Dissertationen zum Thema „Réseaux neuronaux hybrides“
Geben Sie eine Quelle nach APA, MLA, Chicago, Harvard und anderen Zitierweisen an
Machen Sie sich mit Top-42 Dissertationen für die Forschung zum Thema "Réseaux neuronaux hybrides" bekannt.
Neben jedem Werk im Literaturverzeichnis ist die Option "Zur Bibliographie hinzufügen" verfügbar. Nutzen Sie sie, wird Ihre bibliographische Angabe des gewählten Werkes nach der nötigen Zitierweise (APA, MLA, Harvard, Chicago, Vancouver usw.) automatisch gestaltet.
Sie können auch den vollen Text der wissenschaftlichen Publikation im PDF-Format herunterladen und eine Online-Annotation der Arbeit lesen, wenn die relevanten Parameter in den Metadaten verfügbar sind.
Sehen Sie die Dissertationen für verschiedene Spezialgebieten durch und erstellen Sie Ihre Bibliographie auf korrekte Weise.
Reyes, Salgado Gerardo. „Connaissances de haut niveau dans les systèmes hybrides neuro-symboliques“. Grenoble INPG, 2001. http://www.theses.fr/2001INPG0047.
Der volle Inhalt der QuelleAlché-Buc, Florence d'. „Modèles neuronaux et algorithmes constructifs pour l'apprentissage de règles de décision“. Paris 11, 1993. http://www.theses.fr/1993PA112468.
Der volle Inhalt der QuelleSiboni, Didier. „La gestion de service sur les réseaux hétérogènes interconnectes : utilisation des techniques d'intelligence artificielle et architectures hybrides“. Versailles-St Quentin en Yvelines, 1997. http://www.theses.fr/1997VERS0005.
Der volle Inhalt der QuelleRybnik, Mariusz. „Contribution to the modelling and the exploitation of hybrid multiple neural networks systems : application to intelligent processing of information“. Paris 12, 2004. https://athena.u-pec.fr/primo-explore/search?query=any,exact,990003948290204611&vid=upec.
Der volle Inhalt der QuelleFor a great number of actually encountered problems (complex processes modelization, pattern recognition, medical diagnosis support, fault detection) data is presented in form of database. The data is next transformed and processed. This work is concentrated on the development of semi-automatic data processing structures. Proposed approach is based on iterative decomposition of an initial problem. The main idea is to decompose initia!ly complex problems in order to obtain simplification simultaneously on structural level and processing level. Thus, the principal idea of present work is con nected to task decomposition techniques called "Divide to Conquer". A key point of our approach is the integration of Complexity Estimation techniques
Khoyratee, Farad. „Conception d’une plateforme modulable de réseau de neurones biomimétiques pour l’étude des maladies neurodégénératives“. Thesis, Bordeaux, 2019. http://www.theses.fr/2019BORD0351.
Der volle Inhalt der QuelleNeuroscience has been the subject of many studies and has seen new fields of research emerge where technology and biology can be used to find solutions to understand and cure neurological diseases. These illness affect millions of people around the world. The World Health Organization (WHO) predicts a 3 fold increase in the number of patients in the next 30 years.Advances in neuroscience have led to the development of models describing the physiology of neurons and also methods of hardware implementation of these models. Among these methods, neuroprosthesis are devices for restoring certain neuronal functions through communication with the nervous system.This thesis work show that the realization of the biomimetic system was carried out thanks to digital components such as Field Programmable Gate Array (FPGA) which allows to benefit from the flexibility and speed of prototyping of these technologies. The real-time platform of biologically realistic neural networks developed is configurable. It becomes a neuro-computational tool allowing the realization of bio-hybrid experiments for the study of the behavior of the nervous system and more particularly of the neurodegenerative diseases.This work was placed in a larger context. The FPGA digital operator library developed for the platform has been reused for the study of dynamics similar to neural networks such as biochemical network simulation or combinatorial optimization problem solving
Marie-Françoise, Jean-Noël. „Contribution à la commande neuronale et à la gestion d'énergie d'un système hybride batterie-supercondensateurs : application aux transports terrestres“. Besançon, 2004. http://www.theses.fr/2004BESA2030.
Der volle Inhalt der QuelleThis thesis report is a contribution to the command and energy management of a hybrid system battery-ultracapacitors. More precisely, we will present a theoretical study and a realization of a power supply connecting a pack ultracapacitors to a battery, for hybrid vehicles applications. As in all the hybrid systems, the association between several sources needs static converters. They have to adapt the voltage and the current levels of the sources to the load (42V DC bus). This voltage level seems to be recommended by the largest car manufacturers. It is obvious, that the performances of such a device depend mainly on the command method used and also on the strategy for energy management. With this intention, we used two command methods which are for the first one, the classical regulation using PID corrector and for the second one, the techniques based on the artificial neural networks (ANN). The aim is to carry out a comparison between the two approaches
Al, Hajj Mohamad Rami. „Reconnaissance hors ligne de mots manuscrits cursifs par l'utilisation de systèmes hybrides et de techniques d'apprentissage automatique“. Paris, ENST, 2007. http://www.theses.fr/2007ENST0020.
