Literatura académica sobre el tema "Intelligence artificielle (ML/DL)"
Crea una cita precisa en los estilos APA, MLA, Chicago, Harvard y otros
Consulte las listas temáticas de artículos, libros, tesis, actas de conferencias y otras fuentes académicas sobre el tema "Intelligence artificielle (ML/DL)".
Junto a cada fuente en la lista de referencias hay un botón "Agregar a la bibliografía". Pulsa este botón, y generaremos automáticamente la referencia bibliográfica para la obra elegida en el estilo de cita que necesites: APA, MLA, Harvard, Vancouver, Chicago, etc.
También puede descargar el texto completo de la publicación académica en formato pdf y leer en línea su resumen siempre que esté disponible en los metadatos.
Artículos de revistas sobre el tema "Intelligence artificielle (ML/DL)"
1
Brouchet, Edouard, François de Brondeau, Marie-José Boileau y Masrour Makaremi. "Apport de l’intelligence artificielle dans la prévision de croissance mandibulaire : revue systématique de la littérature". Revue d'Orthopédie Dento-Faciale 58, n.º 2 (junio de 2024): 185–209. http://dx.doi.org/10.1051/odf/2024021.
Texto completoResumen
L’orthodontiste intervient principalement auprès d’enfants en cours de croissance. L’examen clinique initial ne fournit qu’une image statique qui doit être interprétée en tenant compte de son évolution potentielle. Une prédiction précise de la croissance mandibulaire, permettrait au praticien d’améliorer le diagnostic, la planification du traitement et ainsi la prise en charge du patient. De nombreux travaux de recherche ont été menés, basés sur des signes structuraux, des analyses céphalométriques et des valeurs d’agrandissement moyen, mais restent imprécis. Les limites rapportées comprennent principalement une variabilité interindividuelle extrême, des schémas de croissance variables selon l’âge, un manque de signes structuraux caractéristiques avant la puberté, l’utilisation de normes statistiques et de résultats dépendant directement de l’expérience du clinicien. À ce jour, il n’existe aucun consensus sur la meilleure méthode pour prédire la croissance mandibulaire, et l’orthodontiste ne peut se fier uniquement à son intuition d’expert. Ces dernières années, la combinaison de l’intelligence artificielle (IA) et des sciences cognitives dans le domaine médical a révolutionné l’interprétation des radiographies. Les techniques d’apprentissage automatique (ML) et profond (DL) représentent une approche novatrice grâce à leur capacité à analyser d’énormes volumes de données tout en éliminant les biais humains. L’objectif de cette revue systématique était d’examiner les différents résultats des prévisions de croissance mandibulaire par intelligence artificielle chez des patients en cours de croissance. Ces résultats suggèrent que nous ne sommes encore qu’aux débuts de l’orthodontie tirant parti du diagnostic et de la prise de décision de l’IA, mais ces modèles de prévision de croissance devraient devenir, dans un avenir proche, des systèmes de support clinique fiables pour les orthodontistes.
2
AFTAB, Ifra, Mohammad DOWAJY, Kristof KAPITANY y Tamas LOVAS. "Artificial Intelligence (AI) – based strategies for point cloud data and digital twins". Nova Geodesia 3, n.º 3 (19 de agosto de 2023): 138. http://dx.doi.org/10.55779/ng33138.
Texto completoResumen
Artificial Intelligence (AI), specifically machine learning (ML) and deep learning (DL), is causing a paradigm shift in coding practices and software solutions across diverse fields. This study focuses on harnessing the potential of ML/DL strategies in the geospatial domain, where geodata possesses characteristics that align with the concept of a “lingual manuscript” in aesthetic theory. By employing ML/DL techniques, such as feature evaluation and extraction from 3D point clouds, we can derive concepts that are specific to software, geographical areas, and tasks. ML/DL-based interpretation of 3D point clouds extends geospatial modelling beyond implicit representations, enabling the resolution of complex heuristic-based reconstructions and abstract concepts. These advancements in artificial intelligence have the potential to optimize and expedite geodata computation and geographic information systems. However, ML/DL encounters notable challenges in this domain, including the need for abundant training data, advanced statistical methods, and the development of effective feature representations. Overcoming these challenges is essential to enhance the performance and efficacy of ML/DL systems. Additionally, ML/DL-based solutions can simplify software engineering processes by replacing certain aspects of current adoption and implementation practices, resulting in reduced complexities in development and management. Through the adoption of ML/DL, many of the existing explicitly coded GIS implementations may gradually be replaced in the long term. Overall, this research illustrates the transformative capabilities of ML/DL in geospatial applications and underscores the significance of addressing associated challenges to drive further advancements in the field.
3
Choudhary, Laxmi y Jitendra Singh Choudhary. "Deep Learning Meets Machine Learning: A Synergistic Approach towards Artificial Intelligence". Journal of Scientific Research and Reports 30, n.º 11 (16 de noviembre de 2024): 865–75. http://dx.doi.org/10.9734/jsrr/2024/v30i112614.
Texto completoResumen
The evolution of artificial intelligence (AI) has progressed from rule-based systems to learning-based models, integrating machine learning (ML) and deep learning (DL) to tackle complex data-driven tasks. This review examines the synergy between ML, which utilizes algorithms like decision trees and support vector machines for structured data, and DL, which employs neural networks for processing unstructured data such as images and natural language. The combination of these paradigms through hybrid ML-DL models has enhanced prediction accuracy, scalability, and automation across domains like healthcare, finance, natural language processing, and robotics. However, challenges such as computational demands, data dependency, and model interpretability remain. This paper discusses the benefits, limitations, and future potential of ML and DL and also provides a review study of a hybrid model makes use of both techniques (machine learning & deep learning) advantages to solve complicated problems more successfully than one could on its own. To boost performance, increase efficiency, or address scenarios where either ML or DL alone would not be able to manage, this approach combines deep learning structures with conventional machine learning techniques.
4
Zhang, Shengzhe. "Artificial Intelligence and Applications in Structural and Material Engineering". Highlights in Science, Engineering and Technology 75 (28 de diciembre de 2023): 240–45. http://dx.doi.org/10.54097/9qknfc57.
Texto completoResumen
The integration of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) has become a vital tool attributed to Structural and Material Engineering and developed the way engineers approach design analysis and optimization. This paper explores the principal models of ML and DL, such as the generative adversarial network (GAN) and the artificial neural networks (ANN) and, and discusses their impacts on the applications of material design, structure damage detection (SDD), and archtecture design. It indicates that the high-quality of database is the essential key to training the model. Thus, the data preprocessing is required for expanding the data source and improving the quality of data. In material design process, ML and DL models reduce the time to predict the properties of construction materials, which makes SDD realistic as well. For architecture design, GAN is used to generate image data, such as drawing of the floor plan and this could be helpful to reduce the labor resources. However, some challenges of ML and DL are found while applying the algorithms to real-life applications. For example, sufficient data is needed to train the DL models and the ethic aspect is also a concern when thinking of AI.
5
Iadanza, Ernesto, Rachele Fabbri, Džana Bašić-ČiČak, Amedeo Amedei y Jasminka Hasic Telalovic. "Gut microbiota and artificial intelligence approaches: A scoping review". Health and Technology 10, n.º 6 (26 de octubre de 2020): 1343–58. http://dx.doi.org/10.1007/s12553-020-00486-7.
Texto completoResumen
Abstract This article aims to provide a thorough overview of the use of Artificial Intelligence (AI) techniques in studying the gut microbiota and its role in the diagnosis and treatment of some important diseases. The association between microbiota and diseases, together with its clinical relevance, is still difficult to interpret. The advances in AI techniques, such as Machine Learning (ML) and Deep Learning (DL), can help clinicians in processing and interpreting these massive data sets. Two research groups have been involved in this Scoping Review, working in two different areas of Europe: Florence and Sarajevo. The papers included in the review describe the use of ML or DL methods applied to the study of human gut microbiota. In total, 1109 papers were considered in this study. After elimination, a final set of 16 articles was considered in the scoping review. Different AI techniques were applied in the reviewed papers. Some papers applied ML, while others applied DL techniques. 11 papers evaluated just different ML algorithms (ranging from one to eight algorithms applied to one dataset). The remaining five papers examined both ML and DL algorithms. The most applied ML algorithm was Random Forest and it also exhibited the best performances.
6
Gokcekuyu, Yasemin, Fatih Ekinci, Mehmet Serdar Guzel, Koray Acici, Sahin Aydin y Tunc Asuroglu. "Artificial Intelligence in Biomaterials: A Comprehensive Review". Applied Sciences 14, n.º 15 (28 de julio de 2024): 6590. http://dx.doi.org/10.3390/app14156590.
Texto completoResumen
The importance of biomaterials lies in their fundamental roles in medical applications such as tissue engineering, drug delivery, implantable devices, and radiological phantoms, with their interactions with biological systems being critically important. In recent years, advancements in deep learning (DL), artificial intelligence (AI), machine learning (ML), supervised learning (SL), unsupervised learning (UL), and reinforcement learning (RL) have significantly transformed the field of biomaterials. These technologies have introduced new possibilities for the design, optimization, and predictive modeling of biomaterials. This review explores the applications of DL and AI in biomaterial development, emphasizing their roles in optimizing material properties, advancing innovative design processes, and accurately predicting material behaviors. We examine the integration of DL in enhancing the performance and functional attributes of biomaterials, explore AI-driven methodologies for the creation of novel biomaterials, and assess the capabilities of ML in predicting biomaterial responses to various environmental stimuli. Our aim is to elucidate the pivotal contributions of DL, AI, and ML to biomaterials science and their potential to drive the innovation and development of superior biomaterials. It is suggested that future research should further deepen these technologies’ contributions to biomaterials science and explore new application areas.
