Teses / dissertações sobre o tema "Apprentissage automatique dynamique"
Crie uma referência precisa em APA, MLA, Chicago, Harvard, e outros estilos
Veja os 50 melhores trabalhos (teses / dissertações) para estudos sobre o assunto "Apprentissage automatique dynamique".
Ao lado de cada fonte na lista de referências, há um botão "Adicionar à bibliografia". Clique e geraremos automaticamente a citação bibliográfica do trabalho escolhido no estilo de citação de que você precisa: APA, MLA, Harvard, Chicago, Vancouver, etc.
Você também pode baixar o texto completo da publicação científica em formato .pdf e ler o resumo do trabalho online se estiver presente nos metadados.
Veja as teses / dissertações das mais diversas áreas científicas e compile uma bibliografia correta.
Quoy, Mathias. "Apprentissage dans les réseaux neuromimétiques à dynamique chaotique". Toulouse, ENSAE, 1994. http://www.theses.fr/1994ESAE0009.
Texto completo da fonteCalvelo, Aros Daniel. "Apprentissage de modèles e la dynamique pour l'aide à la décision en monitorage clinique". Lille 1, 1999. https://pepite-depot.univ-lille.fr/LIBRE/Th_Num/1999/50376-1999-351.pdf.
Texto completo da fonteGelly, Sylvain. "Une contribution à l'apprentissage par renforcement : application au Computer Go". Paris 11, 2007. http://www.theses.fr/2007PA112227.
Texto completo da fonteReinforcement Learning (RL) is at the interface of control theory, supervised and unsupervised learning, optimization and cognitive sciences. While RL addresses many objectives with major economic impact, it raises deep theoretical and practical difficulties. This thesis brings some contributions to RL, mainly on three axis. The first axis corresponds to environment modeling, i. E. Learning the transition function between two time steps. Factored approaches give an efficiently framework for the learning and use of this model. The Bayesian Networks are a tool to represent such a model, and this work brings new learning criterion, either in parametric learning (conditional probabilities) and non parametric (structure). The second axis is a study in continuous space and action RL, thanks to the dynamic programming algorithm. This analysis tackles three fundamental steps: optimization (action choice from the value function), supervised learning (regression) of the value function and choice of the learning examples (active learning). The third axis tackles the applicative domain of the game of Go, as a high dimensional discrete control problem, one of the greatest challenge in Machine Learning. The presented algorithms with their improvements made the resulting program, MoGo, win numerous international competitions, becoming for example the first go program playing at an amateur dan level on 9x9
Soula, Hédi. "Dynamique et plasticité dans les réseaux de neurones à impulsions : étude du couplage temporel réseau / agent / environnement". Lyon, INSA, 2005. http://theses.insa-lyon.fr/publication/2005ISAL0056/these.pdf.
Texto completo da fonteAn «artificial life » approach is conducted in order to assess the neural basis of behaviours. Behaviour is the consequence of a good concordance between the controller, the agent’s sensori-motors capabilities and the environment. Within a dynamical system paradigm, behaviours are viewed as attractors in the perception/action space – derived from the composition of the internal and external dynamics. Since internal dynamics is originated by the neural dynamics, learning behaviours therefore consists on coupling external and internal dynamics by modifying network’s free parameters. We begin by introducing a detailed study of the dynamics of large networks of spiking neurons. In spontaneous mode (i. E. Without any input), these networks have a non trivial functioning. According to the parameters of the weight distribution and provided independence hypotheses, we are able to describe completely the spiking activity. Among other results, a bifurcation is predicted according to a coupling factor (the variance of the distribution). We also show the influence of this parameter on the chaotic dynamics of the network. To learn behaviours, we use a biologically plausible learning paradigm – the Spike-Timing Dependent Plasticity (STDP) that allows us to couple neural and external dynamics. Applying shrewdly this learning law enables the network to remain “at the edge of chaos” which corresponds to an interesting state of activity for learning. In order to validate our approach, we use these networks to control an agent whose task is to avoid obstacles using only the visual flow coming from its linear camera. We detail the results of the learning process for both simulated and real robotics platform
Soula, Hédi Favrel Joel Beslon Guillaume. "Dynamique et plasticité dans les réseaux de neurones à impulsions étude du couplage temporel réseau / agent / environnement /". Villeurbanne : Doc'INSA, 2005. http://docinsa.insa-lyon.fr/these/pont.php?id=soula.
Texto completo da fonteLiu, Zongyi. "Self-Adaptive Bandwidth Control for Balanced QoS and Energy Aware Optimization in Wireless Sensor Network". Thesis, Toulouse, INSA, 2017. http://www.theses.fr/2017ISAT0034/document.
Texto completo da fonteIn the Wireless Multimedia Sensor Networks (WMSNs) field, highly saturated flow increases the probability of collision and congestion in data transmission which dramatically degrade the performance of Quality of Service (QoS). Multi-channels deployment technique is often applied to parallel transmission for QoS guarantee. However, how to make trade-off between QoS requirement and energy efficiency is a challenges to energy-constrained WMSNs. Theoretical analysis of MAC layer and PHY layer structure based on IEEE 802.15.4 standard, aim to study on the cross-layer analytical model in order to provide stronger understanding on the relationship between sensor network parameters and performance, pave the way for new enhancements in succedent multi-channel optimization research. Find effective performance indicator and design efficient performance collection or estimation approach based on the corresponding metrics, which could be used as the parameter input of multi-channel assignment mechanism. Comprehensive dynamically control system is designed for multi-channel assignment task based on light weight and high efficient computation intelligence techniques. We present a fuzzy-based dynamic bandwidth multi-channel assignment mechanism (MCDB_FLS). Cross-layer proactive available bandwidth is estimated as parameters for multi-channel deployment admission control. Reinforcement learning-based approach is proposed for more wisely decision-making in multi- channel allocation mission. Furthermore, fuzzy logic-based bandwidth threshold model provides dynamic optimization on system admission control. Simulations show the MCDB_FLS performs better than benchmark on the metrics of QoS and energy efficiency, achieves the trade-off between energy efficiency and QoS improvement. Finally, we introduce the integration of incremental machine learning approach into multi-channel assignment mechanism with Deep Q Network reinforcement learning method (DQMC). Besides, fully action weight initialization is implemented based on multi-class supervised learning classifier with stacking ensemble approach. DQMC improve the ability of self-adaptive and smart control to learn pattern from different environment of multi-tasks WMSNs
Munos, Rémi. "Apprentissage par renforcement, étude du cas continu". Paris, EHESS, 1997. http://www.theses.fr/1997EHESA021.
Texto completo da fonteNasri, Ridha. "Paramétrage Dynamique et Optimisation Automatique des Réseaux Mobiles 3G et 3G+". Phd thesis, Université Pierre et Marie Curie - Paris VI, 2009. http://tel.archives-ouvertes.fr/tel-00494190.