Der volle Inhalt der QuelleThe automatic offline recognition of handwritten words improves human-machine interaction. It is already used in many business office applications dealing with the automatic processing of documents such as automatic post sorting, and the verification and recognition of bank check amounts. The off line recognition of cursive handwritten words remains an open problem due to difficulties such as :handwriting normalization, word segmentation into compound components and the modeling of these components. The main objective of this thesis, is to propose, design, and implement a system for the automatic offline recognition of Arabic handwritten words. The proposed approach is analytical without explicit segmentation of words into compound characters, and it is based on the stochastic HMM approach (Hidden Markov models). The method is composed of two stages : a recognition stage based on different features, and a combination stage of three HMM-based classifiers. Each individual HMM classifier uses a sliding window with a specific inclination. Different combining strategies are tested, among them the Sum rule, the Majority Vote rule and the Borda Count rule. The best combination strategy consists of using a neural network-based combining classifier. The combination of these classifiers can better cope with the writing inclination, the erroneous positions of diacritical marks and points, and the overlapping of consecutive characters in handwritten words. The reference system based on the proposed method has shown best performance at the competition organized at ICDAR 2005, where a set of state-of art systems were compared and tested on the IFN/ENIT benchmark database
Rynkiewicz, Joseph. „Modèles hybrides intégrant des réseaux de neurones artificiels à des modèles de chaînes de Markov cachées : application à la prédiction de séries temporelles“. Paris 1, 2000. http://www.theses.fr/2000PA010077.
Der volle Inhalt der QuelleThépaut, André. „Contribution à l'étude des machines hybrides : application à la reconnaissance des chiffres manuscrits“. Montpellier 2, 1995. http://www.theses.fr/1995MON20096.
Der volle Inhalt der QuelleCamargo-Pardo, Mauricio. „Estimation paramétrique des coûts des produits finis dans la filière textile-habillement“. Valenciennes, 2004. http://ged.univ-valenciennes.fr/nuxeo/site/esupversions/fe756208-c237-4caf-a060-91109809c4ba.
Der volle Inhalt der QuelleIn supply chains with high degree of product diversity and renewal, there is very difficult to establish economical laws at the design stage, in order to accurately forecast product cost. Nevertheless at this early stage, product cost is defined by 70 to 80% but also, only scarce product information is available, related mainly to product aesthetical features or functionalities. In order to minimise the risk of product reject, it is important for designers to have a cost estimation tool, flexible and easy to adapt. We use the parametric approach in order to develop Cost Estimation Relationships (CER's). First, we define a general methodology and main concepts in order to develop CER's as regression and softcomputing techniques. In particular we developed a Simplified Hybrid Neuro-Fuzzy model, allowing better variables interpretation, mainly for complex systems. Also, we proposed a tool in order to develop a CER by using the described modelling techniques. The candidates CER's could be compared in terms of accuracy, robustness and relevance. This tool allows a maximum of information in order to choose the best CER. This approach has been tested on a particular case concerning the development of specific CER for a textile printing industry
Louis, Thomas. „Conventionnel ou bio-inspiré ? Stratégies d'optimisation de l'efficacité énergétique des réseaux de neurones pour environnements à ressources limitées“. Electronic Thesis or Diss., Université Côte d'Azur, 2025. http://www.theses.fr/2025COAZ4001.
Der volle Inhalt der QuelleIntegrating artificial intelligence (AI) algorithms directly into satellites presents numerous challenges. These embedded systems, which are heavily limited in energy consumption and memory footprint, must also withstand interference. This systematically requires the use of system-on-chip (SoC) solutions to combine two so-called “heterogeneous” systems: a versatile microcontroller and an energy-efficient computing accelerator (such as an FPGA or ASIC). To address the challenges related to deploying such architectures, this thesis focuses on optimizing and deploying neural networks on heterogeneous embedded architectures, aiming to balance energy consumption and AI performance.In Chapter 2 of this thesis, an in-depth study of recent compression techniques for feedforward neural networks (FNN) like MLPs or CNNs was conducted. These techniques, which reduce the computational complexity and memory footprint of these models, are essential for deployment in resource-constrained environments. Spiking neural networks (SNN) were also explored. These bio-inspired networks can indeed offer greater energy efficiency compared to FNNs.In Chapter 3, we adapted and developed innovative quantization methods to reduce the number of bits used to represent the values in a spiking network. This allowed us to compare the quantization of SNNs and FNNs, to understand and assess their respective trade-offs in terms of losses and gains. Reducing the activity of an SNN (e.g., the number of spikes generated during inference) directly improves the energy efficiency of SNNs. To this end, in Chapter 4, we leveraged knowledge distillation and regularization techniques. These methods reduce the spiking activity of the network while preserving its accuracy, ensuring effective operation of SNNs on resource-limited hardware.In the final part of this thesis, we explored the hybridization of SNNs and FNNs. These hybrid networks (HNN) aim to further optimize energy efficiency while enhancing performance. We also proposed innovative multi-timestep networks, which process information with different latencies across layers within the same SNN. Experimental results show that this approach enables a reduction in overall energy consumption while maintaining performance across a range of tasks.This thesis serves as a foundation for deploying future neural network applications in space. To validate our methods, we provide a comparative analysis on various public datasets (CIFAR-10, CIFAR-100, MNIST, Google Speech Commands) as well as on a private dataset for cloud segmentation. Our approaches are evaluated based on metrics such as accuracy, energy consumption, or SNN activity. This research extends beyond aerospace applications. We have demonstrated the potential of quantized SNNs, hybrid neural networks, and multi-timestep networks for a variety of real-world scenarios where energy efficiency is critical. This work offers promising prospects for fields such as IoT devices, autonomous vehicles, and other systems requiring efficient AI deployment
Merasli, Alexandre. „Reconstruction d’images TEP par des méthodes d’optimisation hybrides utilisant un réseau de neurones non supervisé et de l'information anatomique“. Electronic Thesis or Diss., Nantes Université, 2024. http://www.theses.fr/2024NANU1003.