7
Gayatri, T., G. Srinivasu, D. M. K. Chaitanya y V. K. Sharma. "A Review on Optimization Techniques of Antennas Using AI and ML / DL Algorithms". International Journal of Advances in Microwave Technology 07, n.º 02 (2022): 288–95. http://dx.doi.org/10.32452/ijamt.2022.288295.
Texto completoResumen
In recent years, artificial intelligence (AI) aided communications grabbed huge attention to providing solutions for mathematical problems in wireless communications, by using machine learning (ML) and deep learning (DL) algorithms. This paper initially presents a short background on AI, CEM, and the role of AI / ML / DL in antennas. A study on ML / DL algorithms and the optimization techniques of antenna parameters using various ML / DL algorithms are presented. Finally, the application areas of AI in antennas are illustrated.
8
Drikakis, Dimitris y Filippos Sofos. "Can Artificial Intelligence Accelerate Fluid Mechanics Research?" Fluids 8, n.º 7 (19 de julio de 2023): 212. http://dx.doi.org/10.3390/fluids8070212.
Texto completoResumen
The significant growth of artificial intelligence (AI) methods in machine learning (ML) and deep learning (DL) has opened opportunities for fluid dynamics and its applications in science, engineering and medicine. Developing AI methods for fluid dynamics encompass different challenges than applications with massive data, such as the Internet of Things. For many scientific, engineering and biomedical problems, the data are not massive, which poses limitations and algorithmic challenges. This paper reviews ML and DL research for fluid dynamics, presents algorithmic challenges and discusses potential future directions.
9
An, Ruopeng, Jing Shen y Yunyu Xiao. "Applications of Artificial Intelligence to Obesity Research: Scoping Review of Methodologies". Journal of Medical Internet Research 24, n.º 12 (7 de diciembre de 2022): e40589. http://dx.doi.org/10.2196/40589.
Texto completoResumen
Background Obesity is a leading cause of preventable death worldwide. Artificial intelligence (AI), characterized by machine learning (ML) and deep learning (DL), has become an indispensable tool in obesity research. Objective This scoping review aimed to provide researchers and practitioners with an overview of the AI applications to obesity research, familiarize them with popular ML and DL models, and facilitate the adoption of AI applications. Methods We conducted a scoping review in PubMed and Web of Science on the applications of AI to measure, predict, and treat obesity. We summarized and categorized the AI methodologies used in the hope of identifying synergies, patterns, and trends to inform future investigations. We also provided a high-level, beginner-friendly introduction to the core methodologies to facilitate the dissemination and adoption of various AI techniques. Results We identified 46 studies that used diverse ML and DL models to assess obesity-related outcomes. The studies found AI models helpful in detecting clinically meaningful patterns of obesity or relationships between specific covariates and weight outcomes. The majority (18/22, 82%) of the studies comparing AI models with conventional statistical approaches found that the AI models achieved higher prediction accuracy on test data. Some (5/46, 11%) of the studies comparing the performances of different AI models revealed mixed results, indicating the high contingency of model performance on the data set and task it was applied to. An accelerating trend of adopting state-of-the-art DL models over standard ML models was observed to address challenging computer vision and natural language processing tasks. We concisely introduced the popular ML and DL models and summarized their specific applications in the studies included in the review. Conclusions This study reviewed AI-related methodologies adopted in the obesity literature, particularly ML and DL models applied to tabular, image, and text data. The review also discussed emerging trends such as multimodal or multitask AI models, synthetic data generation, and human-in-the-loop that may witness increasing applications in obesity research.
10
Ali, Zulfiqar, Asif Muhammad, Nangkyeong Lee, Muhammad Waqar y Seung Won Lee. "Artificial Intelligence for Sustainable Agriculture: A Comprehensive Review of AI-Driven Technologies in Crop Production". Sustainability 17, n.º 5 (5 de marzo de 2025): 2281. https://doi.org/10.3390/su17052281.
Texto completoResumen
Smart farming leverages Artificial Intelligence (AI) to address modern agricultural sustainability challenges. This study investigates the application of machine learning (ML), deep learning (DL), and time series analysis in agriculture through a systematic literature review following the PRISMA methodology. The review highlights the critical roles of ML and DL techniques in optimizing agricultural processes, such as crop selection, yield prediction, soil compatibility classification, and water management. ML algorithms facilitate tasks like crop selection and soil fertility classification, while DL techniques contribute to forecasting crop production and commodity prices. Additionally, time series analysis is employed for demand forecasting of crops, commodity price prediction, and forecasting crop yield production. The focus of this article is to provide a comprehensive overview of ML and DL techniques within the farming industry. Utilizing crop datasets, ML algorithms are instrumental in classifying soil fertility, crop selection, and various other aspects. DL algorithms, when applied to farming data, enable effective time series analysis and crop selection. By synthesizing the integration of these technologies, this review underscores their potential to enhance decision-making in agriculture and mitigate food scarcity challenges in the future.
Tesis sobre el tema "Intelligence artificielle (ML/DL)"
1
Laguili, Oumaima. "Smart management of combined electric water heaters and self-consumption photovoltaic solar panels (SmartECS)". Electronic Thesis or Diss., Perpignan, 2024. https://theses-public.univ-perp.fr/2024PERP0045.pdf.
Texto completoResumen
Alors que le secteur du bâtiment se montre de plus en plus économe en énergie, les besoins en eau chaude sanitaire (ECS) augmentent, en particulier dans les logements récents. De ce fait, il apparait nécessaire d'améliorer l'efficacité des solutions mises en œuvre pour la production d'ECS, de mieux comprendre les besoins en ECS et d'impliquer l'usager dans la prise de décision. Le projet traite du développement d'algorithmes pour le contrôle/commande « intelligent » d'installations associant chauffe-eau électrique et panneaux solaires photovoltaïques en autoconsommation. Sera mise en œuvre une stratégie fondée sur la théorie de la commande prédictive, mettant à profit les outils de l'apprentissage automatique. Cette stratégie sera généralisée aux systèmes « multi-chauffe-eau », mutualisant une production solaire photovoltaïque, parle développement d'une commande distribuée et hiérarchisée. Une expérimentation permettra d'évaluer les conditions d'acceptabilité de la solution développée et l'impact de l'information sur la prise de décisionWhile the building sector is increasingly energy efficient, the needs in domestic hot water (DHW) is increasing, especially in newer homes. Therefore, improvement of efficiency in the production of DHW, a better understanding of the needs in DHW, and user involvement in the decision-making process are necessary. The project deals with the development of algorithms for the smart control of combined electric water heaters and self-consumption photovoltaic solar panels. A model-based predictive control strategy will be developed and implemented, leveraging machine learning tools. The strategy will be generalized to multi-water heater systems, sharing photovoltaic solar production, through the development of a distributed and hierarchical control approach. An experiment will make it possible to assess the conditions of acceptability of the developed solution and the impact of information on decision-making
2
Djaidja, Taki Eddine Toufik. "Advancing the Security of 5G and Beyond Vehicular Networks through AI/DL". Electronic Thesis or Diss., Bourgogne Franche-Comté, 2024. http://www.theses.fr/2024UBFCK009.
