Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Des obstacles aux stimulants : Optimiser l'ouverture des données gouvernementales grâce à des orientations de publication et des stratégies de protection des données
No full textInternational audienceOpen Government Data (OGD) is a global endeavor, a collaborative effort between governments worldwide to share datasets that encapsulate a wide spectrum of government activities, from environmental issues like pollution and climate to social aspects like education and childcare, and urban concerns like traffic and congestion, and healthcare statistics. As governments, being among the largest producers and collectors of data, are making OGD available online in diverse formats, primarily Word, PDF, or Excel, they are contributing significantly to this global initiative. The OGD initiative holds immense potential to revolutionize the way we access and use government data. Its primary objective is to enhance the discoverability, accessibility, and availability of data in alternative and preferably machine-readable formats. This, in turn, empowers a diverse set of stakeholders to develop innovative data applications under licensing schemes that permit unrestricted reuse. Despite these promising aspects, challenges such as data heterogeneity, data protection, data quality, and data provenance issues persist. This study aims to analyze and categorize these challenges and obstacles that hinder the OGD initiative from realizing its full potential, with a particular emphasis on data protection and security concerns for data providers.L'Open Government Data (OGD) est une initiative mondiale, un effort de collaboration entre les gouvernements du monde entier pour partager des ensembles de données qui englobent un large éventail d'activités gouvernementales, des questions environnementales telles que la pollution et le climat aux aspects sociaux tels que l'éducation et la garde d'enfants, en passant par les préoccupations urbaines telles que le trafic et la congestion, et les statistiques sur les soins de santé. Les gouvernements, qui comptent parmi les plus grands producteurs et collecteurs de données, mettent à disposition les autres ministères en ligne dans divers formats, principalement Word, PDF ou Excel, et contribuent ainsi de manière significative à cette initiative mondiale. L'initiative OGD recèle un immense potentiel pour révolutionner la manière dont nous accédons aux données gouvernementales et les utilisons. Son principal objectif est d'améliorer la découverte, l'accessibilité et la disponibilité des données dans des formats alternatifs et de préférence lisibles par machine. Cela permet à un ensemble diversifié de parties prenantes de développer des applications de données innovantes dans le cadre de régimes de licence permettant une réutilisation sans restriction. Malgré ces aspects prometteurs, des défis tels que l'hétérogénéité des données, la protection des données, la qualité des données et les problèmes de provenance des données persistent. Cette étude a pour but d'analyser et de classer ces défis et obstacles qui empêchent l'initiative OGD de réaliser son plein potentiel, en mettant particulièrement l'accent sur la protection des données et les problèmes de sécurité pour les fournisseurs de données.Traduit avec DeepL.com (version gratuite
Total synthesis of cyclotripeptidic natural products anacine, aurantiomide C, polonimides A and C, and verrucine F (Communication orale)
No full textNational audienc
On the convergence analysis of one-shot inversion methods
No full textInternational audienceWhen an inverse problem is solved by a gradient-based optimization algorithm, the corresponding forward and adjoint problems, which are introduced to compute the gradient, can be also solved iteratively. The idea of iterating at the same time on the inverse problem unknown and on the forward and adjoint problem solutions yields the concept of one-shot inversion methods. We are especially interested in the case where the inner iterations for the direct and adjoint problems are incomplete, that is, stopped before achieving a high accuracy on their solutions. Here, we focus on general linear inverse problems and generic fixed-point iterations for the associated forward problem. We analyze variants of the so-called multi-step one-shot methods, in particular semi-implicit schemes with a regularization parameter. We establish sufficient conditions on the descent step for convergence, by studying the eigenvalues of the block matrix of the coupled iterations. Several numerical experiments are provided to illustrate the convergence of these methods in comparison with the classical gradient descent, where the forward and adjoint problems are solved exactly by a direct solver instead. We observe that very few inner iterations are enough to guarantee good convergence of the inversion algorithm, even in the presence of noisy data
Encoding the Latent Posterior of Bayesian Neural Networks for Uncertainty Quantification
No full textInternational audienceBayesian Neural Networks (BNNs) have long been considered an ideal, yet unscalable solution for improving the robustness and the predictive uncertainty of deep neural networks. While they could capture more accurately the posterior distribution of the network parameters, most BNN approaches are either limited to small networks or rely on constraining assumptions, e.g., parameter independence. These drawbacks have enabled prominence of simple, but computationally heavy approaches such as Deep Ensembles, whose training and testing costs increase linearly with the number of networks. In this work we aim for efficient deep BNNs amenable to complex computer vision architectures, e.g., ResNet-50 DeepLabv3+, and tasks, e.g., semantic segmentation and image classification, with fewer assumptions on the parameters. We achieve this by leveraging variational autoencoders (VAEs) to learn the interaction and the latent distribution of the parameters at each network layer. Our approach, called Latent-Posterior BNN (LP-BNN), is compatible with the recent BatchEnsemble method, leading to highly efficient (in terms of computation and memory during both training and testing) ensembles. LP-BNNs attain competitive results across multiple metrics in several challenging benchmarks for image classification, semantic segmentation, and out-of-distribution detection