Der volle Inhalt der QuellePET is a functional imaging modality used in oncology to obtain a quantitative image of the distribution of a radiotracer injected into a patient. The raw PET data are characterized by a high level of noise and modest spatial resolution, compared to anatomical imaging modalities such as MRI or CT. In addition, standard methods for image reconstruction from the PET raw data introduce a positive bias in low activity regions, especially when dealing with low statistics acquisitions (highly noisy data). In this work, a new reconstruction algorithm, called DNA, has been developed. Using the ADMM algorithm, DNA combines the recently proposed Deep Image Prior (DIP) method to limit noise propagation and improve spatial resolution by using anatomical information, and a bias reduction method developed for low statistics PET imaging. However, the use of DIP and ADMM algorithms requires the tuning of many hyperparameters, which are often selected manually. A study has been carried out to tune some of them automatically, using methods that could benefit other algorithms. Finally, the use of anatomical information, especially with DIP, allows an improvement of the PET image quality, but can generate artifacts when information from one modality does not spatially match with the other. This is particularly the case when tumors have different anatomical and functional contours. Two methods have been developed to remove these artifacts while trying to preserve the useful information provided by the anatomical modality
Farcot, Etienne. „Etude d'une classe d'équations différentielles affines par morceaux modélisant des réseaux de régulation biologique“. Phd thesis, Grenoble INPG, 2005. http://tel.archives-ouvertes.fr/tel-00010463.
Der volle Inhalt der QuelleLe, Franc Yann. „Traitement de l'information sensorielle et nociceptive par le réseau de la corne dorsale de la moelle épinière“. Phd thesis, Université Victor Segalen - Bordeaux II, 2004. http://tel.archives-ouvertes.fr/tel-00548761.
Der volle Inhalt der QuelleDéchelle-Marquet, Marie. „Deep learning based physical-statistics modeling of ocean dynamics“. Electronic Thesis or Diss., Sorbonne université, 2023. https://theses.hal.science/tel-04166816.
Der volle Inhalt der QuelleThe modeling of dynamical phenomena in geophysics and climate is based on a deep understanding of the underlying physics, described in the form of PDEs, and on their resolution by numerical models. The ever-increasing number of observations of physical systems, the recent rise of deep learning and the huge computational power required by numerical solvers, which hinders the resolution of existing models, suggest that the future of physical models could be data-driven. But for this prognosis to come true, deep learning must tackle several challenges, such as the interpretability and physical consistency of deep models, still largely under-addressed by the deep learning community.In this thesis, we address both challenges: we study the prediction of sea surface temperature (SST) using hybrid models combining a data-driven and a physical model. Ensuring the physical plausibility of hybrid models necessitates well-posing their learning: otherwise, the high versatility of neural networks may lead the data-driven part to bypass the physical part.Our study is divided into two parts: a theoretical study on hybrid models, and a practical confrontation of our model on simulations of real data. First, we propose a new generic well- posed learning framework based on the optimization of an upper-bound of a prediction error. Second, we study real-like ocean observations of SST and velocity fields from the Gulf Stream current in the North Atlantic (from the NATL60 model). This application highlights the challenges raised by confronting physics aware learning to the complexity of real-world physics. It also raises issues such as model generalization, which we discuss as a possible perspective
Avrin, Guillaume. „Modélisation du contrôle moteur humain lors de tâches rythmiques hybrides et application à la commande de robots anthropomorphes“. Thesis, Université Paris-Saclay (ComUE), 2017. http://www.theses.fr/2017SACLS334.