Texto completoResumen
L'émergence des réseaux de cinquième génération (5G) et des réseaux véhiculaire (V2X) a ouvert une ère de connectivité et de services associés sans précédent. Ces réseaux permettent des interactions fluides entre les véhicules, l'infrastructure, et bien plus encore, en fournissant une gamme de services à travers des tranches de réseau (slices), chacune adaptée aux besoins spécifiques de ceux-ci. Les générations futures sont même censées apporter de nouvelles avancées à ces réseaux. Cependant, ce progrès remarquable les expose à une multitude de menaces en matière de cybersécurité, dont bon nombre sont difficiles à détecter et à atténuer efficacement avec les contre mesures actuelles. Cela souligne la nécessité de mettre en oeuvre de nouveaux mécanismes avancés de détection d'intrusion pour garantir l'intégrité, la confidentialité et la disponibilité des données et des services.Un domaine suscitant un intérêt croissant à la fois dans le monde universitaire qu'industriel est l'Intelligence Artificielle (IA), en particulier son application pour faire face aux menaces en cybersécurité. Notamment, les réseaux neuronaux (RN) ont montré des promesses dans ce contexte, même si les solutions basées sur l'IA sont accompagnées de défis majeurs.Ces défis peuvent être résumés comme des préoccupations concernant l'efficacité et l'efficience. Le premier concerne le besoin des Systèmes de Détection d'Intrusions (SDI) de détecter avec précision les menaces, tandis que le second implique d'atteindre l'efficacité en termes de temps et la détection précoce des menaces.Cette thèse représente l'aboutissement de nos recherches sur l'investigation des défis susmentionnés des SDI basés sur l'IA pour les systemes 5G en général et en particulier 5G-V2X. Nous avons entamé notre recherche en réalisant une revue de la littérature existante. Tout au long de cette thèse, nous explorons l'utilisation des systèmes d'inférence floue (SIF) et des RN, en mettant particulièrement l'accent sur cette derniere technique. Nous avons utilisé des techniques de pointe en apprentissage, notamment l'apprentissage profond (AP), en intégrant des réseaux neuronaux récurrents et des mécanismes d'attention. Ces techniques sont utilisées de manière innovante pour réaliser des progrès significatifs dans la résolution des préoccupations liées à l'amélioration de l'efficacité et de l'efficience des SDI. De plus, nos recherches explorent des défis supplémentaires liés à la confidentialité des données lors de l'utilisation des SDIs basés sur l'AP. Nous y parvenons en exploitant les algorithmes d'apprentissage fédéré (AF) les plus récentsThe emergence of Fifth Generation (5G) and Vehicle-to-Everything (V2X) networks has ushered in an era of unparalleled connectivity and associated services. These networks facilitate seamless interactions among vehicles, infrastructure, and more, providing a range of services through network slices, each tailored to specific requirements. Future generations are even expected to bring further advancements to these networks. However, this remarkable progress also exposes them to a myriad of security threats, many of which current measures struggle to detect and mitigate effectively. This underscores the need for advanced intrusion detection mechanisms to ensure the integrity, confidentiality, and availability of data and services.One area of increasing interest in both academia and industry spheres is Artificial Intelligence (AI), particularly its application in addressing cybersecurity threats. Notably, neural networks (NNs) have demonstrated promise in this context, although AI-based solutions do come with inherent challenges. These challenges can be summarized as concerns about effectiveness and efficiency. The former pertains to the need for Intrusion Detection Systems (IDSs) to accurately detect threats, while the latter involves achieving time efficiency and early threat detection.This dissertation represents the culmination of our research findings on investigating the aforementioned challenges of AI-based IDSs in 5G systems in general and 5G-V2X in particular. We initiated our investigation by conducting a comprehensive review of the existing literature. Throughout this thesis, we explore the utilization of Fuzzy Inference Systems (FISs) and NNs, with a specific emphasis on the latter. We leveraged state-of-the-art NN learning, referred to as Deep Learning (DL), including the incorporation of recurrent neural networks and attention mechanisms. These techniques are innovatively harnessed to making significant progress in addressing the concerns of enhancing the effectiveness and efficiency of IDSs. Moreover, our research delves into additional challenges related to data privacy when employing DL-based IDSs. We achieve this by leveraging and experimenting state-of-the-art federated learning (FL) algorithms
3
Taghavian, Masoud. "VNF placement in 5G Networks using AI/ML". Electronic Thesis or Diss., Ecole nationale supérieure Mines-Télécom Atlantique Bretagne Pays de la Loire, 2024. http://www.theses.fr/2024IMTA0421.
Texto completoResumen
La transition inévitable des dispositifs matériels physiques vers des modules logiciels légers et réutilisables dans le cadre de la virtualisation des fonctions de réseau (NFV) offre d’innombrables possibilités tout en présentant plusieurs défis sans précédent. La satisfaction des attentes de la NFV dans les réseaux post-5G dépend fortement du placement efficace des services de réseau.L’allocation dynamique des ressources physiques pour les demandes de services en ligne exigeant des ressources hétérogènes selon des exigences de qualité de service spécifiques représente l’une des étapes les plus importantes de la conception NFV et un problème NP-Hard à résoudre. Cette complexité est rencontrée dans divers cas d’utilisation 5G NFV, qui sont liés au placement, à partir de VNF Forwarding-Graphs et Network Slicing, à la virtualisation du réseau central, CDN, IoT, etc., examinant de nombreux objectifs dans la littérature, allant des optimisations multi-objectifs basées sur les ressources à la consommation d’énergie, le coût des revenus, l’acceptation du service, la résilience, la disponibilité, la sécurité,etc. Dans cette thèse, nous nous intéressons au placement des services de réseau virtuel sur le réseau en essayant de maximiser le nombre de services acceptés en tenant compte de leurs exigences de qualité de service. Bien que le problème de placement des VNF soit étudié depuis de nombreuses années, le besoin d’une approche permettant de trouver un juste compromis entre l’optimalité et la scalabilité existe toujours. Dans cette thèse, nous étudions plusieurs problèmes et défis dans le placement de services en réseau et proposons des solutions AL/ML en conséquenceThe inevitable transition from physical hardware devices towards lightweight reusable software modules in Network Function Virtualization (NFV) introduces countless opportunities while presenting several unprecedented challenges. Satisfying NFV expectations in post-5G networks heavily depends on the efficient placement of network services. Dynamic allocation of physical resources for online service requests demanding heterogeneous resources under specific QoS requirements represents one of the most important steps in NFV design, and a NP-Hard problem to solve. This complexity is encountered in various 5G NFV use-cases, which are related to the placement, from VNF Forwarding-Graphs and Network Slicing, to the virtualization of the Core Network, CDN, IoT, etc., investigating numerous objectives in the literature, ranging from resources-based multi-objective optimizations to the energy consumption, cost of revenue, service acceptance, resiliency, availability, security, etc. In this thesis, we are interested in placing the virtual network services over the network by trying to maximize the number of accepted services considering their QoS requirements. Although the VNF placement problem has been studied for many years, the need for an approach that could find a fair compromise between optimality and scalability still exists. In this thesis, we study several problems and challenges in network service placement and propose AL/ML solutions accordingly
4
Delestrac, Paul. "Advanced Profiling Techniques For Evaluating GPU Computing Efficiency Executing ML Applications". Electronic Thesis or Diss., Université de Montpellier (2022-....), 2024. http://www.theses.fr/2024UMONS014.
Texto completoResumen
L'augmentation en complexité des applications d'Intelligence Artificielle (IA) entraîne une demande accrue en puissance de calcul et en énergie pour entraîner et exécuter des modèles d'apprentissage automatique (ML). Les processeurs graphiques (GPU), forts d'architectures améliorées (e.g., ajout de cœurs dédiés à l'IA en 2017), sont devenus le système de prédilection pour de telles tâches. Concevoir des systèmes plus efficients pour l'IA n'est possible qu'avec une connaissance approfondie des limites des systèmes existants, où matériel et logiciel sont étroitement couplés. Mais l'abstraction des plateformes d'IA et la nature fermée des architectures GPU modernes masquent le processus d'exécution, rendant ses performances difficiles à évaluer.L'objectif de cette thèse est caractériser les facteurs limitant la performance et augmentant la consommation énergétique des tâches d'IA exécutées avec des GPUs modernes. Cette thèse adresse trois limitations majeures des outils existants. Premièrement, les outils de caractérisation proposés par les plateformes de développement d'IA sont conçus pour aider les développeurs de modèles ML, mais ne donnent pas d'informations sur la charge additionnelle que représente l'exécution de ces plateformes. Deuxièmement, les outils de caractérisation proposés par les fabricants de GPUs permettent l'accès à des compteurs de performance, mais qui ne permettent pas d'estimer l'efficacité des interactions entre le GPU et l'unité centrale (CPU). Enfin, pour caractériser la consommation énergétique des GPUs lors de l'entraînement d'IA, ces outils ne permettent pas d'obtenir une décomposition détaillée. Pour adresser ces limitations, cette thèse propose trois contributions.Premièrement, nous analysons l'exécution des plateformes d'IA sur un couple CPU-GPU. Nous proposons une nouvelle méthodologie de caractérisation réutilisant les données fournies par des outils existants. Cette méthodologie permet d'extraire de nouvelles informations quant à l'exécution de modèles d'IA. Nous étudions l'exécution de trois modèles d'IA et nos résultats montrent que l'exécution des opérations destinées au GPU doit être suffisamment longue pour masquer le temps d'exécution de la plateforme d'IA, augmentant l'utilisation GPU. Pour autant, cette incitation à utiliser des opérations plus longues conduit à l'utilisation de lots de données plus conséquents, augmentant la demande en mémoire GPU.Deuxièmement, nous analysons l'utilisation des ressources internes au GPU lors de l'entraînement. Nous proposons une nouvelle méthodologie de caractérisation combinant les outils proposés par les fabricants de GPUs et par les plateformes d'IA. Nos résultats suggèrent qu'un plafond de performance a été atteint, annulant les bénéfices à utiliser des lots de données plus larges pour l'entraînement. Nous observons que les cœurs les plus performants du GPU (tensor cores) restent inactifs durant la majorité du temps d'entraînement, limité par les opérations qui n'utilisent pas ces cœurs. Nos résultats suggèrent que les architectures GPU modernes ont atteint un point de saturation.Enfin, nous analysons la consommation énergétique des GPUs lors de l'entraînement. Nous proposons une méthodologie basée sur l'utilisation de microprogrammes afin d'obtenir une décomposition de la consommation énergétique. Nos résultats suggèrent que les transferts de données sont responsables pour la majorité de la consommation énergétique dynamique du GPU (jusqu'à 84%). Ces