Non-Invasive Health Systems based on Advanced Biomedical Signal and Image Processing
No full textInternational audienceThis book contains up-to-date noninvasive monitoring and diagnosing systems closely developed by a set of scientists, engineers, and physicians. The chapters are the results of different biomedical projects and theoretical studies that were coupled by simulations and real-world data. Non-Invasive Health Systems based on Advanced Biomedical Signal and Image Processing provides a multifaceted view of various biomedical and clinical approaches to health monitoring systems. The authors introduce advanced signal- and image-processing techniques as well as other noninvasive monitoring and diagnostic systems such as inertial sensors in wearable devices and novel algorithm-based hybrid learning systems for biosignal processing. The book includes a discussion of designing electronic circuits and systems for biomedical applications and analyzes several issues related to real-world data and how they relate to health technology including ECG signal monitoring and processing in the operating room. The authors also include detailed discussions of different systems for monitoring various conditions and diseases including sleep apnea, skin cancer, deep vein thrombosis, and prosthesis controls. This book is intended for a wide range of readers including scientists, researchers, physicians, and electronics and biomedical engineers. It will cover the gap between theory and real life applications
Rubber cord adhesion inflation test: Influence of envelope/confinement friction
No full textInternational audienceThe Rubber Cord Adhesion Inflation Test (RCAIT), has been previously introduced as an alternative test protocol to traditional pull-out tests for measuring the toughness of rubber matrix to metal reinforcement interfaces. A pressurized fluid is injected in between the matrix and reinforcement in a non-adherend region to initiate and propagate a crack along the specimen. A self-similar crack propagation regime hypothesis is assumed and the critical strain energy release rate is obtained from a simple energy balance analysis. However, to force the crack propagation along the specimen and prevent the development an aneurism in the rubber enveloppe, a confinement tube is used. A lubricant allows “free” longitudinal expansion of the rubber envelope during the contact. Combined theoretical, numerical, and experimental analyses are then proposed here to assess the effectiveness of lubrication conditions. Our results indicate an inversely proportional scaling between the elongation gradient and the friction coefficient in the pre-crack region. The observation confirms the validity of frictionless contact with a greased contact. Meanwhile, the steady-state crack growth assumption holds with the introduction of contact friction
Multi-objective optimization of cycloidal blade-controlled propeller: An experimental approach
No full textInternational audienceIn recent years, innovative naval propulsion systems have been investigated thanks to the growing development of unmanned underwater vehicles. Cycloidal propellers are promising alternative concepts to the usual screw propellers. As bio-inspired technology, these systems use mechanical energy from unsteady hydrodynamic forces generated by blades oscillation like natural marine animal swimmers. As an academic platform, the French Naval Academy Research Institute developed a large-scale experimental cycloidal propeller with the aim of running various pitch motions to evaluate performances of cross-flow propellers. Blades’ pitching is here performed by servo-motors in order to control each blade independently. While common cycloidal propellers use mechanical blade actuators which restrain the blade motion possibilities, this blade-controlled platform allows new investigations of interesting research area in marine propulsion. The platform is widely instrumented with load and torque sensors to measure instantaneous hydrodynamic forces during the rotation of the blades. Experiments, performed in a current flume tank, first reveal that for classical sinusoidal pitch laws, performances are depending on the operating point: the higher the advance parameter, the lower the sinusoidal amplitude must be for a better efficiency. These results confirm the requirement of an adaptable pitch control for cycloidal propeller to improve their performances regarding the operation mode. To go further, an experimental optimization, based on surrogate models (Efficient Global Optimization), is undertaken to surpass the performance of the propeller with parameterized pitch laws. This method authorizes a wide range of possible motion taking account of the platform speed limits. Multi-objective optimization is performed for total thrust and efficiency maximizing for two operating points. Results on the Pareto fronts show that a trade-off is necessary between thrust and efficiency concerning. However, optimized pitching laws reveal high hydrodynamic performances, with gains respectively from 10% to 20% on the hydrodynamic efficiency and the thrust in comparison with classic sinusoidal laws. This confirms the benefit of full electrical blade-controlled propeller and promises interesting further investigations on the experimental optimization
