Browse
Recent Submissions
Item conception et r ´ ealisation d’un ordonnancement des taches dans un syst ` eme embarqu ´ e bas ´ e sur la tol ´ erance aux fautes(2024) Saiad Zineb Saidi AsmaLa conception et la mise en œuvre d’un algorithme d’ordonnancement bas ´ e sur la tol ´ erance aux fautes pour les syst ` emes embarqu ´ es sont des aspects cruciaux dans divers secteurs tel que l’automobile, l’a ´ erospatial, la sant ´ e . . ...etc. cela contribue ` a am ´ eliorer la fiabilit ´ e et la s ´ ecurit ´ e des syst ` emes embarqu ´ e ce qui est essentiel pour assurer le bon fonctionnement de syst ` eme. Ce m ´ emoire traite le probl ` eme d’ordonnancement m ˆ eme en pr ´ esence des fautes ce qui est l’objectif de notre travail. En utilisant le Watch dog Timer pour d ´ etecter les fautes de processeur et en adoptant l’algorithme EDF *(Earliest Deadline First*) EDF*.Item Improving Code Quality based on AI(2024) Messaoudi ChouaibThis thesis addresses the development of a LSTM model for correcting erroneous Python code snippets. The methodology comprises several key steps. First, a carefully curated dataset containing 450 incorrect Python codes with corresponding corrections was constructed, covering a diverse range of code patterns and functionalities. The dataset, structured as a CSV file, encapsulates various aspects of Python programming, including basic function definitions, mathematical calculations, and data processing tasks such as string manipulation (Dataset Construction). Next, the dataset underwent meticulous data processing, involving tokenization, padding, and splitting into training and testing sets to prepare it for model training (Data Processing). The model architecture, following an encoder-decoder framework with Long Short-Term Memory (LSTM) layers and a Dense layer for output, was designed to facilitate effective sequence generation (Model Architecture). Subsequently, the model underwent training using an iterative optimization algorithm such as stochastic gradient descent (SGD), with parameters adjusted based on the gradients of the loss function computed using backpropagation (Training). Evaluation of the trained model's performance on a separate test dataset revealed robust performance metrics, including high accuracy scores and low test loss values, indicating the model's effectiveness in distinguishing between correct and incorrect code snippets (Evaluation). The results underscore the potential of the developed classification and correction models to revolutionize various aspects of software development, including code validation processes and educational platforms, with implications for automated code analysis and correction in real-world applications (Conclusion).Item Prediction and detection of personal information from written text using AI techniques(2024) Helima Sawsen - Ghazel AnfelMachine learning represents a key branch of artificial intelligence, aiming to enable computer systems to learn from data without explicit programming for each task. In various fields, the prediction and detection of personal information from texts have become crucial for information security and privacy protection. Traditional machine learning methods often face the challenge of representing textual data in a numerical form, a problem addressed by text vectorization techniques. These techniques, such as bag-of-words representation and word embeddings, capture the semantics and context of words in a text, thereby improving the accuracy of prediction and detection models. This synergy between machine learning and text vectorization offers promising prospects for privacy protection and compliance with data protection regulations. In this thesis, we will focus on predicting and detecting personal information, such as the age and gender of the author, from written texts and data collected from online blogs. To achieve these objectives, we will adopt machine learning methods, particularly implementing multilayer neural networks for classification, as well as the TF-IDF text vectorization technique for keyword extraction.Item Swarm Intelligence for Sentiment Analysis from Comments(2024) Faiza MERAH Nawel FALEKSentiment analysis is an NLP technique used to identify the emotional tone of a piece of text. This method helps discern the sentiment conveyed in texts like reviews, comments, social media posts, or any other data involving human language. To understand and categorise the sentiment conveyed in textual data, machine learning offers methods and tools necessary to create models capable of this. ML techniques are numerous, for this work Support Vector Machine(svm) is used; but this algorithm relies on a set of hyper-parameters which greatly influence its performance. Therefore, this study takes hyper-parameter tuning as an optimization problem and employs a swarm-based optimization technique to enhance the performance of SVM in sentiment analysis by using the recently introduced Termite Alate Optimization Algorithm. The performance of proposed approach is evaluated using five metrics: accuracy, precision, recall, F1-measure and computation time on five well known datasets in the field of sentiment analysis and compared to another famous swarm-based algorithm: Particle Swarm Optimization. The experimental results show