résultats soutiennent la tendance de recherche pour des architectures cherchant optimiser les transferts de données (e.g., traitement en mémoire ou à proximité, architectures vectorielles).Cette thèse propose une analyse approfondie des limites de performance et de la consommation énergétique des tâches d'IA exécutées à l'aide de GPUs modernes. Nous espérons que ce travail inspirera de futures recherches dans cette direction, pour concevoir des accélérateurs d'IA plus efficientsThe rising complexity of Artificial Intelligence (AI) applications significantly increases the demand for computing power to execute and train Machine Learning (ML) models, thus boosting the energy consumption of data centers. GPUs, enhanced by developments like tensor cores (2017), have become the preferred architecture. Building more efficient ML computing systems relies on a deep understanding of the limits of both parts of a tightly coupled hardware/software paradigm. However, the high abstraction of ML frameworks and the closed-source, proprietary design of state-of-the-art GPU architectures obscure the execution process and make performance evaluation tedious.The main goal of this thesis is to provide new methodologies to evaluate performance and energy bottlenecks of GPU-accelerated ML workloads. Existing profiling solutions are limited in three ways. First, ML framework profiling tools are designed to assist the development of ML models but do not give insights into the runtime execution of the ML framework. While these profiling tools provide high-level metrics on the GPU device execution, these metrics can be misleading and overestimate the utilization of the GPU resources. Second, lower-level profiling tools provide access to performance counters and insights on how to optimize GPU kernels. However, these tools cannot capture the efficiency of host/device interactions occurring at a higher level. Finally, when evaluating energy bottlenecks, the mentioned profiling tools cannot provide a detailed breakdown of the energy consumed by modern GPUs during ML training. To tackle these shortcomings, this thesis makes three key contributions organized as a top-down analysis of GPU-accelerated ML workloads.First, we analyze ML frameworks' runtime execution on a CPU-GPU tandem. We propose a new profiling methodology that leverages data from an ML framework's profiler. We use this methodology to provide new insights into the runtime execution of inference, for three ML models. Our results show that GPU kernels' execution must be long enough to hide the runtime overhead of the ML framework, increasing GPU utilization. However, this strive for longer kernel execution leads to the use of bigger batches of data, seemingly pushing the need for more GPU memory.Second, we analyze the utilization of GPU resources when performing ML training. We propose a new profiling methodology combining the use of high-level and low-level profilers to provide new insights into the utilization of the GPU's inner components. Our experiments, on two modern GPUs, suggest that bigger GPU memory helps enhance throughput and utilization from a high level. However, our results also suggest that a plateau has been reached, eliminating the push for bigger batches. Furthermore, we observe that the fastest GPU cores (tensor cores) are idle most of the time, and the tested workloads are now limited by kernels that do not use these cores. Thus, our results suggest that the current GPU paradigm is reaching a saturation point.Finally, we analyze the energy consumption of GPUs during ML training. We propose an energy model and calibration methodology that uses microbenchmarks to provide a breakdown of the GPU energy consumption. We implement and validate this approach with a modern NVIDIA GPU. Our results suggest that data movement is responsible for most of the energy consumption (up to 84% of the dynamic energy consumption of the GPU). This further motivates the push for newer architectures, optimizing memory accesses (e.g., processing in/near memory, vectorized architectures).This thesis provides a comprehensive analysis of the performance and energy bottlenecks of GPU-accelerated ML workloads. We believe our contributions uncover some of the limitations of current GPU architectures and motivate the need for more advanced profiling techniques to design more efficient ML accelerators. We hope that our work will inspire future research in this direction
5
Pouy, Léo. "OpenNas : un cadre adaptable de recherche automatique d'architecture neuronale". Electronic Thesis or Diss., université Paris-Saclay, 2023. http://www.theses.fr/2023UPASG089.
Texto completoResumen
Lors de la création d'un modèle de réseau de neurones, l'étape dite du "fine-tuning" est incontournable. Lors de ce fine-tuning, le développeur du réseau de neurones doit ajuster les hyperparamètres et l'architecture du réseau pour que ce-dernier puisse répondre au cahier des charges. Cette étape est longue, fastidieuse, et nécessite de l'expérience de la part du développeur. Ainsi, pour permettre la création plus facile de réseaux de neurones, il existe une discipline, l'"Automatic Machine Learning" (Auto-ML), qui cherche à automatiser la création de Machine Learning. Cette thèse s'inscrit dans cette démarche d'automatisation et propose une méthode pour créer et optimiser des architectures de réseaux de neurones (Neural Architecture Search). Pour ce faire, un nouvel espace de recherche basé sur l'imbrication de blocs à été formalisé. Cet espace permet de créer un réseau de neurones à partir de blocs élémentaires connectés en série ou en parallèle pour former des blocs composés qui peuvent eux-mêmes être connectés afin de former un réseau encore plus complexe. Cet espace de recherche à l'avantage d'être facilement personnalisable afin de pouvoir influencer la recherche automatique vers des types d'architectures (VGG, Inception, ResNet, etc.) et contrôler le temps d'optimisation. De plus il n'est pas contraint à un algorithme d'optimisation en particulier. Dans cette thèse, la formalisation de l'espace de recherche est tout d'abord décrite, ainsi que des techniques dîtes d'"encodage" afin de représenter un réseau de l'espace de recherche par un entier naturel (ou une liste d'entiers naturels). Puis, des stratégies d'optimisation applicables à cet espace de recherche sont proposées. Enfin, des expérimentations de recherches d'architectures neuronales sur différents jeux de données et avec différents objectifs en utilisant l'outil développé (nommé OpenNas) sont présentéesWhen creating a neural network, the "fine-tuning" stage is essential. During this fine-tuning, the neural network developer must adjust the hyperparameters and the architecture of the network so that it meets the targets. This is a time-consuming and tedious phase, and requires experience on the part of the developer. So, to make it easier to create neural networks, there is a discipline called Automatic Machine Learning (Auto-ML), which seeks to automate the creation of Machine Learning. This thesis is part of this Auto-ML approach and proposes a method for creating and optimizing neural network architectures (Neural Architecture Search, NAS). To this end, a new search space based on block imbrication has been formalized. This space makes it possible to create a neural network from elementary blocks connected in series or in parallel to form compound blocks which can themselves be connected to form an even more complex network. The advantage of this search space is that it can be easily customized to influence the NAS for specific architectures (VGG, Inception, ResNet, etc.) and control the optimization time. Moreover, it is not constrained to any particular optimization algorithm. In this thesis, the formalization of the search space is first described, along with encoding techniques to represent a network from the search space by a natural number (or a list of natural numbers). Optimization strategies applicable to this search space are then proposed. Finally, neural architecture search experiments on different datasets and with different objectives using the developed tool (named OpenNas) are presented
6
Bouraoui, Zied. "Inconsistency and uncertainty handling in lightweight description logics". Thesis, Artois, 2015. http://www.theses.fr/2015ARTO0408/document.
Texto completoResumen
Cette thèse étudie la dynamique des croyances et la gestion de l’incertitude dans DL-Lite, une des plus importantes familles des logiques de description légères. La première partie de la thèse porte sur la gestion de l’incertitude dans DL-Lite. En premier lieu, nous avons proposé une extension des principaux fragments de DL-Lite pour faire face à l’incertitude associée aux axiomes en utilisant le cadre de la théorie des possibilités. Cette extension est réalisée sans engendrer des coûts calculatoires supplémentaires. Nous avons étudié ensuite la révision des bases DL-Lite possibilistes en présence d’une nouvelle information. Enfin, nous avons proposé un opérateur de fusion lorsque les assertions de ABoxsont fournies par plusieurs sources d’information ayant différents niveaux de priorité. La deuxième partie de la thèse traite le problème de la gestion d’incohérence dans les bases de connaissances DL-Lite. Nous avons étudié, tout d’abord, comment raisonner à partir d’une base DL-Lite standard avec des ABox multiples en introduisant les notions de modificateurs et de stratégies d’inférence. La combinaison des modificateurs et de stratégies d’inférence fournit une liste exhaustive des principales techniques de gestion de l’incohérence. Nous avons proposé ensuite une approche, basée sur un critère de cardinalité, de sélection des réparations, et nous avons identifié les stratégies appropriées pour la gestion de l’incohérence pour les bases DL-Lite stratifiées. Enfin, nous avons effectué une analyse comparative, suivie par des étudesexpérimentales, des différentes techniques de gestion d’incohérence proposées. Finalement, un outil de représentation et de raisonnement à partir des bases DL-Lite possibiliste est réaliséThis thesis investigates the dynamics of beliefs and uncertainty management in DL-Lite, one of the most important lightweight description logics. The first part of the thesis concerns the problem of handling uncertainty in DL-Lite. First, we propose an extension of the main fragments of DL-Lite to deal with the uncertainty associated with axioms using a possibility theory framework without additional extra computational costs. We then study the revision of possibilistic DL-Lite bases when a new piece of information is available. Lastly, we propose a min-based assertional merging operator when assertions of ABox are provided by several sources of information having different levels of priority. The second partof the thesis concerns the problem of inconsistency handling in flat and prioritized DL-Lite knowledge bases. We first propose how to reason from a flat DL-Lite knowledge base, with a multiple ABox, which can be either issued from multiple information sources or resulted from revising DL-Lite knowledge bases. This is done by introducing the notions of modifiers and inference strategies. The combination of modifiers plus inference strategies can be mapped out in order to provide a principled and exhaustive list of techniques for inconsistency management. We then give an approach based on selecting multiple repairs using a cardinality-based criterion, and we identified suitable strategies for handling inconsistencyin the prioritized case. Lastly, we perform a comparative analysis, followed by experimental studies, of the proposed inconsistency handling techniques. A tool for representing and reasoning in possibilistic DL-Lite framework is implemented
7
Giuliani, Luca. "Extending the Moving Targets Method for Injecting Constraints in Machine Learning". Master's thesis, Alma Mater Studiorum - Università di Bologna, 2021. http://amslaurea.unibo.it/23885/.