Der volle Inhalt der QuelleThe identification of the neurbiological principles underlying human motor control is a very active reseach topic. Indeed, human movement has a level of robustness and dexterity still unmatched by robots. The objective is therefore to better understand the origin of this efficiency to replicate these performances in robotics. It has been shown that spinal rhythm generators, known as Central Pattern Generators (CPG), are responsible for the generation of rhythmic movements such as locomotion and respiration in vertebrates. These CPG constitute dynamic nonlinear systems modulated by sensory signals and descending signals from the cortex to adapt the behavior to the changing environment.The present study hypothesizes that visual information is also coupled to the CPG and that these couplings are responsible for the temporal and spatial synchronization observed during rhythmic visuomotor tasks. This assumption is confronted with experimental results from human participants performing ball bouncing, a well-known benchmark in neuroscience and robotics for its intrinsic dynamic properties. This task allows for the investigation of rhythmic movement generation by spinal networks, the temporal synchronization with the environment, the on-line correction of spatial errors and the interception of ballistic projectiles.This thesis proposes an innovative mathematical behavioral model based on a neuronal oscillator whose attractor, which defines the paddle trajectories, is modulated on-line by the visual perception of the ball trajectory. The relevance of the model is validated by comparison with experimental data and models previously proposed in the literature. The robustness of this control strategy is quantified by an asymptotic stability analysis. The bio-inspired controller presented in this thesis harmoniously combines a prospective control of the ball-paddle synchronization, an intermittent parametric control that scales the movement and a control emerging from the coupled system limit cycle. It efficiently reproduces the human modulation in motor action and performance during ball bouncing, without relying on movement planning or explicit internal representation of the environment. The results of this study lead to the realistic assumption that much part of the human behavior during ball bouncing is directly structured by sensory information and on-line error correction processes, in agreement with the behavioral dynamics theory. This control architecture holds promise for the control of humanoid robots as it is able to ensure stability and energy saving through control laws of reduced complexity and computational cost
Patiño, Diego. „Pilotage des cycles limites dans les systèmes dynamiques hybrides : application aux alimentations électriques statiques“. Electronic Thesis or Diss., Vandoeuvre-les-Nancy, INPL, 2009. http://www.theses.fr/2009INPL013N.
Der volle Inhalt der QuelleThis work deals with limit cycle control for one particular class of hybrid dynamical systems (HDS): The cyclic switched systems. The HDS were born because the traditional dynamical models were not able to describe complex behaviors and most of all, behaviors with discontinuities. From an application point of view, one important class of HDS depicts a cyclic behavior in steady state. The main characteristic of these systems is that the operation point cannot be maintained: It does not exist a control that maintains the system on a desired operation point. However, this point can be obtained in average by turning into its neighborhood. Thus, a cycle is produced by switching among the system modes. A switched control law must satisfy stability and dynamic performance. Moreover, criteria related to the waveform must be verified. Nowadays, few methods take into account the cyclic behavior of the system. In this research, some generic methods are studied. They show good performance for controlling the cyclic switched systems. The proposed algorithms can be implemented in real-time. The approaches are based on an affine non-linear model of the system whose control explicitly appears. Two control methods are considered: i) A predictive control, ii) An optimal control. Since the predictive control is a good choice for tracking, it will be able to maintain the system in a cycle. The optimal control yields solutions that can be applied to the transients. Some experiments with both control methods applied to the power converters are shown. These tests were carried out not only in our laboratory (CRAN), but also in other laboratories as part of the HYCON excellence network
Valentin, Nicolas. „Construction d'un capteur logiciel pour le contrôle automatique du procédé de coagulation en traitement d'eau potable“. Compiègne, 2000. http://www.theses.fr/2000COMP1314.
Der volle Inhalt der QuelleMellouk, Abdelhamid. „Un système neuro- prédictif pour la reconnaissance automatique de la parole continue“. Paris 11, 1994. http://www.theses.fr/1994PA112476.
Der volle Inhalt der QuelleDutto, Rémy. „Méthode à deux niveaux et préconditionnement géométrique en contrôle optimal. Application au problème de répartition de couple des véhicules hybrides électriques“. Electronic Thesis or Diss., Université de Toulouse (2023-....), 2024. http://www.theses.fr/2024TLSEP088.
Der volle Inhalt der QuelleMotivated by the torque split and gear shift industrial problem of hybrid electric vehicles, this work mainly proposes two new indirect optimal control problem methods. The first one is the Macro-Micro method, which is based on a bilevel decomposition of the optimal control problem and uses Bellman’s value functions at fixed times. These functions are known to be difficult to create. The main idea of this method is to approximate these functions by neural networks, which leads to a hierarchical resolution of a low dimensional optimization problem and a set of independent optimal control problems defined on smaller time intervals. The second one is a geometric preconditioning method, which allows a more efficient resolution of the optimal control problem. This method is based on a geometrical interpretation of the Pontryagin’s co-state and on the Mathieu transformation, and uses a linear diffeomorphism which transforms an ellipse into a circle. These two methods, presented separately, can be combined and lead together to a fast, robust and light resolution for the torque split and gear shift optimal control problem, closer to the embedded requirements
Yonaba, Harouna. „Modélisation hydrologique hybride : réseau de neurones - modèle conceptuel“. Thesis, Université Laval, 2009. http://www.theses.ulaval.ca/2009/26583/26583.pdf.
Der volle Inhalt der QuelleMalek, Maria. „Un modèle hybride de mémoire pour le raisonnement à partir de cas“. Université Joseph Fourier (Grenoble), 1996. http://www.theses.fr/1996GRE10186.
Der volle Inhalt der QuelleAmbroise, Matthieu. „Hybridation des réseaux de neurones : de la conception du réseau à l’interopérabilité des systèmes neuromorphiques“. Thesis, Bordeaux, 2015. http://www.theses.fr/2015BORD0394/document.