significant improvement in the SVM performance with optimized hyper-parameters by the proposed approach compared to SVM with default hyper-parameters in all datasets. Remarkably, TAOA demonstrated faster tuning compared to PSO, completing its process with notable efficiency.Item Specification and Verification of an IoT terrorist attack system using ThingML/Coloured Petri Nets(2024) M.Ouadaoui Mohamed ElamineThis thesis presents a comprehensive methodology for the specification, verification, and implementation of a security system using a combination of Unified Modeling Language (UML), ThingML, CPN Tools, and Petri Nets. The approach ensures a robust framework for designing secure systems and validating their functionality before deployment. Initially, the security system’s requirements and architecture are modeled using UML, providing a clear and structured representation of system components and interactions. ThingML is then utilized to specify the detailed behavior of the system, allowing for seamless integration with Internet of Things (IoT) devices and ensuring platform-independent design. To ensure the correctness and reliability of the specified system, Colored Petri Nets (CPN) are employed for formal verification. Using CPN Tools, the system’s models are analyzed for potential errors, deadlocks, and inefficiencies, ensuring a high level of confidence in the system’s performance and security. As a case study, the thesis details the development of a web application using Django, designed for intruder detection through facial recognition. This application is integrated with IoT devices, enabling real-time monitoring and response to security breaches. The facial recognition system accurately identifies intruders, and the integration with IoT allows for immediate alerts and automated actions.Item TRANSFER LEARNING APPROACHES FOR IMPROVING MACHINE LEARNING MODEL ON SMALL DATASETS: CASE STUDY SKIN LESION DETECTION(2024) BENSACI FATEH SAI ABDECHAKOUROne of the biggest challenges of working with machine learning is the need for large amounts of data to train accurate models. Small datasets, in particular, can pose significant challenges to machine learning models. Researchers have developed several approaches to address the challenges of working with small datasets, including transfer learning, active learning, and data augmentation. In this study, we employed the transfer learning as a machine learning paradigm that leverages knowledge gained from one task or domain (source domain) to improve performance on a different but related task or domain (target domain). This study focuses on optimizing model accuracy, analyzing feature importance, and understanding the impact of different preprocessing techniques. The results demonstrate significant improvements in diagnostic performance compared to traditional methods, highlighting the potential of deep learning in clinical settings. Future work involves enhancing model generalization and exploring the integration of multimodal data to further refine diagnostic capabilities. We started from VGG16 as a pre-trained model for training a baseline model in a CNN on the domain of skin lesion detection. In the experimental part, we evaluated different structures, including a custom CNN and VGG16, in terms of detection accuracy. The results showed that transfer learning with the pre-trained model VGG16 provided the best performance, achieving an accuracy of 85%, highlighting the importance of transfer learning.Item Validation and implementation of the transformation from Pomsets to LTS(2024) Mr. Kadjouf IslemThis project focuses on the verification and validation of partially ordered multi-sets (Pomsets) in labelled transition systems (LTS). Fully implemented in Python, this transition process is meticulously executed using the tools and fundamentals of the language. The validation phase examines the fidelity and accuracy of the conversion, ensuring that the resulting LTS correctly represents the behavior of the original Pomsets. The verification of an algorithm’s correctness is crucial in software engineering, especially in embedded systems, where it is essential to ensure that the implementation meets the required specifications. For this purpose, we have used Pymodel. Pymodel is a powerful tool for modeling and verifying software systems. It allows for simulating models to detect logical errors and unexpected behaviors in real-time.Item A machine learning approach for energy optimization in Mobile Cloud Computing(2024) Mrs. MORDJANE Maissoun Mrs. MEBARKI AridjeIn an era marked by rapid technological advancement, smartphones have become integral to our daily lives, transforming how we engage with the world. Despite their indispensability, these powerful devices struggle with a variety of performance issues. Among these, limited battery life stands out as a particularly critical challenge. Mobile Cloud Computing offers a promising solution by leveraging remote servers to offload resource-intensive tasks. However, existing approaches grapple to adapt to the dynamic nature of mobile environments, leading to inefficient energy consumption. To address this and enhance the overall system efficiency, this work is dedicated to proposing an approach that integrates machine learning