Texto completoResumen
Informed Machine Learning is an umbrella term that comprises a set of methodologies in which domain knowledge is injected into a data-driven system in order to improve its level of accuracy, satisfy some external constraint, and in general serve the purposes of explainability and reliability. The said topid has been widely explored in the literature by means of many different techniques. Moving Targets is one such a technique particularly focused on constraint satisfaction: it is based on decomposition and bi-level optimization and proceeds by iteratively refining the target labels through a master step which is in charge of enforcing the constraints, while the training phase is delegated to a learner. In this work, we extend the algorithm in order to deal with semi-supervised learning and soft constraints. In particular, we focus our empirical evaluation on both regression and classification tasks involving monotonicity shape constraints. We demonstrate that our method is robust with respect to its hyperparameters, as well as being able to generalize very well while reducing the number of violations on the enforced constraints. Additionally, the method can even outperform, both in terms of accuracy and constraint satisfaction, other state-of-the-art techniques such as Lattice Models and Semantic-based Regularization with a Lagrangian Dual approach for automatic hyperparameter tuning.
8
Hmedoush, Iman. "Connectionless Transmission in Wireless Networks (IoT)". Electronic Thesis or Diss., Sorbonne université, 2022. https://accesdistant.sorbonne-universite.fr/login?url=https://theses-intra.sorbonne-universite.fr/2022SORUS143.pdf.
Texto completoResumen
L'origine concernant l'idée d'ajouter de l'intelligence aux objets de base et de les faire communiquer n'est pas connue précisément. Mais ces derniers temps, l'émergence d'Internet en tant que réseau de communication global a aussi motivé l'utilisation de son architecture et de ses protocoles pour connecter des objets. C'est par exemple le cas célèbre du distributeur automatique de sodas connecté à l'ARPANET dans les années 1980. Au cours des deux dernières décennies, de nombreuses améliorations technologiques ont été développées pour rendre possible l'Internet des objets (IoT). Un scénario d'un réseau IoT typique consiste à connecter des dispositifs embarqués composés de capteurs environnementaux, de microcontrôleurs et de matériel de communication à un nœud de collecte central. L'ensemble des données recueillies par ces nœuds permettra d'analyser et de comprendre précisément les phénomènes et comportements se produisant dans cet environnement. Les applications des technologies IoT sont infinies, car elles sont adaptables à presque tous les systèmes, que l'on doit surveiller et contrôler à distance, pouvant fournir des informations sur son état, son fonctionnement et son environnement. Les villes intelligentes, les soins, l'automatisation industrielle et la technologie portable sont quelques-unes des applications de l'IoT qui promettent de rendre notre vie plus sûre et plus facile. Certains défis en matière de recherche et de technologie doivent être relevés pour la mise en œuvre et la large dissémination des applications de l'IoT comme le déploiement, la mise en réseau, la sécurité, la résilience et le contrôle de l'alimentation des équipements. Cette demande massive de connexion dans les réseaux IoT introduit de nouveaux défis en termes de connectivité, de fiabilité et de technologie. Au niveau de la radio, les réseaux IoT représentent un énorme afflux de divers appareils qui communiquent via le même support radio partagé. Cependant, bon nombre de ces appareils sont difficiles à sécuriser et à manipuler. L'un des principaux défis du déploiement des réseaux IoT est le manque de solutions efficaces qui permettent un nombre massif de connexions tout en répondant en même temps aux exigences de faible latence et de faible coût. De plus, il y a eu récemment une tendance vers des systèmes de communication à longue portée pour l'IoT et aussi pour les réseaux cellulaires. Pour de nombreux cas d'utilisation, tels que les communications massives de type machine (mMTC), les performances peuvent être améliorées en s'éloignant du modèle classique d'établissement de connexion et en adoptant des méthodes d'accès aléatoire sans attribution prédéterminée. Associé à des techniques de couche physique telles que l'annulation successive des interférences (SIC) ou l'accès multiple non orthogonal (NOMA), les performances de l'accès aléatoire peuvent être améliorées, donnant lieu à de nouvelles conceptions de protocoles d'accès aléatoire. Dans cette thèse, nous nous concentrons sur l'un des candidats modernes pour les protocoles d'accès aléatoire bien adaptés à l'IoT :ALOHA à répétition irrégulière (IRSA). Comme des solutions sont nécessaires pour surmonter les défis de l'IoT, nous étudions le schéma d'accès aléatoire IRSA sous de nouveaux points de vue et nous commençons par une analyse des performances des différentes variantes grâce à l'outil de l'évolution de la densité du débit. Précisément, nous commençons par revisiter le scénario du protocole IRSA avec la capacité de réception de paquets multiples (MPR) au niveau du récepteur. Ensuite, nous étudions IRSA dans différents scénarios où des hypothèses plus réalistes sont considérées comme : IRSA avec plusieurs puissances de transmission, avec effet de capture et avec des erreurs de décodage. Dans la deuxième partie de la thèse, nous nous concentrons sur l'apprentissage et l'ajustement dynamique des paramètres du protocole IRSA. Dans un premier temps, nous analysons les performances [...]The origin of the idea of adding intelligence to basic objects and making them communicate has been lost to history. But in recent times, the emergence of the Internet as a global communication network has also motived the use of its architecture and protocols to connect objects (such as the soda vending machine famously connected to the ARPANET in the 1980s). In the past two decades, many technological enhancements have been developed to enable the ``Internet of Things'' (IoT). A scenario of a typical IoT network is to connect embedded devices composed of environmental sensors, microcontrollers, and communication hardware, to a central collection node. The set of data gathered by these nodes will increasingly help in analyzing and precisely understanding the phenomenons and behaviors occurring in this environment. The applications of IoT technologies are endless because they are adaptable to almost any system that can provide information about its status, operation, and the environment and that one needs to monitor and control at a distance. Smart cities, healthcare, industrial automation, and wearable technology are some IoT applications that promise to make our life safer and easier. Some research and technology challenges need to be addressed for the implementation and full popularization of IoT applications including deployment, networking, security, resilience, and power control. This massive demand for connection in IoT networks will introduce new challenges in terms of connectivity, reliability, and technology. At the radio network level, IoT networks represent a huge inflow of various devices that communicate through the same shared radio medium. However, many of these devices are difficult to secure and handle. One major challenge to deploying IoT networks is the lack of efficient solutions that allow for a massive number of connections while meeting the low-latency and low-cost demands at the same time. In addition, recently, there has been a trend towards long-range communications systems for the IoT, including cellular networks. For many use cases, such as massive machine-type communications (mMTC), performance can be gained by moving away from the classical model of connection establishment and adopting grant-free, random access methods. Associated with physical layer techniques such as Successive Interference Cancellation (SIC), or Non-Orthogonal Multiple Access (NOMA), the performance of random access can be dramatically improved, giving rise to novel random access protocol designs. In this thesis, we focus on one of the modern candidates for random access protocols ``well-fitted'' to the IoT: Irregular Repetition Slotted ALOHA (IRSA). As solutions are needed to overcome the challenges of IoT, we study the IRSA random access scheme from new points of view and we start with an analysis of the performance of different variations through the density evolution tool. Precisely, we start by revisiting the scenario of the IRSA protocol in the case of Multiple Packet Reception (MPR) capability at the receiver. Then, we study IRSA in different scenarios where more realistic assumptions are considered, such as IRSA with multiple transmissions powers, with capture effect, and with decoding errors. In the second part of the thesis, we concentrate on learning and dynamically adjusting IRSA protocol parameters. First, we analyze the protocol performance in a centralized approach through a variant of Reinforcement Learning and in a distributed approach through Game Theory. We also optimize short frame length IRSA through a Deep Reinforcement Learning approach. Finally, we introduce a sensing capability to IRSA, in line with carrier sense principles, and we tentatively explore how one can learn part of sensing protocols with the help of Deep Learning tools
9
Kaplan, Caelin. "Compromis inhérents à l'apprentissage automatique préservant la confidentialité". Electronic Thesis or Diss., Université Côte d'Azur, 2024. http://www.theses.fr/2024COAZ4045.
