http://archives.univ-biskra.dz/handle/123456789/1092 Mon, 06 Apr 2026 22:43:42 GMT 2026-04-06T22:43:42Z http://archives.univ-biskra.dz/handle/123456789/31594 Titre: Incorporating Deep Learning and Optimization Techniques with Data Augmentation for Improved Image Analysis and Classificatio Auteur(s): Nouara BOUDOUH Résumé: Deep learning methods often face challenges due to unbalanced or non-representative data, and in many cases, data scarcity limits model effectiveness. We advocate that improving data quality can lead to significant performance enhancements. This thesis presents new methods for data augmentation. Our first method involves randomly create f ilters to remove certain rows and columns from the original image to generate smaller, more informative images. This method was applied to the Cats vs. Dogs dataset to train the Basic CNN and ResNet50 models, showing improved results compared to the original dataset. However, random filter generation can sometimes produce images that are too similar to the originals, reducing diversity. To address this, we developed a secondary technique incorporating a random optimization algorithm to select optimal generated im ages based on entropy, yielding promising results when applied to the VGG16 model. Nevertheless, image selection remains dependent on filter quality, potentially limiting diversity. Therefore, our third method employs a genetic algorithm to enhance filter generation and ensure greater diversity. Additionally, we improved the architectures of the VGG16 and VGG19 models. When applied to the Cats vs. Dogs and Chest X-ray datasets and used to train a set of seven models (VGG16, VGG19, their enhanced ver sions, EfficientNet-B0, Inception-V3, and Vision Transformer), we observed promising improvements in model performance compared to the second method. Since optimization techniques require considerable time and resources, we proposed an alternative method to enhance model performance without increasing data size. This approach leverages the unique capabilities of each model to extract features by merging their outputs into a uni f ied representation used to train a single classifier. The integrated models using VGG16, VGG19, EfficientNet-B0, and Inception-V3 showed clear performance superiority com pared to each model’s individual performance Description: Image and Artificial Life Wed, 01 Jan 2025 00:00:00 GMT http://archives.univ-biskra.dz/handle/123456789/31594 2025-01-01T00:00:00Z http://archives.univ-biskra.dz/handle/123456789/31593 Titre: Deep Learning technique and parallel optimization algorithm for intelligent pattern recognition Auteur(s): Hakima Rym RAHAL Résumé: Misdiagnosis poses a significant challenge within the healthcare sector, carrying potentially severe consequences for patients, including delayed or inappropriate treatment, unnecessary me dical procedures, emotional distress, financial burdens, and legal repercussions. To address this issue, we propose the utilization of deep learning algorithms to enhance the precision of medi cal diagnoses. However, the development of accurate deep learning models for medical purposes necessitates substantial quantities of top-quality data, a resource that can be challenging for individual healthcare entities to acquire. Consequently, there is a need to aggregate data from various sources to create a diverse dataset suitable for effective model training. Nevertheless, the sharing of medical data across different healthcare sectors is fraught with security concerns due to the sensitive nature of the information and stringent privacy regulations. To tackle these complex challenges, we advocate for the adoption of Blockchain technology, which offers a se cure, decentralized, and privacy-centric approach to sharing locally trained deep learning models, thereby obviating the need to exchange raw data. Our proposed technique, known as model en sembling, combines the strengths of multiple local deep learning models by aggregating their weights to construct a unified global model. This global model enables accurate diagnosis of intricate medical conditions across various locations while safeguarding patient privacy and data integrity. Our research serves as a testament to the efficacy of this approach, achieving high accuracy rates in the diagnosis of three diseases (accuracy of 97.44 % for the Breast Cancer, 97.14 % for the Diabetes, and 98.51 % for the Lung Cancer) that surpass those of individual local models. Furthermore, we have successfully developed a multi-diagnosis application as an outcome of this innovative methodology Description: Artificial intelligence and image processing Wed, 01 Jan 2025 00:00:00 GMT http://archives.univ-biskra.dz/handle/123456789/31593 2025-01-01T00:00:00Z http://archives.univ-biskra.dz/handle/123456789/31591 Titre: Towards Green Security for E-health Applications in The Internet of Things Auteur(s): Meriem GASMI Résumé: The Internet of Things (IoTs) is an emerging technology that connects various de vices and