techniques such as Reinforcement Learning and K-means clustering. This approach dynamically partitions mobile applications into offloadable tasks for remote execution on cloud servers and unoffloadable tasks for local execution on the mobile device. Through experimentation and evaluation, we demonstrate the efficacy of our approach in improving energy efficiency and performance.Item Design and Implementation of a Skyline-Based System for Recommending Offers to Customers of a Telecommunications Operator(2024) Nezzar Salaheddine Zaabeb ChamsseddineIn the field of recommendation systems (RS), the primary objective is to provide users with relevant suggestions based on their preferences and past behaviours. This study explores the use of the skyline technique to enhance the efficiency and quality of recommendations. The skyline method aids in selecting recommendations by considering multiple user preferences or characteristics, such as price, category, etc. Its versatility allows it to be adapted to various decision-making scenarios involving multiple criteria, reducing complexity by quickly identifying non-dominated choices. Additionally, we examine the integration of TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) and cosine similarity into the recommendation process. TOPSIS ranks alternatives based on their proximity to ideal and anti-ideal solutions, while cosine similarity measures the degree of similarity between two vectors. By combining these methods, we develop a robust and flexible method to multi-criteria decision making, capable of identifying alternatives that are close to the ideal, distinct from the anti-ideal, and reflective of similarity relationships between criteria. The aim of this work is to design and implement an offer recommendation system for a telecommunications operator using the skyline, TOPSIS, and cosine similarity methods. Such a system will help customers select packages, services, and promotions that best meet their needs and usage patterns, while enabling the operator to maximize sales and customer loyalty. The choice of the skyline method is justified by its ability to streamline decisionmaking by focusing on a relevant and non-dominated subset of data, thereby eliminating clearly inferior choices across all considered criteria.Item L’analyse des sentiments en utilisant des algorithmes d’essaims (Swarm Intelligence Algorithm) à partir de contenus en ligne(2024) Bouchami Ilias Gasmi Nour ErrafaDe nos jours, l'analyse de texte ou la fouille des textes revêt une grande importance, notamment dans des domaines sensibles comme la politique, la santé, etc. Les réseaux sociaux jouent un rôle crucial à cet égard, car ils regorgent plein de textes sur des sujets variés. C'est pourquoi l’analyse des sentiments est primordiale car elle permet de gérer efficacement les opinions et offre un aperçu rapide de l'opinion publique. Le but de notre travail est de proposer une approche ou un modèle simple pour l’analyse de sentiments des commentaires des internautes. Ces commentaires sont exprimés en plusieurs langues, notamment l'arabe, le dialecte algérien, le français et l'anglais. Notre modèle permet de classifier les sentiments en deux classes positive et negative. Pour réaliser notre travail, Nous avons utilisé deux algorithmes basés sur les essaims (Swarm algorithms) combinés avec une technique d’apprentissage automatique (Machine Learning) à base de l’algorithme SVM (Support Vector Machine), qui est un classifieur.Item L’analyse des sentiments en utilisant des algorithmes d’essaims (Swarm Intelligence Algorithm) à partir de contenus en ligne(2024) Bouchami Ilias Gasmi Nour ErrafaDe nos jours, l'analyse de texte ou la fouille des textes revêt une grande importance, notamment dans des domaines sensibles comme la politique, la santé, etc. Les réseaux sociaux jouent un rôle crucial à cet égard, car ils regorgent plein de textes sur des sujets variés. C'est pourquoi l’analyse des sentiments est primordiale car elle permet de gérer efficacement les opinions et offre un aperçu rapide de l'opinion publique. Le but de notre travail est de proposer une approche ou un modèle simple pour l’analyse de sentiments des commentaires des internautes. Ces commentaires sont exprimés en plusieurs langues, notamment l'arabe, le dialecte algérien, le français et l'anglais. Notre modèle permet de classifier les sentiments en deux classes positive et negative. Pour réaliser notre travail, Nous avons utilisé deux algorithmes basés sur les essaims (Swarm algorithms) combinés avec une technique d’apprentissage automatique (Machine Learning) à base de l’algorithme SVM (Support Vector Machine), qui est un classifieur.Item Generative artificial intelligence for content synthesis and information retrieval(2024) Chabbi Islam - Falek LamisseGenerative artificial intelligence has emerged as a revolutionary technology capable of transforming how content is created and information is retrieved. In this project, we explored the capabilities of advanced generative AI models, such as GPT-4, to develop four distinct services: a PDF chatbot, a blog chatbot, a YouTube video summarization service, and a SQL chatbot. These services are hosted on an accessible web server, providing users with an interactive interface to query and obtain