Texto completoResumen
À mesure que les modèles d'apprentissage automatique (ML) sont de plus en plus intégrés dans un large éventail d'applications, il devient plus important que jamais de garantir la confidentialité des données des individus. Cependant, les techniques actuelles entraînent souvent une perte d'utilité et peuvent affecter des facteurs comme l'équité et l'interprétabilité. Cette thèse vise à approfondir la compréhension des compromis dans trois techniques de ML respectueuses de la vie privée : la confidentialité différentielle, les défenses empiriques, et l'apprentissage fédéré, et à proposer des méthodes qui améliorent leur efficacité tout en maintenant la protection de la vie privée. La première étude examine l'impact de la confidentialité différentielle sur l'équité entre les groupes définis par des attributs sensibles. Alors que certaines hypothèses précédentes suggéraient que la confidentialité différentielle pourrait exacerber l'injustice dans les modèles ML, nos expériences montrent que la sélection d'une architecture de modèle optimale et le réglage des hyperparamètres pour DP-SGD (Descente de Gradient Stochastique Différentiellement Privée) peuvent atténuer les disparités d'équité. En utilisant des ensembles de données standards dans la littérature sur l'équité du ML, nous montrons que les disparités entre les groupes pour les métriques telles que la parité démographique, l'égalité des chances et la parité prédictive sont souvent réduites ou négligeables par rapport aux modèles non privés. La deuxième étude se concentre sur les défenses empiriques de la vie privée, qui visent à protéger les données d'entraînement tout en minimisant la perte d'utilité. La plupart des défenses existantes supposent l'accès à des données de référence — un ensemble de données supplémentaire provenant de la même distribution (ou similaire) que les données d'entraînement. Cependant, les travaux antérieurs n'ont que rarement évalué les risques de confidentialité associés aux données de référence. Pour y remédier, nous avons réalisé la première analyse complète de la confidentialité des données de référence dans les défenses empiriques. Nous avons proposé une méthode de défense de référence, la minimisation du risque empirique pondéré (WERM), qui permet de mieux comprendre les compromis entre l'utilité du modèle, la confidentialité des données d'entraînement et celle des données de référence. En plus d'offrir des garanties théoriques, WERM surpasse régulièrement les défenses empiriques de pointe dans presque tous les régimes de confidentialité relatifs. La troisième étude aborde les compromis liés à la convergence dans les systèmes d'inférence collaborative (CIS), de plus en plus utilisés dans l'Internet des objets (IoT) pour permettre aux nœuds plus petits de décharger une partie de leurs tâches d'inférence vers des nœuds plus puissants. Alors que l'apprentissage fédéré (FL) est souvent utilisé pour entraîner conjointement les modèles dans ces systèmes, les méthodes traditionnelles ont négligé la dynamique opérationnelle, comme l'hétérogénéité des taux de service entre les nœuds. Nous proposons une approche FL novatrice, spécialement conçue pour les CIS, qui prend en compte les taux de service variables et la disponibilité inégale des données. Notre cadre offre des garanties théoriques et surpasse systématiquement les algorithmes de pointe, en particulier dans les scénarios où les appareils finaux gèrent des taux de requêtes d'inférence élevés. En conclusion, cette thèse contribue à l'amélioration des techniques de ML respectueuses de la vie privée en analysant les compromis entre confidentialité, utilité et autres facteurs. Les méthodes proposées offrent des solutions pratiques pour intégrer ces techniques dans des applications réelles, en assurant une meilleure protection des données personnellesAs machine learning (ML) models are increasingly integrated into a wide range of applications, ensuring the privacy of individuals' data is becoming more important than ever. However, privacy-preserving ML techniques often result in reduced task-specific utility and may negatively impact other essential factors like fairness, robustness, and interpretability. These challenges have limited the widespread adoption of privacy-preserving methods. This thesis aims to address these challenges through two primary goals: (1) to deepen the understanding of key trade-offs in three privacy-preserving ML techniques—differential privacy, empirical privacy defenses, and federated learning; (2) to propose novel methods and algorithms that improve utility and effectiveness while maintaining privacy protections. The first study in this thesis investigates how differential privacy impacts fairness across groups defined by sensitive attributes. While previous assumptions suggested that differential privacy could exacerbate unfairness in ML models, our experiments demonstrate that selecting an optimal model architecture and tuning hyperparameters for DP-SGD (Differentially Private Stochastic Gradient Descent) can mitigate fairness disparities. Using standard ML fairness datasets, we show that group disparities in metrics like demographic parity, equalized odds, and predictive parity are often reduced or remain negligible when compared to non-private baselines, challenging the prevailing notion that differential privacy worsens fairness for underrepresented groups. The second study focuses on empirical privacy defenses, which aim to protect training data privacy while minimizing utility loss. Most existing defenses assume access to reference data---an additional dataset from the same or a similar distribution as the training data. However, previous works have largely neglected to evaluate the privacy risks associated with reference data. To address this, we conducted the first comprehensive analysis of reference data privacy in empirical defenses. We proposed a baseline defense method, Weighted Empirical Risk Minimization (WERM), which allows for a clearer understanding of the trade-offs between model utility, training data privacy, and reference data privacy. In addition to offering theoretical guarantees on model utility and the relative privacy of training and reference data, WERM consistently outperforms state-of-the-art empirical privacy defenses in nearly all relative privacy regimes.The third study addresses the convergence-related trade-offs in Collaborative Inference Systems (CISs), which are increasingly used in the Internet of Things (IoT) to enable smaller nodes in a network to offload part of their inference tasks to more powerful nodes. While Federated Learning (FL) is often used to jointly train models within CISs, traditional methods have overlooked the operational dynamics of these systems, such as heterogeneity in serving rates across nodes. We propose a novel FL approach explicitly designed for CISs, which accounts for varying serving rates and uneven data availability. Our framework provides theoretical guarantees and consistently outperforms state-of-the-art algorithms, particularly in scenarios where end devices handle high inference request rates.In conclusion, this thesis advances the field of privacy-preserving ML by addressing key trade-offs in differential privacy, empirical privacy defenses, and federated learning. The proposed methods provide new insights into balancing privacy with utility and other critical factors, offering practical solutions for integrating privacy-preserving techniques into real-world applications. These contributions aim to support the responsible and ethical deployment of AI technologies that prioritize data privacy and protection
10
Callebert, Lucile. "Activités collaboratives et génération de comportements d'agents : moteur décisionnel s'appuyant sur un modèle de confiance". Thesis, Compiègne, 2016. http://www.theses.fr/2016COMP2299/document.
Texto completoResumen
Lorsqu’ils travaillent en équipe, les humains ont rarement des comportements optimaux : ils peuvent faire des erreurs, manquer de motivation ou de compétence. Dans les domaines des environnements virtuels ou des systèmes multi-agents, de nombreux travaux ont cherché à reproduire les comportements d’équipes humaines : un agent représente alors un membre de l’équipe. Cependant, ces travaux ont très souvent pour objectif la performance de l’équipe, et non la fidélité des comportements produits. Pour former un apprenant en environnement virtuel à prêter attention et à s’adapter aux autres, nous avons cherché dans cette thèse à reproduire des comportements humains réalistes et non-optimaux de travail d’équipe. Plus particulièrement, nous nous sommes intéressés aux équipes auto-organisées, c’est-à-dire aux équipes dans lesquelles le pouvoir de décision est réparti entre les membres, et dans lesquelles l’organisation est implicite. Dans de telles équipes, l’organisation se fait non pas au travers des communications mais par l’observation et l’anticipation des comportements des autres. Pour s’organiser, chaque agent doit se demander ce qu’il est préférable de faire en fonction de ce que pourraient faire les autres, et donc se poser des questions telles que Ai-je confiance en la compétence de mon coéquipier pour faire cette tâche ? Les relations de confiance permettent donc à chacun de prendre en compte les autres. Pour générer de tels comportements, nous proposons un système permettant aux agents de raisonner d’une part sur un modèle de l’activité à effectuer et d’autre part sur les relations de confiance qui les lient aux autres agents de l’environnement. Dans ce cadre, notre première contribution porte sur l’augmentation du langage de description de l’activité Activity-DL de manière à permettre la description d’activités collectives. Nous proposons également des mécanismes de propagation de contraintes et d’informations qui faciliteront le raisonnement des agents. Ces contraintes et informations permettront par exemple aux agents de répondre à la question Avons-nous les compétences nécessaires pour faire la tâche qui permettra de réaliser notre but collectif ? Notre seconde contribution porte sur la proposition d’un modèle d’agent opérationnalisant le modèle de confiance de Mayer et al. (1995), sélectionné après une étude de la littérature en sciences humaines et sociales sur la confiance. La confiance d’un agent en un autre est décrite selon trois dimensions : l’intégrité, la bienveillance et les compétences. Chaque agent est donc défini par ces trois dimensions et a des croyances sur l’intégrité, la bienveillance et la compétence des autres agents. De plus chaque agent possède des buts qui lui sont personnels ainsi que des buts collectifs et devra donc choisir quel but privilégier. Finalement nous proposons un moteur décisionnel qui permet à chaque agent de calculer l’importance qu’il accorde à ses buts afin de sélectionner une tâche. Nous avons défini les mécanismes de calcul de l’importance des buts de manière à modéliser l’influence sur l’agent de ses croyances sur les autres, et pour sélectionner une tâche, l’agent raisonne à la fois sur les modèles d’activité et sur ses attentes à propos du comportement des autres, également générées à partir des croyances de l’agent sur les autres. Nous avons implémenté notre système et constaté qu’il répond à nos objectifs de génération de comportements d’équipe réalistes et non optimaux. Nous avons également conduit une évaluation perceptive préliminaire au cours de laquelle les participants ont notamment été capables de percevoir la confiance ou le manque de confiance d’un agent en un autre grâce à son comportementWhen working in teams, humans rarely display optimal behaviors: they sometimes make mistakes, lack motivation or competence. In virtual environments or in multi-agent systems, many studies have tried to reproduce human teamwork: each agent acts as a team member. However, the main objective in those studies is the performance of the team: each agent should display optimal behavior, and the realism of those simulated behaviors is not a concern. To train someone in a virtual environment to pay attention to and to adapt to their teammates, we built a decision-making system for agents to display realistic and non-optimal behaviors. More specifically, we are interested in self-organized teams (i.e. teams where the decision power is decentralized among its members) and in implicit organization (i.e. when team members do not interact through communications but rather through the observation of others’ behaviors). In such a team, each agent has to think about what it should do given what others could do. Agents then have to ask themselves questions such as Do I trust my teammate’s competence to perform this task? Trust relationships therefore allow agents to take others into account. We propose a system that allows agents to reason, on the first hand, on models of the activity they have to do, and on the other hand, on trust relationships they share with others. In that context, we first augmented the Activity-Description Language so that it supports the description of collective activities. We also defined mechanisms for constraint generation that facilitates agent reasoning, by giving them the answer to questions like Do we have the required abilities to perform the task which will achieve our goal? We then proposed an agent model based on the model of interpersonal trust of Mayer et al. (1995) that we selected after a study of trust in social science. This model describes trust relationship with three dimensions: the trustor trusts the trustee’s integrity, benevolence and abilities. An agent is therefore defined through those three dimensions, and has a mental model of each other agent; i.e. has trust beliefs about others’ integrity, benevolence and abilities. Moreover each agent has both personal and collective goals (i.e. goals that are shared with other members of the team), and thus will have to decide which goal to focus on. Finally we proposed a decision-making system that allows an agent to compute the importance it gives to its goals and then to select a task. When computing goal importance, the agent is influenced by its trust beliefs about others, and to select a task, it reasons on the activity models and on its expectations about what others could do. Those expectations are generated from the agents’ trust beliefs. We implemented our system and observed that it produces realistic and non-optimal behaviors. We also conducted a preliminary perceptive evaluation which showed that participants were able to recognize one agent’s trust or lack of trust in another through the behaviors of the first one
Capítulos de libros sobre el tema "Intelligence artificielle (ML/DL)"
1
Rajendran, Sindhu, Alen Aji John, B. Suhas y B. Sahana. "Role of ML and DL in Detecting Fraudulent Transactions". En Artificial Intelligence for Societal Issues, 59–82. Cham: Springer International Publishing, 2023. http://dx.doi.org/10.1007/978-3-031-12419-8_4.