Der volle Inhalt der QuelleHYBRID experiments allow to connect a biological neural network with an artificial one,used in neuroscience research and therapeutic purposes. During these three yearsof PhD, this thesis focused on hybridization in a close-up view (bi-diretionnal direct communication between the artificial and the living) and in a broader view (interoperability ofneuromorphic systems). In the early 2000s, an analog neuromorphic system has been connected to a biological neural network. This work is firstly, about the design of a digital neural network, for hybrid experimentsin two multi-disciplinary projects underway in AS2N team of IMS presented in this document : HYRENE (ANR 2010-Blan-031601), aiming the development of a hybrid system for therestoration of motor activity in the case of a spinal cord lesion,BRAINBOW (European project FP7-ICT-2011-C), aiming the development of innovativeneuro-prostheses that can restore communication around cortical lesions. Having a configurable architecture, a digital neural network was designed for these twoprojects. For the first project, the artificial neural network emulates the activity of CPGs (Central Pattern Generator), causing the locomotion in the animal kingdom. This activity will trigger aseries of stimuli in the injured spinal cord textit in vitro and recreating locomotion previously lost. In the second project, the neural network topology will be determined by the analysis anddecryption of biological signals from groups of neurons grown on electrodes, as well as modeling and simulations performed by our partners. The neural network will be able to repair the injuredneural network. This work show the two different networks design approach and preliminary results obtained in the two projects.Secondly, this work hybridization to extend the interoperability of neuromorphic systems. Through a communication protocol using Ethernet, it is possible to interconnect electronic neuralnetworks, computer and biological. In the near future, it will increase the complexity and size of networks
Fock, Éric. „Modélisation hybride en physique du bâtiment : proposition d'outils d'optimisation des paramètres neuronaux : application à la modélisation des systèmes de traitement d'air“. La Réunion, 2004. http://elgebar.univ-reunion.fr/login?url=http://thesesenligne.univ.run/04_22_Fock.pdf.
Der volle Inhalt der QuelleThe electricity demand side management at Reunion island is still of growing interest because of the limited power supply. Thus, the french electricity utility leads action in order to reduce consumption. According, the tools for simulation of the thermal behavior of the building make it possible to consider an alternative architectural response to air-conditioning. Their implementation requires however powerful methods of modelling. The aim of this work is the contribution of neural networks for black box and hybrid mode in building physics. Behind an apparent simplicity, one can see that neural networks requires nevertheless a fine setup of the network architecture. Original tools for variable selection and node pruning are proposed. The tuned neuronal approach is applied to the modelling of split system and is validated within the framework of the international procedure IEA HVAC BESTEST
Saadia, Nadia. „Contribution à la commande hybride force-position des robots compliants selon une approche neuronale“. Paris 12, 1997. http://www.theses.fr/1997PA120080.
Der volle Inhalt der QuelleWacquant, Sylvie. „Contribution à l'étude d'un modèle de réseaux d'automates corticaux : principes & outils logiciels“. Rouen, 1993. http://www.theses.fr/1993ROUES063.
Der volle Inhalt der QuellePoisson, Émilie. „Architecture et apprentissage d'un système hybride neuro-markovien pour la reconnaissance de l'écriture manuscrite en-ligne“. Nantes, 2005. http://www.theses.fr/2005NANT2082.
Der volle Inhalt der QuelleThis thesis deals with the study, the conception, the development and the test of an online unconstrained handwriting word recognition system for an omni-writer application. The proposed system is based on a hybrid architecture including on the one hand, a neural convolutional network (TDNN and/or SDNN), and on the other hand Hidden Markov Models (HMM). The neural network has a global vision and works at the character level, while the HMM works on a more local description and allows the extension from the character level to the word level. The system was first dedicated for processing isolated characters (digits, lowercase letters, uppercase letters). This architecture has been optimized in terms of performances and size. The second part of this work concerns the extension to the word level. In this case, we have defined a global training scheme directly at the word level. It allows to insure the global convergence of the system. It relies on an objective function that combines two main criteria: one based on generative models (typically by maximum likelihood estimation) and the second one based on discriminant criteria (maximum mutual information). Several results are presented on MNIST, IRONOFF and UNIPEN databases. They show the influence of the main parameters of the system, either in terms of topologies, information sources, and training models (number of states, criteria weighting, duration)
Chane, Kuang Sang Laurent. „Stratégie de contrôle hybride d'un magnétron verrouillé par injection pour un Transport d'Énergie Sans Fil par onde hyperfréquence“. La Réunion, 2002. http://tel.archives-ouvertes.fr/tel-00464105/fr/.
Der volle Inhalt der QuelleWith the aim to put forward an alternative renewable and large-scale energy source to Mankind P. Glaser presented the project of Solar Power Satellite to the american spatial agency. This scheme consists in collecting directly in space the solar energy before being targeted on a terrestrial reception base by means of a focused microwave beam. This principle is founded on the concept of Wireless Power Transportation (WP1). To complete this project successfully, a preliminary "earthwork" strategy is adopted by the international researchers community, before upgrading to a spatial project. In terrestrial point-to-point WPT systems prototypes or proposals, one of the preferred microwave power projection system consists in a phased array antenna supplied by individual mid-power range microwave sources : magnetron. To be efficiently coupled to projecting systems and to allow electronic steering and beam-forming, magnetrons have to be synchronised to a reference frequency and controlled in phase and amplitude. For this purpose, this research wQrk presents a new approach of the control of the output parameters of an injection /ocked magnetron. Ln order to take into account the non linear behaviour of this microwave tube, an hybrid control strategy was designed to control the amplitude and frequency of a magnetron in fixed-load operations. This control algorithm involves a non linear artificial neural network modelling the plant inversion mapping, in combination with a classical linear PID feedback controller. Supervised and Generalized learning with experimental databases collected from a magnetron measurement bench developed in our laboratory was adopted to identify the neural controller. A dynamical - control architecture, which switches either on a non linear control loop or a classical linear PID feedback loop, allows to drive the frequency and amplitude of the magnetron, while its phase remains steady, all over the injection locking bandwith
Donà, Jérémie. „Statistical learning of physical dynamics“. Electronic Thesis or Diss., Sorbonne université, 2022. http://www.theses.fr/2022SORUS166.