systems to the Internet, allowing them to communicate and share data. This interconnected network of devices has the potential to revolutionize industries, improve efficiency, and enhance our daily lives. The IoT relies on physical sensors that gather data and transmit it to remote cloud computing platforms for analysis and storage. One of the most promising fields for IoT applications is healthcare, where significant advancements are anticipated. For instance, devices can be implanted in patients’ bodies to monitor vital signs, transmitting the data to the cloud for storage, processing, and informed decision making. The data should then be accessed by healthcare professionals in real-time, al lowing for immediate intervention when necessary. With the potential to revolutionize the healthcare industry, the integration of IoT technology in healthcare applications is becoming increasingly prevalent and essential. This research focuses on security aspects, specifically addressing the confidentiality and access control of data (e.g., EMR) during transmission and storage in the cloud. In addition, we examine the energy consumption of IoT devices in healthcare applications, ensuring their optimization for better efficiency. To ensure the confidentiality and security of these sensitive data, encryption techniques and access control protocols must be carefully implemented. Energy efficiency is another crucial aspect to consider, as IoT devices in healthcare applications must be able to op erate continuously and autonomously without draining excessive power. By addressing these challenges, we can fully harness the potential of IoT technology to improve patient outcomes and streamline healthcare delivery. Ciphertext-Policy Attribute-Based Encryp tion (CP-ABE) is widely regarded as an ideal method for implementing fine-grained access control. However, existing CP-ABE solutions are not efficient or well-suited for the IoT environment, as data producers are typically highly resource-constrained and unable to perform the public-key cryptographic functions required by CP-ABE. In this thesis, we suggest enhancing the standard CP-ABE to make it more suitable for resource-constrained sensors by offloading the intensive encryption tasks to multiple cooperative nodes. This approach ensures a balanced workload distribution between the sensors and the assistant nodes, taking into account the capacity of the assistant nodes for optimal distribution. We evaluated our proposal through a series of experiments, and the results demonstrate that, compared to existing outsourcing ABE solutions, our approach significantly improves both computation time and energy efficiency, guarantees data confidentiality, prolongs the sensor’s battery life, and ensures fine-grained access control Description: Networking and Distributed Systems Wed, 01 Jan 2025 00:00:00 GMT http://archives.univ-biskra.dz/handle/123456789/31591 2025-01-01T00:00:00Z http://archives.univ-biskra.dz/handle/123456789/29326 Titre: Contribution au développement de concepts et outils d’aide à la décision pour l’optimisation via le plongement dans un réseau d’interconnexion parallèle. Auteur(s): SELMI _Aymen_TakieEddine Résumé: In a world characterized by complex and interconnected challenges, effective decision-making is paramount for addressing issues spanning environmental sustainability, transportation infrastructure improvement, and medical innovation. However, the growing complexity of these problems often exceeds traditional reasoning capabilities. Decision support systems, leveraging artificial intelligence techniques, offer promising avenues for navigating these challenges. This thesis focuses on addressing one such complex problem, the Traveling Salesman Problem (TSP), which finds applications in logistics, network planning, and bioinformatics. Despite advancements in TSP-solving methods, scalability and adaptability to dynamic scenarios remain persistent challenges. This research proposes a parallel simulation via an interconnection network topologybased optimization tool integrating advanced artificial intelligence techniques to tackle these issues. The methodology includes hierarchical clustering representations, graph embeddings, and hybrid parallel-solving strategies. Key contributions include novel clustering algorithms tailored for TSP optimization, integration with parallel computing architectures, and experimental validation showcasing superior performance compared to traditional methods. The thesis outlines theoretical foundations, explores parallel computing architectures and graph embedding techniques with the best quality, and presents a comprehensive evaluation of the proposed methodology. The findings contribute to enhancing decision-making processes and offer a robust framework for addressing complex optimization challenges in dynamic real-world settings. Description: Image et Vie Artificielle Mon, 01 Jan 2024 00:00:00 GMT http://archives.univ-biskra.dz/handle/123456789/29326 2024-01-01T00:00:00Z