Texto completo
2
Wittenberg, Thomas, Thomas Lang, Thomas Eixelberger y Roland Grube. "Acquisition of Semantics for Machine-Learning and Deep-Learning based Applications". En Unlocking Artificial Intelligence, 153–75. Cham: Springer Nature Switzerland, 2024. http://dx.doi.org/10.1007/978-3-031-64832-8_8.
Texto completoResumen
AbstractFor the development, training, and validation of machine learning (ML) and deep learning (DL) based methods, such as, e.g., image analysis, prediction of critical events, extraction or reconstruction of information from disrupted data streams, searching similarities in data collections, or planning of procedures, a lot of data is needed. Additionally to this data (images, bio-signals, vital-signs, text records, machine states, trajectories, antenna data, ...) adequate supplementary information about the meaning encoded in the data is required. Only with this additional information – the meaning or knowledge – a tight relation between the raw data and the human-understandable concepts – the semantics – from the real world can be established. Nevertheless, as the amount of data needed to develop robust ML or DL methods is strongly increasing, the assessment and acquisition of the related knowledge becomes more and more challenging. Within this chapter, an overview of concepts of knowledge acquisition applied to the different examples of applications is described and evaluated. Six main groups of knowledge acquisition related to AI-based technologies have been identified, namely (1) manual annotation methods, (2) data augmentation, (3) generative networks or simulation techniques, (4) synchronized sensors, (5)Active Learning approaches, and (6) explicit knowledge modeling using semantic networks.
3
Gadri, Said y Erich Neuhold. "Building Best Predictive Models Using ML and DL Approaches to Categorize Fashion Clothes". En Artificial Intelligence and Soft Computing, 90–102. Cham: Springer International Publishing, 2020. http://dx.doi.org/10.1007/978-3-030-61401-0_9.
Texto completo
4
Das, Priya y Sohail Saif. "Intrusion Detection in IoT-Based Healthcare Using ML and DL Approaches: A Case Study". En Artificial Intelligence and Cyber Security in Industry 4.0, 271–94. Singapore: Springer Nature Singapore, 2023. http://dx.doi.org/10.1007/978-981-99-2115-7_12.
Texto completo
5
Kotios, Dimitrios, Georgios Makridis, Silvio Walser, Dimosthenis Kyriazis y Vittorio Monferrino. "Personalized Finance Management for SMEs". En Big Data and Artificial Intelligence in Digital Finance, 215–32. Cham: Springer International Publishing, 2012. http://dx.doi.org/10.1007/978-3-030-94590-9_12.
Texto completoResumen
AbstractThis chapter presents Business Financial Management (BFM) tools for Small Medium Enterprises (SMEs). The presented tools represent a game changer as they shift away from a one-size-fits-all approach to banking services and put emphasis on delivering a personalized SME experience and an improved bank client’s digital experience. An SME customer-centric approach, which ensures that the particularities of the SME are taken care of as much as possible, is presented. Through a comprehensive view of SMEs’ finances and operations, paired with state-of-the-art ML/DL models, the presented BFM tools act as a 24/7 concierge. They also operate as a virtual smart advisor that delivers in a simple, efficient, and engaging way business insights to the SME at the right time, i.e., when needed most. Deeper and better insights that empower SMEs contribute toward SMEs’ financial health and business growth, ultimately resulting in high-performance SMEs.
6
Trocin, Cristina, Jan Gunnar Skogås, Thomas Langø y Gabriel Hanssen Kiss. "Operating Room of the Future (FOR) Digital Healthcare Transformation in the Age of Artificial Intelligence". En Digital Transformation in Norwegian Enterprises, 151–72. Cham: Springer International Publishing, 2022. http://dx.doi.org/10.1007/978-3-031-05276-7_9.
Texto completoResumen
AbstractNew technologies are emerging under the umbrella of digital transformation in healthcare such as artificial intelligence (AI) and medical analytics to provide insights beyond the abilities of human experts. Because AI is increasingly used to support doctors in decision-making, pattern recognition, and risk assessment, it will most likely transform healthcare services and the way doctors deliver those services. However, little is known about what triggers such transformation and how the European Union (EU) and Norway launch new initiatives to foster the development of such technologies. We present the case of Operating Room of the Future (FOR), a research infrastructure and an integrated university clinic which investigates most modern technologies such as artificial intelligence (AI), machine learning (ML), and deep learning (DL) to support the analysis of medical images. Practitioners can benefit from strategies related to AI development in multiple health fields to best combine medical expertise with AI-enabled computational rationality.
7
Das, Pritam, Hakam Singh, Nilamadhab Mishra, Nagesh Kumar, Ramamani Tripathy, Rudra Kalyan Nayak y Saroja Kumar Rout. "The Impact and Evolution of Deep Learning in Contemporary Real-World Predictive Applications". En Advances in Computational Intelligence and Robotics, 1–32. IGI Global, 2024. https://doi.org/10.4018/979-8-3693-6230-3.ch001.
Texto completoResumen
Deep learning (DL) is making a significant impact on the lives of human beings, either directly or indirectly; we benefit from Artificial Intelligence (AI), Machine Learning (ML), DL and other technologies/networks in our day-to-day lives. DL not only can mimic a human brain and carry out tasks like a human being but has also outworked the approaches of ML, making itself the most efficient technology used nowadays. Maybe this is the reason why various fields like Cybersecurity, Medical Treatments, Traffic Control, Weather Forecast, Bioinformatics, Fraud Detection, Robotics, Vocal AI, Computer Vision, Autonomous Vehicles, E-Commerce and so on are using DL techniques/algorithms rather than the traditional ML approach. This review will illuminate the History of DL, i.e. what DL is, how and why it has evolved so far, and what the various parts of DL are. Though DL is considered the most efficient, it has some limitations, and we will investigate them gradually. We will also study all pre-researched DL Techniques and their implications for contemporary real-world predictive applications.
8
Nagula, Jagan Mohan, Murugan R. y Tripti Goel. "Role of Machine and Deep Learning Techniques in Diabetic Retinopathy Detection". En Multidisciplinary Applications of Deep Learning-Based Artificial Emotional Intelligence, 32–46. IGI Global, 2022. http://dx.doi.org/10.4018/978-1-6684-5673-6.ch003.
Texto completoResumen
Machine learning (ML) and deep learning (DL) techniques play a significant role in diabetic retinopathy (DR) detection via grading the severity levels or segmenting the retinal lesions. High sugar levels in the blood due to diabetes causes DR, a leading cause of blindness. Manual detection or grading of the DR requires ophthalmologists' expertise and consumes time prone to human errors. Therefore, using fundus images, the ML and DL algorithms help automatic DR detection. The fundus imaging analysis helps the early DR detection, controlling, and treatment evaluation of DR conditions. Knowing the fundus image analysis requires a strong knowledge of the system and ML and DL functionalities in computer vision. DL in fundus imaging is a rapidly expanding research area. This chapter presents the fundus images, DR, and its severity levels. Also, this chapter explains the performance analysis of the various ML DL-based DR detection techniques. Finally, the role of ML and DL techniques in DR detection or severity grading is discussed.