Texto completo da fonteAmadou, Boubacar Habiboulaye. "Classification Dynamique de données non-stationnaires :Apprentissage et Suivi de Classes évolutives". Phd thesis, Université des Sciences et Technologie de Lille - Lille I, 2006. http://tel.archives-ouvertes.fr/tel-00106968.
Texto completo da fonteAlami, Réda. "Bandits à Mémoire pour la prise de décision en environnement dynamique. Application à l'optimisation des réseaux de télécommunications". Electronic Thesis or Diss., université Paris-Saclay, 2021. http://www.theses.fr/2021UPASG063.
Texto completo da fonteIn this PhD thesis, we study the non-stationary multi-armed bandit problem where the non-stationarity behavior of the environment is characterized by several abrupt changes called "change-points". We propose Memory Bandits: a combination between an algorithm for the stochastic multi-armed bandit and the Bayesian Online Change-Point Detector (BOCPD). The analysis of the latter has always been an open problem in the statistical and sequential learning theory community. For this reason, we derive a variant of the Bayesian Online Change-point detector which is easier to mathematically analyze in term of false alarm rateand detection delay (which are the most common criteria for online change-point detection). Then, we introduce the decentralized exploration problem in the multi-armed bandit paradigm where a set of players collaborate to identify the best arm by asynchronously interacting with the same stochastic environment. We propose a first generic solution called decentralized elimination: which uses any best arm identification algorithm as a subroutine with the guar-antee that the algorithm ensures privacy, with a low communication cost. Finally, we perform an evaluation of the multi-armed bandit strategies in two different context of telecommunication networks. First, in LoRaWAN (Long Range Wide Area Network) context, we propose to use multi-armed bandit algorithms instead of the default algorithm ADR (Adaptive Data Rate) in order to minimize the energy consumption and the packet losses of end-devices. Then, in a IEEE 802.15.4-TSCH context, we perform an evaluation of 9 multi-armed bandit algorithms in order to select the ones that choose high-performance channels, using data collected through the FIT IoT-LAB platform. The performance evaluation suggests that our proposal can significantly improve the packet delivery ratio compared to the default TSCH operation, thereby increasing the reliability and the energy efficiency of the transmissions
Lurette, Christophe. "Développement d'une technique neuronale auto-adaptative pour la classification dynamique de données évolutives : application à la supervision d'une presse hydraulique". Lille 1, 2003. https://ori-nuxeo.univ-lille1.fr/nuxeo/site/esupversions/aed48e38-323f-425b-b6ff-c8e75ff5d4b6.
Texto completo da fonteLurette, Christophe Lecœuche Stéphane Vasseur Christian. "Développement d'une technique neuronale auto-adaptative pour la classification dynamique de données évolutives application à la supervision d'une presse hydraulique /". [S.l.] : [s.n.], 2003. http://www.univ-lille1.fr/bustl-grisemine/pdf/extheses/50376-2003-77-78.pdf.
Texto completo da fonteBeroule, Dominique. "Un modèle de mémoire adaptative, dynamique et associative pour le traitement automatique de la parole". Paris 11, 1985. http://www.theses.fr/1985PA112317.
Texto completo da fonteSalaün, Camille. "Apprentissage De Modèles Pour La Commande De La Mobilité Interne En Robotique". Phd thesis, Université Pierre et Marie Curie - Paris VI, 2010. http://tel.archives-ouvertes.fr/tel-00545534.
Texto completo da fonteGauriau, Olivier. "Fouille de règles numériques pour la prédiction de la dynamique des maladies des plantes". Electronic Thesis or Diss., Université de Rennes (2023-....), 2024. https://ged.univ-rennes1.fr/nuxeo/site/esupversions/23cd92a0-e9ff-4c23-82f2-6c8af1104dde.
Texto completo da fonteThis thesis focuses on the prediction of plant disease dynamics and the information that can be derived from the resulting models. To find a compromise between model performance and complexity, pattern-based models of intermediate complexity were used. The aim was to obtain sufficiently efficient models of this type, based on meteorological (such as rainfall and sunshine) and agronomic data. These models were compared with models commonly used in crop protection from the point of view of the tradeoff between model complexity and performance. Given the hybrid nature of pattern-based models, we sought to compare their structures and the information they provide with other models. This enabled us to confirm that pattern-based models come close to more complex models (RF, Gradient-Boosting...) while remaining simpler to understand. This allows us to assume that the explanations provided by these models are relevant. Finally, these models were used in the development of a visualization tool: this tool was developed in collaboration with agronomic experts from technical institutes to obtain a result adapted to their needs. This made it possible to isolate important principles for them, such as the notion of contrast in the information provided. The tool visualizes the agronomic and meteorological factors with the greatest impact on a defined set of plots
Sochacki, Stéphane. "De la combinaison/compétition de classifieurs vers la sélection dynamique d'opérateurs de traitement d'image". Poitiers, 2011. http://nuxeo.edel.univ-poitiers.fr/nuxeo/site/esupversions/7a942260-2262-43e4-b4a4-8873726dde8b.
Texto completo da fonteTechnology evolutions bring multimedia and huge heterogeneous databases (texts, images, sounds, video etc. ) development. Nevertheless, this evolution amplified the need of automatic processing solutions. Whereas many years of research resulted in more and more powerful operators development, their interoperability and sometimes their tuning remain supported by the expert. This limit harms the specialized multi‐media and consumer systems development. This research work is divided in three main parts. The automatic processing chain management, from preprocessing to decision, leads a reverse looped system to be build. Each stage needs the conception of information relative to the quality or the accuracy of the done action, and information indicating the relevance of this action according to the desired goal as well. This first part is based on the Theory of Evidence. An intermediate validation phase will be achieved on classes of attributes computing and classification operators. Patrimonial documents, and by extension multimedia databases, intrinsic variability introduce the choice trouble of a unique operators sequence. In order to dynamically decide the best chain for the input document, allowing to reach the desired goal, the system would have to put in competition the different operators from a same class. This second part exploits accuracy/relevance previously proposed. An intermediate validation about graphical primitives description attributes and hierarchical classifiers management will be used. As classification is based on a set of data, the attributes, the processing chain will have to manipulate operators like texture or shape analysis. But for their integration, we need to know the accuracy by usefulness and information preservation. This will be the third main step, via a shape descriptors attributes study. The progress of this thesis work will help to develop this different points within a complete processing chain. All those development will be integrated in an application dedicated to the analysis and valorization of patrimonial documents. Intermediate validation elements will be developed to evaluate the quality and the stability of all proposed solutions. This thesis work comes next to D. Arrivault's between the XLIM‐SIC laboratory and the RC‐Soft company and comes within works which results are exploited in the Interactive Documentary Environments to which the two entities participated
Novello, Paul. "Combining supervised deep learning and scientific computing : some contributions and application to computational fluid dynamics". Thesis, Institut polytechnique de Paris, 2022. http://www.theses.fr/2022IPPAX005.
