (ML) Adversarial Machine Learning (AML) Fast Gradient Sign Method (FGSM)
Index. décimale :	004 - Informatique
Résumé :	Machine Learning (ML), a subset of Artificial Intelligence (AI), have revolutionized various domains by enabling machines to learn from data and make intelligent decisions autonomously. However, the integration of ML into critical applications, such as healthcare, raises concerns about security and robustness. This study investigates the intersection of ML and security through a deep learning-based brain tumor classification model. We develop a model capable of accurately classifying brain tumor types. Furthermore, we explore the vulnerability of the model to adversarial attacks, employing techniques such as Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD). Our experiments encompass comprehensive evaluations of model performance metrics, including accuracy, precision, recall, FPR and F1-score. Through meticulous analysis, we uncover insights into the impact of perturbation strength and subset size on the model’s susceptibility to attacks. Our findings underscore the importance of robust defense mechanisms in mitigating adversarial threats in ML-driven systems. By elucidating the challenges and opportunities in brain tumor classification using ML techniques, this study contributes to the broader understanding of AI’s role in healthcare and the imperative of ensuring security in ML applications.
Note de contenu :	Sommaire Table of Contents ii List of Figures v List of Tables vii Table of Abbreviations viii General Introduction 1 1 State-of-the-art 3 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2 Overview of Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.1 Types of Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.2 Types of Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.2.3 Most Popular Algorithms . . . . . . . . . . . . . . . . . . . . . . . 6 1.2.4 ML Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.2.5 ML Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.3 Adversarial Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3.2 Adversarial ML Model . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.3 Adversarial Attacks Types . . . . . . . . . . . . . . . . . . . . . . . 15 1.3.4 Adversarial Attacks Methods . . . . . . . . . . . . . . . . . . . . . 18 1.4 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2 Contribution 22 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2 Problem Statement, Objectives, and Hypothesis . . . . . . . . . . . . . . 22 2.3 Proposed DL-based Healthcare Model . . . . . . . . . . . . . . . . . . . . 23 2.3.1 Dataset Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.3.2 Dataset Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.3 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.3.4 Medical Image Classification . . . . . . . . . . . . . . . . . . . . . . 26 2.4 Proposed Adversarial DL-based Healthcare Model . . . . . . . . . . . . . 28 2.4.1 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.4.2 Evasion Attack Steps . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3 Experiments and Results 35 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.2 Hardware and Software Environments . . . . . . . . . . . . . . . . . . . . 35 3.2.1 Experiment Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.3 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.4.1 Proposed DL-based Healthcare Model . . . . . . . . . . . . . . . . 39 3.4.2 Proposed Adversarial DL-based Healthcare Model . . . . . . . . . 42 3.5 Comparison to Related Works . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Côte titre :	MAI/0887