precise and contextual responses. The PDF chatbot allows users to ask questions about the content of PDF documents and receive accurate answers. The blog chatbot enables users to interact with blog post content, asking questions and receiving instant, contextual responses. The YouTube video summarization service extracts and summarizes video content, offering users concise summaries and the ability to interact via chat. The SQL chatbot is designed to interact with SQL databases, allowing users to ask questions about the data and receive precise answers. Our work demonstrates the significant potential of generative artificial intelligence in automating and enhancing content creation and information retrieval processes. By developing and deploying advanced AI services for PDFs, blogs, YouTube videos, and SQL databases, we contribute to more intelligent and efficient information systems, benefiting users across various fields.Item The reinforcement learning of an autonomous agent, Case study: Taxi Game(2024) Boudouha Islem Hoggas Salah EddineReinforcement learning (RL) is a powerful machine learning technique that enables autonomous agents to learn and make decisions by interacting with their environment. It is inspired by the behavioural psychology concept of learning through trial-and-error, where an agent takes actions, receives rewards or penalties, and adjusts its behaviour accordingly to achieve specific goals. In this thesis, we focus on the mechanisms, techniques, and applications of this exciting discipline, We apply the Q-learning algorithm wich is a reinforcement learning algorithm that seeks to find the best possible next action given its current state, in order to maximise the reward it receives from the enviroment Taxi game wich is one of many environments available on OpenAI Gym, The aim of this algorithm is to make sure the taxi can get to the passenger, pick him up and bring him to the drop-off location in the fastest way possible, The core concept of Q-learning revolves around the notion of a Q-table, a data structure that stores the estimated values of taking a specific action in a particular state. As the agent (taxi) interacts with the environment, it continuously updates the Q-table based on the rewards it receives. This iterative process enables the agent to gradually learn which actions lead to favorable outcomes and which ones should be avoided.Item Reconnaissance Faciale avec Machine Learning(2024) — GHALMI Hanane — AFOUFOU SabrinaLa reconnaissance faciale a émergé comme un domaine fascinant et indispensable dans le domaine de la vision par ordinateur et de l'intelligence artificielle, avec des applications allant de la sécurité et la surveillance à l'authentification biométrique. Dans le contexte de cette évolution rapide, l’objectif de ce mémoire est d’explorer et d’évaluer l’utilisation conjointe de l’apprentissage automatique (des Machine à Vecteurs de Support SVM) et de deux techniques d’extraction des caractéristiques les Histogrammes des Gradients Orientés (HOG) et le Motif Binaire Local (LBP). Nous présenterons un état de l’art des approches récentes dans ce domaine, et nous allons évaluer ces techniques en termes d'efficacité et de performance. Notre étude est basée sur l'extraction des caractéristiques qui identifie les informations des zones du visage et leur classification, à partir de ces informations extraites. Enfin, nous avons essayé d'obtenir de meilleurs résultats en effectuant plusieurs tests avec les descripteurs HOG et LBP où nous avons obtenus un taux de reconnaissance (98,7 %) et (93,4 %) sur les bases de données de référence ORL et LFW en utilisant le classifieur SVM.Item Spécification formelle des systèmes complexes par l’utilisation des réseaux de Petri colorés basés sur Python(2024) Sekaoui Amir Dendouga SafaCe mémoire est consacré à la spécification formelle des systèmes complexes en utilisant les réseaux de Pétri colorés basés sur Python. Cela implique la spécification du comportement, des états et des transitions du système en utilisant le formalisme des réseaux de Pétri colorés (CPN) et en les implémentant en Python. En utilisant les réseaux de Pétri colorés, qui constituent un moyen formel de représenter visuellement la façon dont les choses circulent et interagissent, ainsi que Python, un langage de programmation, nous nous sommes capables de créer un modèle clair et structuré du fonctionnement des systèmes complexes. Cela nous aide à comprendre et à prévoir leur comportement, ce qui les rend plus fiables et plus faciles à gérer.Item An ontology framework to allow semantic communications between cyber-physical systems(2024) Aroua SmailThis thesis focuses on the development of an ontology framework to enable semantic communication between Cyber-Physical Systems (CPS). CPS, merging physical processes with computational technologies, offers significant potential for efficiency and innovation. The framework addresses the gap between the cyber and physical worlds within CPS, facilitating seamless data exchange and informed decision-making. Key topics include the design methodology, implementation steps, and the role of Semantic Web technologies like RDF, RDFS, and OWL. The framework's impact on communication, interoperability, and decisionmaking in CPS is discussed, with