Texto completo da fonteRecent innovations in mathematics, computer science, and engineering have enabled more and more sophisticated numerical simulations. However, some simulations remain computationally unaffordable, even for the most powerful supercomputers. Lately, machine learning has proven its ability to improve the state-of-the-art in many fields, notoriously computer vision, language understanding, or robotics. This thesis settles in the high-stakes emerging field of Scientific Machine Learning which studies the application of machine learning to scientific computing. More specifically, we consider the use of deep learning to accelerate numerical simulations.We focus on approximating some components of Partial Differential Equation (PDE) based simulation software by a neural network. This idea boils down to constructing a data set, selecting and training a neural network, and embedding it into the original code, resulting in a hybrid numerical simulation. Although this approach may seem trivial at first glance, the context of numerical simulations comes with several challenges. Since we aim at accelerating codes, the first challenge is to find a trade-off between neural networks’ accuracy and execution time. The second challenge stems from the data-driven process of the training, and more specifically, its lack of mathematical guarantees. Hence, we have to ensure that the hybrid simulation software still yields reliable predictions. To tackle these challenges, we thoroughly study each step of the deep learning methodology while considering the aforementioned constraints. By doing so, we emphasize interplays between numerical simulations and machine learning that can benefit each of these fields.We identify the main steps of the deep learning methodology as the construction of the training data set, the choice of the hyperparameters of the neural network, and its training. For the first step, we leverage the ability to sample training data with the original software to characterize a more efficient training distribution based on the local variation of the function to approximate. We generalize this approach to general machine learning problems by deriving a data weighting methodology called Variance Based Sample Weighting. For the second step, we introduce the use of sensitivity analysis, an approach widely used in scientific computing, to tackle neural network hyperparameter optimization. This approach is based on qualitatively assessing the effect of hyperparameters on the performances of a neural network using Hilbert-Schmidt Independence Criterion. We adapt it to the hyperparameter optimization context and build an interpretable methodology that yields competitive and cost-effective networks. For the third step, we formally define an analogy between the stochastic resolution of PDEs and the optimization process at play when training a neural network. This analogy leads to a PDE-based framework for training neural networks that opens up many possibilities for improving existing optimization algorithms. Finally, we apply these contributions to a computational fluid dynamics simulation coupled with a multi-species chemical equilibrium code. We demonstrate that we can achieve a time factor acceleration of 21 with controlled to no degradation from the initial prediction
Hadjerci, Oussama. "Détection automatique du nerf dans les images échographiques". Thesis, Orléans, 2017. http://www.theses.fr/2017ORLE2006/document.
Texto completo da fonteRegional anesthesia presents an interesting alternative or complementary act to general anesthesia in many surgical procedures. It reduces pain scores, improves postoperative mobility and facilitates earlier hospital discharge. Ultrasound-Guided Regional Anesthesia (UGRA) has been gaining importance in the last few years, offering numerous advantages over alternative methods of nerve localization (neurostimulation or paraesthesia). However, nerve detection is one of the most difficult tasks that anesthetists can encounter in the UGRA procedure. The context of the present work is to provide practitioners with a method to facilitate and secure the practice of UGRA. However, automatic detection and segmentation in ultrasound images is still a challenging problem in many medical applications. This work addresses two main issues. The first one, we propose an algorithm for nerve detection and segmentation in ultrasound images, this method is composed of a pre-processing, texture analysis and machine learning steps. In this part of work, we explore two new approaches ; one to characterize the nerve and the second for selecting the minimum redundant and maximum relevant features. The second one, we studied the nerve detection in consecutive ultrasound frames. We have demonstrated that the development of an algorithm based on the temporal coherence of the position, the shape and the confidence measure of the classification, allows to generate a robust segmentation. In this work, we also propose a new model of shape based on a set of intervals landmarks able to adapt to the nerve shape under a morphological variations
Ramdane, Saïd. "Identification automatique de types de formulaires par des méthodes stochastiques markoviennes". Le Havre, 2002. http://www.theses.fr/2002LEHA0018.
Texto completo da fonteIdentification of forms is a significant operation of an automatic system of reading. No distinctive sign is supposed to mark the form. The treatment starts with the extraction of the rectangular blocks of texts or rectangles including the drawings or the images. Since the forms include handwritten fields, the position, dimensions of the rectangular blocks present are variable. The phenomena of merging and fragmentation resulting from the segmentation induce an additional variability in the number of the rectangles. This double variability of the rectangles is naturally random. A first statistical method carries out the recognition by the calculation of a distance, which generalizes the Mahalanobis distance. Learning requires taking care of the phenomenon of merging/fragmentation. This statistical model appears to be actually a Markovian stochastic model of order 0. A second stochastic method rests on the construction of planar hidden Markov models (PHMM: Pseudo-2D Hidden Markov Model). We describe in particular a new unsupervised training of the states number by a dynamic aggregation method. The recognition is based on the estimation of the conditional probability calculated by an extension of a doubly imbricated Viterbi algorithm. For the two methods, we sought to make automatic all the phases of the training and the recognition. The experimental results confirm the validity of the two methods
Ngo, Ha Nhi. "Apprentissage continu et prédiction coopérative basés sur les systèmes de multi-agents adaptatifs appliqués à la prévision de la dynamique du trafic". Electronic Thesis or Diss., Université de Toulouse (2023-....), 2024. http://www.theses.fr/2024TLSES043.