Impact of Sampling Strategies on the Monitoring of Climate Regime Shifts with a Learning Data Assimilation Method
No full textInternational audienceIn oceanography, the acquisition and processing of observations are crucial to improve the understanding of complex oceanic processes. Considering an idealized model of the North Atlantic ocean circulation, we propose to implement a variational data assimilation method optimized by deep learning to reconstruct abrupt changes in ocean circulation, representing Dansgaard-Oeschger climate events. We show that this assimilation method leads to improved reconstruction performances, particularly at low sampling frequencies. Focusing on this difficult latter case, four sampling strategies are studied more specifically. Our experiments highlight that clusters of three consecutive observations regularly sampled leads to a better monitoring of the ocean circulation regime shifts. These results pave the way for further research in optimal ocean observation.</div
Marine Renewable-Driven Green Hydrogen Production toward a Sustainable Solution and a Low-Carbon Future in Morocco
No full textInternational audienceOceanic energy sources, notably offshore wind and wave power, present a significant opportunity to generate green hydrogen through water electrolysis. This approach allows for offshore hydrogen production, which can be efficiently transported through existing pipelines and stored in various forms, offering a versatile solution to tackle the intermittency of renewable energy sources and potentially revolutionize the entire electrical grid infrastructure. This research focusses on assessing the technical and economic feasibility of this method in six strategic coastal regions in Morocco: Laayoune, Agadir, Essaouira, Eljadida, Casablanca and Larache. Our proposed system integrates offshore wind turbines, oscillating water column wave energy converters, and PEM electrolyzers, to meet energy demands while aligning with global sustainability objectives. Significant electricity production estimates are observed across these regions, ranging from 14 MW to 20 MW. Additionally, encouraging annual estimates of hydrogen production, varying between 20 and 40 tonnes for specific locations, showcase the potential of this approach. The system's performance demonstrates promising efficiency rates, ranging from 13% to 18%, while maintaining competitive production costs. These findings underscore the ability of oceanic energy-driven green hydrogen to diversify Morocco's energy portfolio, bolster water resilience, and foster sustainable development.Ultimately, this research lays the groundwork for comprehensive energy policies and substantial infrastructure investments, positioning Morocco on a trajectory towards a decarbonized future powered by innovative and clean technologies.</p
Apprentissage profond et extraction de descripteurs pour l’analyse la maladie veineuse thromboembolique et du syndrome de Gougerot-Sjögren à partir d’échographies
No full textDeep venous thrombosis (DVT) is linked to the obstruction of a vein by a blood clot. The objective of this thesis is to evaluate the possibility of predicting a pulmonary embolism (PE) which results from the migration of the clot towards a pulmonary artery, based on ultrasound images. It should be emphasized that there is no medical expertise for the detection PE from these images. We proposed two methods, the first is based on the extraction of texture descriptors, the second relies on deep learning models. We developed a learning scheme for deep neural networks based on a joint training on a classification and segmentation task, and then a specialization of the network on the classification task. Alternatively, we built a model combining images and clinical data. Beyond the techniques used, significant work has been carried out to sort the database studied and select images. We obtained conclusive accuracy on the detection of PE. In parallel, we used the same methods to classify the Gougerot-Sjögren syndrome on ultrasound imaging of salivary glands and demonstrated a great adaptability of the model to very different databases.La thrombose veineuse profonde (TVP), est liée à l’obstruction d’une veine par un caillot. L’objectif de cette thèse est de détecter une complication de TVP en embolie pulmonaire (EP), résultant de la migration d’un caillot vers une artère pulmonaire, à partir d’échographies. Actuellement, il n’existepas d’expertise médicale pour la détection d’EP à partir de ces images. Les images utilisées ont été sauvegardées durant des examens d’échographie évaluant la présence de TVP. Nous évaluons deux approches, la première, basée sur l’extraction de descripteurs de texture, la seconde, basée sur des modèles d’aprentissage profond. Nous avons développé un schéma d’apprentissage pour les réseaux de neurones profonds basé sur un apprentissage conjoint sur une tâche de classification et de segmentation, puis une spécialisation sur la classification seule. Alternativement, nous avons construit un modèle combinant des images et des données cliniques. Au delà des techniques utilisées, des travaux importants ont été effectués pour trier la base de données et sélectionner les images. Nous avons obtenu une exactitude intéressante sur la détection d’EP. En parallèle nous avons utilisé ces méthodes pour classer le Syndrome de Gougerot sur des échographies de glandes salivaires et démontré une grande adaptabilité du modèle obtenu à des bases de données très différentes