suggestions for future refinement and application.Item An Annotation tool for Natural language processing tasks(2024) M. Bezza aya M. Maache NadjetThis dissertation creates an annotation tool used in Natural Language Processing (NLP) tasks. NLP has the appropriate techniques to exploit valuable information, provided that a large amount of annotated textual data is available. Most annotation processes rely on manually handling a large body of text for development and evaluation. Creating a large annotated corpus is tedious and requires adequate computational support. Although many annotation tools are available, their primary weaknesses lie in their specific purposes and commercial licenses. Since the quality of the data used to train the NLP model directly affects the quality of the results, ensuring quality control of the annotations is essential. To facilitate this process, we have created a tool that helps us to collect comments from social media platforms such as YouTube and Reddit and annotate them.Item Generative artificial intelligence for content synthesis and information retrieval(2024) Chabbi Islam Falek LamisseGenerative artificial intelligence has emerged as a revolutionary technology capable of transforming how content is created and information is retrieved. In this project, we explored the capabilities of advanced generative AI models, such as GPT-4, to develop four distinct services: a PDF chatbot, a blog chatbot, a YouTube video summarization service, and a SQL chatbot. These services are hosted on an accessible web server, providing users with an interactive interface to query and obtain precise and contextual responses. The PDF chatbot allows users to ask questions about the content of PDF documents and receive accurate answers. The blog chatbot enables users to interact with blog post content, asking questions and receiving instant, contextual responses. The YouTube video summarization service extracts and summarizes video content, offering users concise summaries and the ability to interact via chat. The SQL chatbot is designed to interact with SQL databases, allowing users to ask questions about the data and receive precise answers. Our work demonstrates the significant potential of generative artificial intelligence in automating and enhancing content creation and information retrieval processes. By developing and deploying advanced AI services for PDFs, blogs, YouTube videos, and SQL databases, we contribute to more intelligent and efficient information systems, benefiting users across various fields.Item The reinforcement learning of an autonomous agent, Case study: Taxi Game(2024) Boudouha Islem Hoggas Salah EddineReinforcement learning (RL) is a powerful machine learning technique that enables autonomous agents to learn and make decisions by interacting with their environment. It is inspired by the behavioural psychology concept of learning through trial-and-error, where an agent takes actions, receives rewards or penalties, and adjusts its behaviour accordingly to achieve specific goals. In this thesis, we focus on the mechanisms, techniques, and applications of this exciting discipline, We apply the Q-learning algorithm wich is a reinforcement learning algorithm that seeks to find the best possible next action given its current state, in order to maximise the reward it receives from the enviroment Taxi game wich is one of many environments available on OpenAI Gym, The aim of this algorithm is to make sure the taxi can get to the passenger, pick him up and bring him to the drop-off location in the fastest way possible, The core concept of Q-learning revolves around the notion of a Q-table, a data structure that stores the estimated values of taking a specific action in a particular state. As the agent (taxi) interacts with the environment, it continuously updates the Q-table based on the rewards it receives. This iterative process enables the agent to gradually learn which actions lead to favorable outcomes and which ones should be avoided.Item Reconnaissance Faciale avec Machine Learning(2024) — GHALMI Hanane — AFOUFOU SabrinaLa reconnaissance faciale a émergé comme un domaine fascinant et indispensable dans le domaine de la vision par ordinateur et de l'intelligence artificielle, avec des applications allant de la sécurité et la surveillance à l'authentification biométrique. Dans le contexte de cette évolution rapide, l’objectif de ce mémoire est d’explorer et d’évaluer l’utilisation conjointe de l’apprentissage automatique (des Machine à Vecteurs de Support SVM) et de deux techniques d’extraction des caractéristiques les Histogrammes des Gradients Orientés (HOG) et le Motif Binaire Local (LBP). Nous présenterons un état de l’art des approches récentes dans ce domaine, et nous allons évaluer ces techniques en termes d'efficacité et de performance. Notre étude est basée sur l'extraction des caractéristiques qui identifie les informations des zones du visage et leur classification, à partir de ces informations extraites. Enfin, nous avons essayé d'obtenir de meilleurs résultats en effectuant plusieurs tests avec les descripteurs HOG et LBP où nous avons obtenus un taux de reconnaissance (98,7 %) et (93,4 %) sur les bases de données de référence ORL et LFW en utilisant le classifieur SVM.