Texto completo da fonteLe développement rapide des technologies matérielles, logicielles et de communication des systèmes de transport ont apporté des opportunités prometteuses et aussi des défis importants pour la société humaine. Parallèlement à l'amélioration de la qualité des transports, l'augmentation du nombre de véhicules a entraîné de fréquents embouteillages, en particulier dans les grandes villes aux heures de pointe. Les embouteillages ont de nombreuses conséquences sur le coût économique, l'environnement, la santé mentale des conducteurs et la sécurité routière. Il est donc important de prévoir la dynamique du trafic et d'anticiper l'apparition des embouteillages, afin de prévenir et d'atténuer les situations de trafic perturbées, ainsi que les collisions dangereuses à la fin de la queue d'un embouteillage. De nos jours, les technologies innovatives des systèmes de transport intelligents ont apporté des ensembles de données diverses et à grande échelle sur le trafic qui sont continuellement collectées et transférées entre les dispositifs sous forme de flux de données en temps réel. Par conséquent, de nombreux services de systèmes de transport intelligents ont été développés basé sur l'analyse de données massives, y compris la prévision du trafic. Cependant, le trafic contient de nombreux facteurs variés et imprévisibles qui rendent la modélisation, l'analyse et l'apprentissage de l'évolution historique du trafic difficiles. Le système que nous proposons vise donc à remplir les cinq composantes suivantes d'un système de prévision du trafic : textbf{analyse temporelle, analyse spatiale, interprétabilité, analyse de flux et adaptabilité à plusieurs échelles de données} pour capturer les patterns historiques de trafic à partir des flux de données, fournir une explication explicite de la causalité entrée-sortie et permettre différentes applications avec divers scénarios. Pour atteindre les objectifs mentionnés, nous proposons un modèle d'agent basé sur le clustering dynamique et la théorie des systèmes multi-agents adaptatifs afin de fournir des mécanismes d'apprentissage continu et de prédiction coopérative. Le modèle d'agent proposé comprend deux processus interdépendants fonctionnant en parallèle : textbf{apprentissage local continu} et textbf{prédiction coopérative}. Le processus d'apprentissage vise à détecter, au niveau de l'agent, différents états représentatifs à partir des flux de données reçus. Basé sur le clustering dynamique, ce processus permet la mise à jour continue de la base de données d'apprentissage en s'adaptant aux nouvelles données. Simultanément, le processus de prédiction exploite la base de données apprise, dans le but d'estimer les futurs états potentiels pouvant être observés. Ce processus prend en compte l'analyse de la dépendance spatiale en intégrant la coopération entre les agents et leur voisinage. Les interactions entre les agents sont conçues sur la base de la théorie AMAS avec un ensemble de mécanismes d'auto-adaptation comprenant textbf{l'auto-organisation}, textbf{l'autocorrection} et textbf{l'auto-évolution}, permettant au système d'éviter les perturbations, de gérer la qualité de la prédiction et de prendre en compte les nouvelles informations apprises dans le calcul de la prédiction. Les expériences menées dans le contexte de la prévision de la dynamique du trafic évaluent le système sur des ensembles de données générées et réelles à différentes échelles et dans différents scénarios. Les résultats obtenus ont montré la meilleure performance de notre proposition par rapport aux méthodes existantes lorsque les données de trafic expriment de fortes variations. En outre, les mêmes conclusions retirées de différents cas d'étude renforcent la capacité du système à s'adapter à des applications multi-échelles
Reis, Ana Flávia dos. "New Baseband Architectures Using Machine Learning and Deep Learning in the Presence of Nonlinearities and Dynamic Environment". Electronic Thesis or Diss., Sorbonne université, 2024. http://www.theses.fr/2024SORUS023.
Texto completo da fonteThe forthcoming sixth generation (6G) of wireless communication systems is expected to enable a wide range of new applications in vehicular communication, which is accompanied by a diverse set of challenges and opportunities resulting from the demands of this cutting-edge technology. In particular, these challenges arise from dynamic channel conditions, including time-varying channels and nonlinearities induced by high-power amplifiers. In this complex context, wireless channel estimation emerges as an essential element in establishing reliable communication. Furthermore, the potential of machine learning and deep learning in the design of receiver architectures adapted to vehicular communication networks is evident, given their capabilities to harness vast datasets, model complex channel conditions, and optimize receiver performance. Throughout the course of this research, we leveraged these potential tools to advance the state-of-the-art in receiver design for vehicular communication networks. In this manner, we delved into the characteristics of wireless channel estimation and the mitigation of nonlinear distortions, recognizing these as significant factors in the communication system performance. To this end, we propose new methods and flexible receivers, based on hybrid approaches that combine mathematical models and machine learning techniques, taking advantage of the unique characteristics of the vehicular channel to favor accurate estimation. Our analysis covers both conventional wireless communications waveform and a promising 6G waveform, targeting the comprehensiveness of our approach. The results of the proposed approaches are promising, characterized by substantial enhancements in performance and noteworthy reductions in system complexity. These findings hold the potential for real-world applications, marking a step toward the future in the domain of vehicular communication networks
Chalabi, Asma. "Processus d'analyse dynamique pour l'imagerie de cellules vivantes permettant la détection des réponses cellulaires aux anticancéreux, par traitement de l'image et du signal et apprentissage automatique profond". Electronic Thesis or Diss., Université Côte d'Azur, 2024. http://www.theses.fr/2024COAZ6004.
Texto completo da fonteCell division and cell death are the main indicators to evaluate cancer drug action, and only their accurate measures can reveal the actual potency and efficacy of a compound. The detection of cell division and cell death events in live-cell assays has the potential to produce robust metrics of drug pharmacodynamics and return a more comprehensive understanding of tumor cells responses to cancer therapeutic combinations. Knowing precisely when a cell death or a cell division occurs in a live-cell experiment allows to study the relative contribution of different drug effects -such as cytotoxic or cytostatic effects, on a cell population. Yet, classical methods require dyes to measure cell viability as an end-point assay with whole population counts, where the proliferation rates can only be estimated when both viable and dead cells are labeled simultaneously.Live-cell imaging is a promising cell-based assay to determine drug efficacies, with the main limitation being the accuracy and depth of the analyses to detect and predict automatically cellular response phenotypes (cell death and division, which share some morphological features).This thesis introduces a method integrating deep learning using neural networks, and image and signal processing to perform dynamic image analyses of single-cell events in time-lapse microscopy experiments of drug pharmacological profiling. This method works by automatically tracking the cells, extracting radiometric and morphologic cell features, and analyzing the temporal evolution of these features for each cell so as to detect cellular events such as division and cell death, as well as acquiring signaling pathway dynamics.A case of study comprising the analyses of caspase-8 single-cell dynamics and other cell responses to cancer drugs is presented. The aim is to achieve automatically, at a large scale the necessary analyses to augment the phenotype prediction method available in the lab (Fateseq) and to apply it to various cancer cell lines of a human cancer cell line panel to improve our live-cell OMICS profiling approaches, and, in a longer term, to scale up pharmacological screening of new cancer drugs
Infantes, Guillaume. "Apprentissage de modèles de comportement pour le contrôle d'exécution et la planification robotique". Phd thesis, Université Paul Sabatier - Toulouse III, 2006. http://tel.archives-ouvertes.fr/tel-00129505.
Texto completo da fonteBayoudh, Meriam. "Apprentissage de connaissances structurelles à partir d’images satellitaires et de données exogènes pour la cartographie dynamique de l’environnement amazonien". Thesis, Antilles-Guyane, 2013. http://www.theses.fr/2013AGUY0671/document.
Texto completo da fonteClassical methods for satellite image analysis are inadequate for the current bulky data flow. Thus, automate the interpretation of such images becomes crucial for the analysis and management of phenomena changing in time and space, observable by satellite. Thus, this work aims at automating land cover cartography from satellite images, by expressive and easily interpretable mechanism, and by explicitly taking into account structural aspects of geographic information. It is part of the object-based image analysis framework, and assumes that it is possible to extract useful contextual knowledge from maps. Thus, a supervised parameterization methods of a segmentation algorithm is proposed. Secondly, a supervised classification of geographical objects is presented. It combines machine learning by inductive logic programming and the multi-class rule set intersection approach. These approaches are applied to the French Guiana coastline cartography. The results demonstrate the feasibility of the segmentation parameterization, but also its variability as a function of the reference map classes and of the input data. Yet, methodological developments allow to consider an operational implementation of such an approach. The results of the object supervised classification show that it is possible to induce expressive classification rules that convey consistent and structural information in a given application context and lead to reliable predictions, with overall accuracy and Kappa values equal to, respectively, 84,6% and 0,7. In conclusion, this work contributes to the automation of the dynamic cartography from remotely sensed images and proposes original and promising perpectives
Seroussi, Brigitte. "Alex : resolution de problemes par analogie basee sur un apprentissage de strategies par la construction dynamique d'une memoire indexee des exemples". Paris 7, 1988. http://www.theses.fr/1988PA077153.