9
Andrae, Silvio. "The Use of Artificial Intelligence to Curb Deforestation in the Brazilian Rainforest". En Artificial Intelligence and Data Science for Sustainability, 81–122. IGI Global, 2025. https://doi.org/10.4018/979-8-3693-6829-9.ch004.
Texto completoResumen
Tropical rainforests like the Amazon are invaluable ecosystems for human society and biodiversity. However, they are facing unprecedented threats, primarily from deforestation. This chapter explores the use of machine learning (ML) and deep learning (DL) to address this pressing environmental problem. By analyzing different ML/DL methods, we show how these tools can be used to understand deforestation patterns in the Brazilian Amazon better. Specifically, we discuss how ML/DL can help identify the drivers of deforestation, improve remote sensing-based monitoring, and predict future deforestation trends. Our results, particularly the role of ML/DL in providing actionable insights, empower decision-makers and policymakers with the knowledge to make informed choices. Ultimately, these strategies contribute to more effective forest conservation measures and sustainable land use, reassuring the audience about the reliability of our research.
10
Mishra, Alok y Preeti Mishra. "MACHINE AND DEEP LEARNING APPLICATIONS: ADVANCEMENTS, CHALLENGES, AND FUTURE DIRECTIONS". En Futuristic Trends in Artificial Intelligence Volume 3 Book 1, 89–98. Iterative International Publisher, Selfypage Developers Pvt Ltd, 2024. http://dx.doi.org/10.58532/v3bfai1p1ch8.
Texto completoResumen
Advancements in the realm of artificial intelligence (AI) have been nothing short of extraordinary, and the fields of machine learning (ML) and deep learning (DL) have been at the forefront, driving transformative changes in various sectors such as computer vision, natural language processing, healthcare, finance, and autonomous systems. This research paper presents a comprehensive survey of the most current applications of ML and DL techniques, engages in a discourse regarding the difficulties tied to their execution, and explores the future trajectories within this domain. It encompasses a detailed examination of essential ML and DL algorithms, architectures, and methodologies, underscoring their pragmatic utility and societal ramifications. By conducting an exhaustive review of pertinent literature and research endeavors, the core objective of this scholarly endeavor is to elucidate the progress, obstacles, and untapped potential of ML and DL in catalyzing innovation and addressing complex challenges.
Actas de conferencias sobre el tema "Intelligence artificielle (ML/DL)"
1
Arunachalam, N., S. Rukmani Devi, Niyas Ahamed A, Nazrin Salma S y M. Meikandan. "IoT Wearable Medical Device for Heart Disease Recognition Based Ml and Dl: A Bovw Based MDCNN Classification Approach". En 2024 International Conference on Intelligent Algorithms for Computational Intelligence Systems (IACIS), 1–6. IEEE, 2024. http://dx.doi.org/10.1109/iacis61494.2024.10721749.
Texto completo
2
Muniraj, Inbarasan. "Investigating the efficacy of deep learning networks for 3D imaging and processing". En 3D Image Acquisition and Display: Technology, Perception and Applications, DW1H.4. Washington, D.C.: Optica Publishing Group, 2024. http://dx.doi.org/10.1364/3d.2024.dw1h.4.
Texto completoResumen
Artificial intelligence techniques, such as machine learning (ML) and deep learning (DL), are now widely used in various vision-based applications. Here, we summarize some of the most recent advances in Computational Integral Imaging using DL networks.
3
Khan, Ibrahim y Zahid Ahmed. "ML and DL Classifications of Route Conditions Using Accelerometers and Gyroscope Sensors". En 2023 3rd International Conference on Artificial Intelligence (ICAI). IEEE, 2023. http://dx.doi.org/10.1109/icai58407.2023.10136666.
Texto completo
4
Sekhar, Ch, K. Pavani y M. Srinivasa Rao. "Comparative analysis on Intrusion Detection system through ML and DL Techniques: Survey". En 2021 International Conference on Computational Intelligence and Computing Applications (ICCICA). IEEE, 2021. http://dx.doi.org/10.1109/iccica52458.2021.9697291.
Texto completo
5
Wang, Han y Zefeng Li*. "The application of machine learning and deep learning to Ophthalmology: A bibliometric study (2000-2021)". En Human Interaction and Emerging Technologies (IHIET-AI 2022) Artificial Intelligence and Future Applications. AHFE International, 2022. http://dx.doi.org/10.54941/ahfe100885.
Texto completoResumen
Machine learning (ML) and deep learning (DL) are an advanced technology for the latest 20 years, which has been applied for multiple fields. This study utilizes methods of text mining and bibliometric analysis to explore applications of ML and DL to Ophthalmology. 50-ML-related and 60-DL-related papers from Web of Science (WOS), 15-ML-related and 38-DL-related articles from China Nation-al Knowledge Infrastructure (CNKI) are explored from 2000 to 2021. A descriptive analysis of major article, developing trends, journal releasing, topic mapping and quotation relationships is implemented in this paper. Leading authors, institutions, and journals in the related research are identified. Findings show that there is a significant difference between DL and ML studies pertaining to the application of Artificial Intelligence (AI) for Ophthalmology, especially for the hot-topic mapping and institutions.
6
Palei, Shantilata, Rakesh Kumar Lenka, Swoyam Siddharth Nayak, Rohan Mohanty, Biswajit Jena y Sanjay Saxena. "Precision Agriculture: ML and DL-Based Detection and Classification of Agricultural Pests". En 2023 2nd International Conference on Ambient Intelligence in Health Care (ICAIHC). IEEE, 2023. http://dx.doi.org/10.1109/icaihc59020.2023.10431427.
Texto completo
7
Rani, Jyoti, Jaswinder Singh y Jitendra Virmani. "Mammographic mass Classification using DL based ROI segmentation and ML based Classification". En 2023 International Conference on Device Intelligence, Computing and Communication Technologies, (DICCT). IEEE, 2023. http://dx.doi.org/10.1109/dicct56244.2023.10110098.
Texto completo
8
Shankar, T., Gummadapu Sreelekha, Challa Sai Tejaswini, P. Sivasankar, N. Lavanya y J. Murali. "Prediction of Parkinson’s disease with various ML and DL techniques on speech data". En 2024 3rd International Conference on Artificial Intelligence For Internet of Things (AIIoT). IEEE, 2024. http://dx.doi.org/10.1109/aiiot58432.2024.10574537.
Texto completo
9
Shukla, Jyoti S. y Rahul Jashvantbhai Pandya. "Predictive Modeling of Vegetative Drought Using ML/DL Approach on Temporal Satellite Data". En 2023 IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT). IEEE, 2023. http://dx.doi.org/10.1109/iaict59002.2023.10205851.
Texto completo
10
El-Attar, Noha E. y Yehia A. El-Mashad. "Artificial intelligence models for genomics analysis: review article". En Agria Média 2023 és ICI-17 Információ- és Oktatástechnológiai konferencia, 134–50. Eszterházy Károly Katolikus Egyetem Líceum Kiadó, 2024. http://dx.doi.org/10.17048/am.2023.134.
Texto completoResumen
Artificial intelligence (AI) including machine learning (ML), and deep learning (DL) models have become powerful tools for analyzing genomics data in recent years. These models can process large amounts of data and identify complex patterns that may not be apparent through traditional statistical methods. ML and DL models have been used for a wide range of genomics applications, including gene expression analysis, variant detection, and drug discovery. One popular approach for using ML and DL models in genomics is to train these models on large datasets of genomic information. These datasets may include information on gene expression levels, DNA sequences, and epigenetic modifications. By training these models on large datasets, researchers can identify patterns and correlations that may be used to predict disease risk, identify potential drug targets, and develop personalized treatments. Generally, the use of different AI models in genomics has the potential to transform the field by enabling more accurate and personalized medical treatments. As these models continue to evolve and improve, researchers will be able to extract even more information from genomic data and accelerate the pace of discovery in genomics.
Informes sobre el tema "Intelligence artificielle (ML/DL)"
1
Alhasson, Haifa F. y Shuaa S. Alharbi. New Trends in image-based Diabetic Foot Ucler Diagnosis Using Machine Learning Approaches: A Systematic Review. INPLASY - International Platform of Registered Systematic Review and Meta-analysis Protocols, noviembre de 2022. http://dx.doi.org/10.37766/inplasy2022.11.0128.
Texto completoResumen
Review question / Objective: A significant amount of research has been conducted to detect and recognize diabetic foot ulcers (DFUs) using computer vision methods, but there are still a number of challenges. DFUs detection frameworks based on machine learning/deep learning lack systematic reviews. With Machine Learning (ML) and Deep learning (DL), you can improve care for individuals at risk for DFUs, identify and synthesize evidence about its use in interventional care and management of DFUs, and suggest future research directions. Information sources: A thorough search of electronic databases such as Science Direct, PubMed (MIDLINE), arXiv.org, MDPI, Nature, Google Scholar, Scopus and Wiley Online Library was conducted to identify and select the literature for this study (January 2010-January 01, 2023). It was based on the most popular image-based diagnosis targets in DFu such as segmentation, detection and classification. Various keywords were used during the identification process, including artificial intelligence in DFu, deep learning, machine learning, ANNs, CNNs, DFu detection, DFu segmentation, DFu classification, and computer-aided diagnosis.