Der volle Inhalt der QuelleThe modeling of natural processes relies on a physical description that prescribes the changes in the state of the studied system. The use of domain specific knowledge about the system allows the translation of physical principles into models, which are then validated by experimental data. With its successes in many domain like image classification, deep learning has become a powerful tool for the modeling of physical processes, thanks to the significant increase in the amount of data available from sensors. Statistical learning of physical processes by a sole data-driven approach suffers from several limitations such as interpretation difficulties, stability during training and reduced generalization capabilities. The objective of this work is to provide tools in order to perform data-driven learning of physical processes. In particular, we study spatio-temporal phenomena which dynamics obey a differential equation and focus on incorporating domain and physical knowledge in learning algorithms. This leads us to study hybrid physical-statistical systems for the modeling of physical processes. We will identify the problems related to the learning of hybrid dynamics and propose a framework including constraints adapted to deep networks to improve the interpretability and the performance of the learned algorithms. Conversely, dynamical systems have provided numerous tools to improve statistical models. However, neural networks remain qualified as "black boxes" because they are not interpretable. Thus, we will attempt to open the black box and propose more interpretable neural network architectures with increased generalization performances for the modeling of spatio-temporal systems
Yin, Yuan. „Physics-Aware Deep Learning and Dynamical Systems : Hybrid Modeling and Generalization“. Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS161.
Der volle Inhalt der QuelleDeep learning has made significant progress in various fields and has emerged as a promising tool for modeling physical dynamical phenomena that exhibit highly nonlinear relationships. However, existing approaches are limited in their ability to make physically sound predictions due to the lack of prior knowledge and to handle real-world scenarios where data comes from multiple dynamics or is irregularly distributed in time and space. This thesis aims to overcome these limitations in the following directions: improving neural network-based dynamics modeling by leveraging physical models through hybrid modeling; extending the generalization power of dynamics models by learning commonalities from data of different dynamics to extrapolate to unseen systems; and handling free-form data and continuously predicting phenomena in time and space through continuous modeling. We highlight the versatility of deep learning techniques, and the proposed directions show promise for improving their accuracy and generalization power, paving the way for future research in new applications
Orsier, Bruno. „Etude et application de systèmes hybrides neurosymboliques“. Phd thesis, Université Joseph Fourier (Grenoble), 1995. http://tel.archives-ouvertes.fr/tel-00005057.
Der volle Inhalt der QuelleSaïghi, Sylvain. „Circuits et systèmes de modélisation analogique de réseaux de neurones biologiques : application au développement d'outils pour les neurosciences computationnelles“. Phd thesis, Université Sciences et Technologies - Bordeaux I, 2004. http://tel.archives-ouvertes.fr/tel-00326005.
Der volle Inhalt der QuelleFrad, M'hamed. „Etude et mise en oeuvre d'un système d'interaction adaptatif pour les applications de réalité virtuelle“. Thesis, Université Paris-Saclay (ComUE), 2016. http://www.theses.fr/2016SACLE053.
Der volle Inhalt der QuelleOver last decades, virtual reality has been widely used in many disciplines. It is able to plunge the user at the heart of an artificial environment created digitally through interaction and immersion paradigms. These paradigms are based on the use of very specific interfaces that help user to interact and performspecific tasks in the virtual environment. Nevertheless, many technical problems are often present and may penalize the quality of that interaction and may break user immersion in the virtual environment.The goal of this thesis is to build a comprehensive procedure to guide the user to calibrate a virtual reality interface and therefore attempt to overcome some technological shortcomings. The originality of the thesis is the use of an approach that combines two areas of research that will combine very rarely, that of data processing and the virtual reality. This approach will provide theoretical and technical framework for the design of a comprehensive calibration procedure to ensure continuous and precise interaction in the virtual environment.To overcome problems described above, the work was conducted on several fronts :data acquisition, processing and validation. The first step is by the use of a new protocol insofar as it is based on virtual reality techniques to collect calibration data. In second step, two calibration methods have been proposed to improve the absolute accuracy of the virtual reality interface. Both methods are universal approximators as well as their ability to estimate the outputs of the involved system from inputs even the model of the system being calibrated remains unknown. In the last step, two virtual reality applications prototypes were developed in order to assess the relevance of our approach
Haykal, Vanessa. „Modélisation des séries temporelles par apprentissage profond“. Thesis, Tours, 2019. http://www.theses.fr/2019TOUR4019.