Texto completo da fonteMabed, Mehdi. "Application de méthodes d'apprentissage profond à l'analyse et modélisation de la dynamique non-linéaire dans les fibres optiques". Electronic Thesis or Diss., Bourgogne Franche-Comté, 2024. http://www.theses.fr/2024UBFCD021.
Texto completo da fonteNonlinear pulse propagation in optical fibers exhibits a wide range of complex dynamics. Depending on the initial conditions and parametres of the fiber used, it is possible to observe a range of instability and solitonic dynamics leading to pronounced spectral broadening, and even supercontinuum generation. The complexity of these effects can make their analysis and modeling difficult, and the aim of this thesis has been to demonstrate the application of machine learning techniques to analyse different scenarios of nonlinear pulse propagation governed by the nonlinear Schrödinger equation. In particular, the prediction of the temporal properties of optical "rogue waves" based solely on spectral intensity analysis, without any phase information, was accomplished using a feedforward neural network. These predictions proved highly satisfactory in systems of pure modulation instability, as well as noise-like pulse laser operation in narrow- and wide-band spectral regimes. Finally, the feedforward neural network algorithm also proved remarkably accurate in reconstructing the dynamics of the nonlinear Schrödinger equation
Benazzouz, Yazid. "Découverte de contexte pour une adaptation automatique de services en intelligence ambiante". Phd thesis, Ecole Nationale Supérieure des Mines de Saint-Etienne, 2011. http://tel.archives-ouvertes.fr/tel-00733013.
Texto completo da fonteBernard, Simon. "Forêts aléatoires : de l’analyse des mécanismes de fonctionnement à la construction dynamique". Phd thesis, Rouen, 2009. http://www.theses.fr/2009ROUES011.
Texto completo da fonteThis research work is related to machine learning and more particularlydealswiththeparametrizationofRandomForests,whichareclassifierensemble methods that use decision trees as base classifiers. We focus on two important parameters of the forest induction : the number of features randomly selected at each node and the number of trees. We first show that the number of random features has to be chosen regarding to the feature space properties, and we propose hence a new algorithm called Forest-RK that exploits those properties. We then show that a static induction process implies that some of the trees of the forest make the ensemble generalisation error decrease, by deteriorating the strength/correlation compromise. We finaly propose an original random forest dynamic induction algorithm that favorably compares to static induction processes
Bézenac, Emmanuel de. "Modeling physical processes with deep learning : a dynamical systems approach". Electronic Thesis or Diss., Sorbonne université, 2021. http://www.theses.fr/2021SORUS203.
Texto completo da fonteDeep Learning has emerged as a predominant tool for AI, and has already abundant applications in fields where data is abundant and access to prior knowledge is difficult. This is not necessarily the case for natural sciences, and in particular, for physical processes. Indeed, these have been the object of study since centuries, a vast amount of knowledge has been acquired, and elaborate algorithms and methods have been developped. Thus, this thesis has two main objectives. The first considers the study of the role that deep learning has to play in this vast ecosystem of knowledge, theory and tools. We will attempt to answer this general question through a concrete problem: the one of modelling complex physical processes, leveraging deep learning methods in order to make up for lacking prior knowledge. The second objective is somewhat its converse: it focuses on how perspectives, insights and tools from the field of study of physical processes and dynamical systems can be applied in the context of deep learning, in order to gain a better understanding and develop novel algorithms
Pastor, Philippe. "Étude et application des méthodes d'apprentissage pour la navigation d'un robot en environnement inconnu". Toulouse, ENSAE, 1995. http://www.theses.fr/1995ESAE0013.
Texto completo da fonteMeghnoudj, Houssem. "Génération de caractéristiques à partir de séries temporelles physiologiques basée sur le contrôle optimal parcimonieux : application au diagnostic de maladies et de troubles humains". Electronic Thesis or Diss., Université Grenoble Alpes, 2024. http://www.theses.fr/2024GRALT003.
Texto completo da fonteIn this thesis, a novel methodology for features generation from physiological signals (EEG, ECG) has been proposed that is used for the diagnosis of a variety of brain and heart diseases. Based on sparse optimal control, the generation of Sparse Dynamical Features (SDFs) is inspired by the functioning of the brain. The method's fundamental concept revolves around sparsely decomposing the signal into dynamical modes that can be switched on and off at the appropriate time instants with the appropriate amplitudes. This decomposition provides a new point of view on the data which gives access to informative features that are faithful to the brain functioning. Nevertheless, the method remains generic and versatile as it can be applied to a wide range of signals. The methodology's performance was evaluated on three use cases using openly accessible real-world data: (1) Parkinson's Disease, (2) Schizophrenia, and (3) various cardiac diseases. For all three applications, the results are highly conclusive, achieving results that are comparable to the state-of-the-art methods while using only few features (one or two for brain applications) and a simple linear classifier supporting the significance and reliability of the findings. It's worth highlighting that special attention has been given to achieving significant and meaningful results with an underlying explainability
Pierrot, David. "Détection dynamique des intrusions dans les systèmes informatiques". Thesis, Lyon, 2018. http://www.theses.fr/2018LYSE2077.
Texto completo da fonteThe expansion and democratization of the digital world coupled with the effect of the Internet globalization, has allowed individuals, countries, states and companies to interconnect and interact at incidence levels never previously imagined. Cybercrime, in turn, is unfortunately one the negative aspects of this rapid global interconnection expansion. We often find malicious individuals and/or groups aiming to undermine the integrity of Information Systems for either financial gain or to serve a cause. The consequences of an intrusion can be problematic for the existence of a company or an organization. The impacts are synonymous with financial loss, brand image degradation and lack of seriousness. The detection of an intrusion is not an end in itself, the reduction of the delta detection-reaction has become a priority. The different existing solutions prove to be cumbersome to set up. Research has identified more efficient data mining methods, but integration into an information system remains difficult. Capturing and converting protected resource data does not allow detection within acceptable time frames. Our contribution helps to detect intrusions. Protect us against Firewall events which reduces the need for computing power while limiting the knowledge of the information system by intrusion detectors. We propose an approach taking into account the technical aspects by the use of a hybrid method of data mining but also the functional aspects. The addition of these two aspects is grouped into four phases. The first phase is to visualize and identify network activities. The second phase concerns the detection of abnormal activities using data mining methods on the source of the flow but also on the targeted assets. The third and fourth phases use the results of a risk analysis and a safety verification technique to prioritize the actions to be carried out. All these points give a general vision on the hygiene of the information system but also a direction on monitoring and corrections to be made.The approach developed to a prototype named D113. This prototype, tested on a platform of experimentation in two architectures of different size made it possible to validate our orientations and approaches. The results obtained are positive but perfectible. Prospects have been defined in this direction
Djebra, Yanis. "Accelerated Dynamic MR Imaging Using Linear And Non-Linear Machine Learning-Based Image Reconstruction Models". Electronic Thesis or Diss., Institut polytechnique de Paris, 2023. http://www.theses.fr/2023IPPAT011.
Texto completo da fonteDynamic Magnetic Resonance (MR) imaging is of high value in medical diagnosis thanks to its contrast versatility, high spatial resolution, high Signal-to-Noise Ratio (SNR), and allows for non-invasive multi-planar images of the body. It can be particularly useful for imaging the brain, heart, spine, and joints, as well as for detecting abnormalities. In addition, the increasing availability of Positron Emission Tomography (PET)/MR machines enables simultaneous acquisition of PET and MR data for better reconstruction and complementary information. However, a key challenge in dynamic MRI is reconstructing high-dimensional images from sparse k-space data sampled below the Nyquist sampling rate. Many methods have been proposed for accelerated imaging with sparse sampling, including parallel imaging and compressed sensing.The first objective of this thesis is to show the potential and usefulness of the linear subspace model for free-breathing MR imaging. Such a model can in principle capture regular respiratory and cardiac motion. However, when dealing with lengthy scans, irregular motion patterns can occur, such as erratic breathing or bulk motion caused by patient discomfort. A first question thus naturally arises: can such a model capture irregular types of motion and, if so, can it reconstruct images from a dynamic MR scan presenting bulk motion and irregular respiratory motion? We demonstrate in this thesis how the subspace model can efficiently reconstruct artifact-free images from highly undersampled k-space data with various motion patterns. A first application is presented where we reconstruct high-resolution, high frame-rate dynamic MR images from a PET/MR scanner and use them to correct motion in PET data, capturing complex motion patterns such as irregular respiratory patterns and bulk motion. A second application on cardiac T1 mapping is presented. Undersampled k-space data were acquired using a free-breathing, ECG-gated inversion recovery sequence, and dynamic 3D MR images of the whole heart were reconstructed leveraging the linear subspace model.The second objective of this thesis is to understand the limits of the linear subspace model and develop a novel dynamic MR reconstruction scheme that palliates these limitations. More specifically, the subspace model assumes that high-dimensional data reside in a low-dimensional linear subspace that captures the spatiotemporal correlations of dynamic MR images. This model relies on a linear dimensionality reduction model and does not account for intrinsic non-linear features of the signal, which may show its limits with higher undersampling rates. Manifold learning-based models have therefore been explored for image reconstruction in dynamic MRI and aim at learning the intrinsic structure of the input data that are embedded in a high-dimensional signal space by solving non-linear dimensionality reduction problems. We present in this thesis an alternative strategy for manifold learning-based MR image reconstruction. The proposed method learns the manifold structure via linear tangent space alignment (LTSA) and can be interpreted as a non-linear generalization of the subspace model. Validation on numerical simulation studies as well as in vivo 2D and 3D cardiac imaging experiments were performed, demonstrating improved performances compared to state-of-the-art techniques.The two first objectives present respectively linear and non-linear models yet both methods use conventional linear optimization techniques to solve the reconstruction problem. In contrast, using deep neural networks for optimization may procure non-linear and better representation power. Early results on deep learning-based approaches are presented in this thesis and state-of-the-art techniques are discussed. The last chapter then presents conclusions, discusses the author's contributions, and considers the potential research perspectives that have been opened up by the work presented in this thesis
Alwan, Azzam. "Vieillissement du système cardio-vasculaire – Etude de l’activité des peptides d’élastine". Thesis, Troyes, 2018. http://www.theses.fr/2018TROY0034.
Texto completo da fonteThis work aims to propose a statistical methodology to study the degradation products of arterial elastin. The proposed approach consists in analyzing simulation data of molecular dynamics of peptides resulting from the degradation of elastin proteins. Biological approaches indicate that some of these peptides can be considered as molecular signals and can influence the evolution of vascular pathologies. Moreover, experiments show that the biological properties of peptides are linked to their 3D structures. In this context, the objective of our work consists in analyzing the 3D structures of these peptides to identify the structures (conformations) related to their biological activities and next predict the activity of new peptides. It is therefore necessary to identify the "key" conformations for the activity of these peptides using a database of dynamic simulations of their molecular behaviour. Among the simulated peptides, some are known to have a biological effect while others are not. First, it is extremely important to identify the main conformations of each peptide from molecular simulations. A process combining several statistical methods is proposed for this purpose and demonstrates its effectiveness on the basis of existing data. Second, a peptide activity detector is proposed. It is able to predict the activity of new unlabelled peptides. The proposed detector is simple and can be applied to large databases
Leclerc, Sarah Marie-Solveig. "Automatisation de la segmentation sémantique de structures cardiaques en imagerie ultrasonore par apprentissage supervisé". Thesis, Lyon, 2019. http://www.theses.fr/2019LYSEI121.
Texto completo da fonteThe analysis of medical images plays a critical role in cardiology. Ultrasound imaging, as a real-time, low cost and bed side applicable modality, is nowadays the most commonly used image modality to monitor patient status and perform clinical cardiac diagnosis. However, the semantic segmentation (i.e the accurate delineation and identification) of heart structures is a difficult task due to the low quality of ultrasound images, characterized in particular by the lack of clear boundaries. To compensate for missing information, the best performing methods before this thesis relied on the integration of prior information on cardiac shape or motion, which in turns reduced the adaptability of the corresponding methods. Furthermore, such approaches require man- ual identifications of key points to be adapted to a given image, which makes the full process difficult to reproduce. In this thesis, we propose several original fully-automatic algorithms for the semantic segmentation of echocardiographic images based on supervised learning ap- proaches, where the resolution of the problem is automatically set up using data previously analyzed by trained cardiologists. From the design of a dedicated dataset and evaluation platform, we prove in this project the clinical applicability of fully-automatic supervised learning methods, in particular deep learning methods, as well as the possibility to improve the robustness by incorporating in the full process the prior automatic detection of regions of interest
Boutiba, Karim. "On enforcing Network Slicing in the new generation of Radio Access Networks". Electronic Thesis or Diss., Sorbonne université, 2024. https://accesdistant.sorbonne-universite.fr/login?url=https://theses-intra.sorbonne-universite.fr/2024SORUS003.pdf.