Der volle Inhalt der QuelleTime series prediction is a problem that has been addressed for many years. In this thesis, we have been interested in methods resulting from deep learning. It is well known that if the relationships between the data are temporal, it is difficult to analyze and predict accurately due to non-linear trends and the existence of noise specifically in the financial and electrical series. From this context, we propose a new hybrid noise reduction architecture that models the recursive error series to improve predictions. The learning process fusessimultaneouslyaconvolutionalneuralnetwork(CNN)andarecurrentlongshort-term memory network (LSTM). This model is distinguished by its ability to capture globally a variety of hybrid properties, where it is able to extract local signal features, to learn long-term and non-linear dependencies, and to have a high noise resistance. The second contribution concerns the limitations of the global approaches because of the dynamic switching regimes in the signal. We present a local unsupervised modification with our previous architecture in order to adjust the results by adapting the Hidden Markov Model (HMM). Finally, we were also interested in multi-resolution techniques to improve the performance of the convolutional layers, notably by using the variational mode decomposition method (VMD)
Benatia, Mohamed Amin. „Optimisation multi-objectives d’une infrastructure réseau dédiée aux bâtiments intelligents“. Thesis, Rouen, INSA, 2016. http://www.theses.fr/2016ISAM0024/document.
Der volle Inhalt der QuelleIn this thesis, we studied the Wireless Sensor Network deployment for indoor environments with a focus on smart building application. The goal of our work was to develop a WSN deployment tool which is able to assist network designers in the deployment phase. We begin this thesis with network modeling of all the deployment parameters and requirement, such as : cost, coverage, connectivity and network lifetime. Thereafter, we implement five optimisation methods, including three multi-objective optimization agorithms, to resolve WSN deployment problem. Then, two realistics study cases were identified to test the performances of the aforementioned algorithms. The obtained results shows that these algorithms are very efficient for deploying a small scale network in small buildings. However, when the building surface becomes more important the algorithms tends to converge to local optimum while consuming high processing time. To resolve this problem, we develop and implement a new Hybrid multi-objectif optimization algorithm wich limits the number of direct evaluation. This algorithm is based on data-mining methods (Artificial Neural Networks and K-means) and tries to approximate the fitness value of each individual in each generation. At every generation of the algorithm, the population is divided to K clusters and we evaluate only the closest individual to cluster centroide. The fitness value of the rest of population is approximated using a trained ANN. A comparative study was made and the obtained results show that our method outperformes others in the two sudy cases (small and big buildings)
Ramachandra, Rao Sanjay Kamath. „Question Answering with Hybrid Data and Models“. Thesis, université Paris-Saclay, 2020. http://www.theses.fr/2020UPASS024.
Der volle Inhalt der QuelleQuestion Answering is a discipline which lies in between natural language processing and information retrieval domains. Emergence of deep learning approaches in several fields of research such as computer vision, natural language processing, speech recognition etc. has led to the rise of end-to-end models.In the context of GoASQ project, we investigate, compare and combine different approaches for answering questions formulated in natural language over textual data on open domain and biomedical domain data. The thesis work mainly focuses on 1) Building models for small scale and large scale datasets, and 2) Leveraging structured and semantic information into question answering models. Hybrid data in our research context is fusion of knowledge from free text, ontologies, entity information etc. applied towards free text question answering.The current state-of-the-art models for question answering use deep learning based models. In order to facilitate using them on small scale datasets on closed domain data, we propose to use domain adaptation. We model the BIOASQ biomedical question answering task dataset into two different QA task models and show how the Open Domain Question Answering task suits better than the Reading Comprehension task by comparing experimental results. We pre-train the Reading Comprehension model with different datasets to show the variability in performance when these models are adapted to biomedical domain. We find that using one particular dataset (SQUAD v2.0 dataset) for pre-training performs the best on single dataset pre-training and a combination of four Reading Comprehension datasets performed the best towards the biomedical domain adaptation. We perform some of the above experiments using large scale pre-trained language models like BERT which are fine-tuned to the question answering task. The performance varies based on the type of data used to pre-train BERT. For BERT pre-training on the language modelling task, we find the biomedical data trained BIOBERT to be the best choice for biomedical QA.Since deep learning models tend to function in an end-to-end fashion, semantic and structured information coming from expert annotated information sources are not explicitly used. We highlight the necessity for using Lexical and Expected Answer Types in open domain and biomedical domain question answering by performing several verification experiments. These types are used to highlight entities in two QA tasks which shows improvements while using entity embeddings based on the answer type annotations. We manually annotated an answer variant dataset for BIOASQ and show the importance of learning a QA model with answer variants present in the paragraphs.Our hypothesis is that the results obtained from deep learning models can further be improved using semantic features and collective features from different paragraphs for a question. We propose to use ranking models based on binary classification methods to better rank Top-1 prediction among Top-K predictions using these features, leading to an hybrid model that outperforms state-of-art-results on several datasets. We experiment with several overall Open Domain Question Answering models on QA sub-task datasets built for Reading Comprehension and Answer Sentence Selection tasks. We show the difference in performance when these are modelled as overall QA task and highlight the wide gap in building end-to-end models for overall question answering task
Chane, Kuang Sang Laurent. „Stratégie de contrôle hybride d'un magnétron verrouillé par injection pour un Transport d'Energie Sans Fil par onde hyperfréquence“. Phd thesis, Université de la Réunion, 2002. http://tel.archives-ouvertes.fr/tel-00464105.