Texto completo da fonteThe emerging 5G networks and beyond promise to support novel use cases such as immersive holographic communication, Internet of Skills, and 4D Interactive mapping [usecases]. These use cases require stringent requirements in terms of Quality of Service (QoS), such as low latency, high Downlink (DL)/Uplink (UL) throughput and low energy consumption. The 3rd Generation Partnership Project (3GPP) specifications introduced many features in 5G New Radio (NR) to improve the physical efficiency of 5G to meet the stringent and heterogeneous requirements of beyond 5G services. Among the key 5G NR features, we can mention the numerology, BandWidth Part (BWP), dynamic Time Duplex Division (TDD) and Connected-mode Discontinuous Reception (C-DRX). However, the specifications do not provide how to configure the next Generation Node B (gNB)/User Equipment (UE) in order to optimize the usage of the 5G NR features. We enforce the 5G NR features by applying Machine Learning (ML), particularly Deep Reinforcement Learning (DRL), to fill this gap. Indeed, Artificial Intelligence (AI)/ML is playing a vital role in communications and networking [1] thanks to its ability to provide a self-configuring and self-optimizing network.In this thesis, different solutions are proposed to enable intelligent configuration of the Radio Access Network (RAN). We divided the solutions into three different parts. The first part concerns RAN slicing leveraging numerology and BWPs. In contrast, the second part tackles dynamic TDD, and the last part goes through different RAN optimizations to support Ultra-Reliable and Low-Latency Communication (URLLC) services.In the first part, we propose two contributions. First, we introduce NRflex, a RAN slicing framework aligned with Open RAN (O-RAN) architecture. NRflex dynamically assigns BWPs to the running slices and their associated User Equipment (UE) to fulfill the slices' required QoS. Then, we model the RAN slicing problem as a Mixed-Integer Linear Programming (MILP) problem. To our best knowledge, this is the first MILP modeling of the radio resource management featuring network slicing, taking into account (i) Mixed-numerology, (ii) both latency and throughput requirements (iii) multiple slices attach per UE (iv) Inter-Numerology Interference (INI). After showing that solving the problem takes an exponential time, we consider a new approach in a polynomial time, which is highly required when scheduling radio resources. The new approach consists of formalizing this problem using a DRL-based solver.In the second part of this thesis, we propose a DRL-based solution to enable dynamic TDD in a single 5G NR cell. The solution is implemented in OAI and tested using real UEs. Then, we extend the solution by leveraging Multi-Agent Deep Reinforcement Learning (MADRL) to support multiple cells, considering cross-link interference between cells.In the last part, we propose three solutions to optimize the RAN to support URLLC services. First, we propose a two-step ML-based solution to predict Radio Link Failure (RLF). We combine Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) to find the correlation between radio measurements and RLF. The RLF prediction model was trained with real data obtained from a 5G testbed. In the second contribution, we propose a DRL-based solution to reduce UL latency. Our solution dynamically allocates the future UL grant by learning from the dynamic traffic pattern. In the last contribution, we introduce a DRL-based solution to balance latency and energy consumption by jointly deriving the C-DRX parameters and the BWP configuration
Versini, Raphaëlle. "Structural basis of outer-mitochondrial membrane mitofusin-guided fusion". Electronic Thesis or Diss., Sorbonne université, 2023. https://accesdistant.sorbonne-universite.fr/login?url=https://theses-intra.sorbonne-universite.fr/2023SORUS653.pdf.
Texto completo da fonteThe Phd project is the structural study of mitofusins (Mfn1/2 in humans and Fzo1 in yeasts) using mainly modeling-based methods such as molecular dynamics or structure prediction methods based on artificial intelligence (mainly AlphaFold). This project is a part of an ANR (MITOFUSION) shared between different partners (Laboratoire de Biochimie Théorique: Antoine Taly, Marc Baaden, Laboratoire des Biomolécules: Patrick Fuchs, Laboratoire de Biologie Moléculaire et Cellulaire des Eucaryotes: Mickaël Cohen, Institut de Psychiatrie et Neurosciences de Paris: David Tareste) whose goal is to understand the structure-function relationships of the mitofusin. Mitochondria form a complex network inside the cells, undergoing continuous fusion and fission events. These processes shape mitochondrial dynamics and are essential for the maintenance, function, distribution and inheritance of mitochondria. The morphology of the latter therefore respond to the ever-changing physiological changes of the cell. The large GTPase involved in the tethering and fusion of the mitochondrial outer membranes (OM) are transmembrane proteins called mitofusins. The mitofusins Mfn1 and Mfn2 can be found in mammals. Fzo1 (Fuzzy Onion 1) is the unique mitofusin homologue in Saccharomyces cerevisiae. The mitochondrial inner membrane fusion and cristea organisation is mediated by human OPA1 (Optic Atrophy 1) and yeast Mgm1 (Mitochondrial Genome Maintenance 1). Mitochondrial fusion dysfonction is related to several neurodegenerative disorders, such as Parkinson, Alzheimer and Huntingtion diseases. As a matter of fact, research has shown that mutations in Mfn2 induce the development and progression of muscular dystrophies, such as Charcot-Marie-Tooth Type 2A, the most common form of axonal CMT disease. The exact mechanism by which the mitofusins contributes to mitochondria dysfunction as well as the exact molecular fusion mechanism is not fully understood yet. Overall, mitochondrial fusion plays an important role in CMT2A, it is thus of paramount importance to get a full understanding of the process at the molecular level. The structure of both Mfn1 and Mfn2 was partially solved, the transmembrane domain being excluded, and no solved structure are available for Fzo1. With our ANR partners, we decided to work on the yeast version of Mitofusin (named Fzo1) as it is a good model (of homology with human Mfn1 and Mfn2) as yeast are convenient hosts for testing how other protein partners are involved in the process (e.g. Ugo1). Fzo1 is embedded in the mitochondrial OM as it possesses two transmembrane domains, exposing N- and C- terminal portions towards the cytosol and a loop towards the intermembrane space. On the N-terminal side can be found two coiled-coil heptad repeats (HRs) domains, HRN (in yeast only) and HR1, flanking a GTPase domain. A third coiled-coil heptad repeats domain HR2 is on the C-terminal portion. Some models of Fzo1 were built based on the mitofusin related bacterial dynamin-like protein (BDLP). BDLP is involved in membrane remodelling and exists in two conformational states, a closed compact version which changes to an opened extended structure, upon GTP-binding, on which the built models were based. The goal of the PhD is to update the model of Fzo1 built in 2017, by working on the transmembrane domains using multiscale molecular dynamics, and then update the overall structure using artificial intelligence methods. An other project consisted in studying the amphipathic helix of HR1 domain of Mfn1 (MfnA-AH), test its membrane binding capabilities. Initially, we employed coarse-grained simulations, establishing a robust foundation for evaluating the predictive capacity of the MARTINI family of force fields. Using other simulations ran with the penetratin, we were able to provide a comparative analysis for the AH-membranes interactions in the MARTINI force-fields. The Mfn1-AH was then further characterized using all-atom simulations
Devergne, Timothée. "Machine learning methods for computational studies in origins of life". Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS376.