Der volle Inhalt der QuelleRojas, Castro Dalia Marcela. „The RHIZOME architecture : a hybrid neurobehavioral control architecture for autonomous vision-based indoor robot navigation“. Thesis, La Rochelle, 2017. http://www.theses.fr/2017LAROS001/document.
Der volle Inhalt der QuelleThe work described in this dissertation is a contribution to the problem of autonomous indoor vision-based mobile robot navigation, which is still a vast ongoing research topic. It addresses it by trying to conciliate all differences found among the state-of-the-art control architecture paradigms and navigation strategies. Hence, the author proposes the RHIZOME architecture (Robotic Hybrid Indoor-Zone Operational ModulE) : a unique robotic control architecture capable of creating a synergy of different approaches by merging them into a neural system. The interactions of the robot with its environment and the multiple neural connections allow the whole system to adapt to navigation conditions. The RHIZOME architecture preserves all the advantages of behavior-based architectures such as rapid responses to unforeseen problems in dynamic environments while combining it with the a priori knowledge of the world used indeliberative architectures. However, this knowledge is used to only corroborate the dynamic visual perception information and embedded knowledge, instead of directly controlling the actions of the robot as most hybrid architectures do. The information is represented by a sequence of artificial navigation signs leading to the final destination that are expected to be found in the navigation path. Such sequence is provided to the robot either by means of a program command or by enabling it to extract itself the sequence from a floor plan. This latter implies the execution of a floor plan analysis process. Consequently, in order to take the right decision during navigation, the robot processes both set of information, compares them in real time and reacts accordingly. When navigation signs are not present in the navigation environment as expected, the RHIZOME architecture builds new reference places from landmark constellations, which are extracted from these places and learns them. Thus, during navigation, the robot can use this new information to achieve its final destination by overcoming unforeseen situations.The overall architecture has been implemented on the NAO humanoid robot. Real-time experimental results during indoor navigation under both, deterministic and stochastic scenarios show the feasibility and robustness of the proposed unified approach
Grassia, Filippo Giovanni. „Silicon neural networks : implementation of cortical cells to improve the artificial-biological hybrid technique“. Thesis, Bordeaux 1, 2013. http://www.theses.fr/2013BOR14748/document.
Der volle Inhalt der QuelleThis work has been supported by the European FACETS-ITN project. Within the frameworkof this project, we contribute to the simulation of cortical cell types (employingexperimental electrophysiological data of these cells as references), using a specific VLSIneural circuit to simulate, at the single cell level, the models studied as references in theFACETS project. The real-time intrinsic properties of the neuromorphic circuits, whichprecisely compute neuron conductance-based models, will allow a systematic and detailedexploration of the models, while the physical and analog aspect of the simulations, as opposedthe software simulation aspect, will provide inputs for the development of the neuralhardware at the network level. The second goal of this thesis is to contribute to the designof a mixed hardware-software platform (PAX), specifically designed to simulate spikingneural networks. The tasks performed during this thesis project included: 1) the methodsused to obtain the appropriate parameter sets of the cortical neuron models that can beimplemented in our analog neuromimetic chip (the parameter extraction steps was validatedusing a bifurcation analysis that shows that the simplified HH model implementedin our silicon neuron shares the dynamics of the HH model); 2) the fully customizablefitting method, in voltage-clamp mode, to tune our neuromimetic integrated circuits usinga metaheuristic algorithm; 3) the contribution to the development of the PAX systemin terms of software tools and a VHDL driver interface for neuron configuration in theplatform. Finally, it also addresses the issue of synaptic tuning for future SNN simulation
Li, Haifeng. „Traitement de la variabilité et développement de systèmes robustes pour la reconnaissance de l'écriture manuscrite en-ligne“. Paris 6, 2002. http://www.theses.fr/2002PA066230.
Der volle Inhalt der QuellePham, Hoang Anh. „Coordination de systèmes sous-marins autonomes basée sur une méthodologie intégrée dans un environnement Open-source“. Electronic Thesis or Diss., Toulon, 2021. http://www.theses.fr/2021TOUL0020.
Der volle Inhalt der QuelleThis thesis studies the coordination of autonomous underwater robots in the context of coastal seabed exploration or facility inspections. Investigating an integrated methodology, we have created a framework to design and simulate low-cost underwater robot controls with different model assumptions of increasing complexity (linear, non-linear, and finally non-linear with uncertainties). By using this framework, we have studied algorithms to solve the problem of formation control, collision avoidance between robots and obstacle avoidance of a group of underwater robots. More precisely, we first consider underwater robot models as linear systems of simple integrator type, from which we can build a formation controller using consensus and avoidance algorithms. We then extend these algorithms for the nonlinear dynamic model of a Bluerov robot in an iterative design process. Then we have integrated a Radial Basis Function neural network, already proven in convergence and stability, with the algebraic controller to estimate and compensate for uncertainties in the robot model. Finally, we have presented simulation results and real basin tests to validate the proposed concepts. This work also aims to convert a remotely operated ROV into an autonomous ROV-AUV hybrid
Wu, Zhao Xin. „Un modèle computationnel d'intelligence culturelle ouvert et extensible“. Thèse, 2013. http://www.archipel.uqam.ca/5475/1/D2431.pdf.
Der volle Inhalt der Quelle