Texto completo da fontePrebiotic chemistry is the study of chemical reactions at the origins of life on earth. It is a very wide subject that requires the contribution of many scientific fields, including numerical physics. Indeed, highly accurate molecular dynamics simulations are performed to test the influence of different environnement on the synthesis of molecules: could this componant appear in interstellar medium? Is its formation impacted by the presence of mineral surfaces? They can also be used to identify intermediates that are too unstable to be observed experimentally and better understand the mechanism of formation. To do so, enhanced sampling (ES) methods such as metadynamics or umbrella sampling are used to explore and sample the chemical landscape. These methods can be used to study the synthesis of amino acids that are the building blocks of proteins. This was done by Magrino et al, with the study of the Strecker synthesis of glycine, the simplest amino acid, in water. This allowed to identify all the intermediates and characterize their stability. However, these simulations called ab initio which take into account the electronic degrees of freedom are computationnally expensive and only small systems of a few hundreds atoms can be studied. To solve this problem, machine learning (ML) methods have been put into place that allow to reduce the computational time for equilibrium system. Only few ML methods have been suggested to study reactive events because this requires an accurate model on the whole chemical space. In a first time, we use the existing data from the study of the prebiotic synthesis of glycine to devise a training methods for ML models for chemical reactions in solution. We start by training a set of models, called committee, with the same training set but different initial conditions. During a simulation we track the evolution of deviation of the prediction of forces, and we see that when the model is out of its training zone, this deviation drastically increases. This allows us to define a time during which the model behaves like an ab initio simulation. Thanks to this, we can target in the chemical space what new data to put in the training set to have a more accurate model. By using this method we obtained results close to ab initio accuracy. We then apply this method to a new prebiotic pathway to glycine in water that has never been studies before. However, the method previously developped requires a prior knowledge of the transition mechanism. In the second part of this thesis, we use ab initio transition path sampling trajectories which are trajectories starting from the transition state and relaxing into the equilibrium basins. They cover all the chemical space and are therefore suitable to train a ML model. By using such model we managed not only to recover the thermodynamics of the reaction but also the kinetics. We obtained results close to ab initio accuracy
Mokhtari, Aimed. "Diagnostic des systèmes hybrides : développement d'une méthode associant la détection par classification et la simulation dynamique". Phd thesis, INSA de Toulouse, 2007. http://tel.archives-ouvertes.fr/tel-00200034.
Texto completo da fonteVauchel, Nicolas. "Estimation des indices de Sobol à l'aide d'un métamodèle multi-éléments : application à la dynamique du vol". Electronic Thesis or Diss., Université de Lille (2022-....), 2023. http://www.theses.fr/2023ULILN008.
Texto completo da fonteThe thesis is addressing a concrete issue on aircrafts safety. The post-stall flight domain is a complex flight domain where flows around an airfoil may be highly unstable and massively stalled. In this domain, which can be reached on purpose or accidentally, usual controls are less efficient or completely inefficient, which can endanger the pilot and its passengers. The thesis is about the determination of the flight predictions in the post-stall flight domain, their dependences to the selected model structure and about the uncertainties of the experimental data the model relies on. The dynamic of the motion of the aircraft is governed by a dynamic system of ordinary non-linear differential equations. In these equations, the effects from the fluid on the aircraft are traduced by the global aerodynamic coefficients, the dimensionless forces and moments applied by the fluid on the aircraft. These coefficients depend on a high number of variables in a non-linear fashion. Among these variables are the geometry of the aircraft, its velocity and its rotation rates compared to earth, and characteristics of the surrounding flow. A representation model having a selected structure is determined for every aerodynamic coefficient, in order to represent these complex dependences. This model rely on experimental data obtained on a scale model, free flight data on a real aircraft being too expensive and too risky to get in the post-stall domain. Another way of obtaining data would be to use computational simulations. Nevertheless, the complex and unsteady flows around the 3D geometry of the aircraft makes the simulation too expensive with the current ressources, even if some recent studies begin to explore this direction of research. The selected models in the thesis are built on experimental data only. In the dynamic system, the global aerodynamic coefficients are evaluated by interpolation in these databases according to the selected model structure. The fact of selecting a simplified structure of the model makes it deficient. Moreover, as these models rely on experimental data, they are uncertain. The gaps and the uncertainties of the model have some impacts on the flight predictions. The initial objective of the thesis is therefore to study these impacts.During the thesis, new scientific objectives appeared, objectives going beyond the scope of Flight Dynamics. First, a new multi-element surrogate model for Uncertainty Quantification based on modern Machine learning methods is developed. Multi-element surrogate models were developed to address the loss of accuracy of Polynomial Chaos model in presence of discontinuities. Then, a formula linking the sensitivity Sobol indices to the coefficient of a multi-element surrogate model is derived. These results are used in the case of Flight Dynamics in order to address the issue raised in the initial objective of the thesis. The numerous bifurcations of the dynamic system can be traduced by discontinuities and/or irregularities in the evolution of the state variables compared to the uncertain parameters. The methods of Sensitivity Analysis and of Uncertainty Quantification developed in the thesis are therefore good candidates to analyse the system
Bessafa, Hichem. "Advanced Estimation Algorithms for Connected and Autonomous Vehicle Applications". Electronic Thesis or Diss., Université de Lorraine, 2024. http://www.theses.fr/2024LORR0075.
Texto completo da fonteThis thesis is dedicated to the development of advanced estimation algorithms specifically designed for autonomous vehicle applications. Initially, we provide a comprehensive overview of various vehicle controllers and advanced driving assistance systems, setting the stage for an in-depth discussion of vehicle dynamics and kinematics models. We then explore both classical (model-based) and machine learning-based (data-driven) observers, examining their literature and applications within vehicular and robotics contexts. Our research introduces several novel methodologies: first, a finite time interval estimation approach for discrete Linear Parameter Varying (LPV) systems, applied to the vehicle's lateral dynamics to estimate side slip despite uncertainties in cornering stiffness. Next, we propose a neuro-adaptive observer that combines neural networks with concurrent learning to estimate unknown forces in the vehicle's longitudinal model. Furthermore, we present a generalized high-gain observer, incorporating Linear Matrix Inequality (LMI) conditions and a threshold constraint on the high-gain parameter, designed to handle additional measurements and constraints. This observer ensures Input-to-State Stability (ISS) bounds on measurement noise and adapts to non-canonical systems via output transformation and augmented system design. Finally, we validate our methods through extensive simulations using the CARLA simulator and trajectory estimation with the KITTI dataset, demonstrating superior performance in terms of accuracy, convergence speed, and robustness in various vehicular scenarios. The outcomes illustrate significant improvements over traditional methods, highlighting the practical potential of our advanced estimation techniques in enhancing autonomous vehicle performance
Damay, Gabriel. "Dynamic Decision Trees and Community-based Graph Embeddings : towards Interpretable Machine Learning". Electronic Thesis or Diss., Institut polytechnique de Paris, 2024. http://www.theses.fr/2024IPPAT047.
Texto completo da fonteMachine Learning is the field of computer science that interests in building models and solutions from data without knowing exactly the set of instructions internal to these models and solutions. This field has achieved great results but is now under scrutiny for the inability to understand or audit its models among other concerns. Interpretable Machine Learning addresses these concerns by building models that are inherently interpretable. This thesis contributes to Interpretable Machine Learning in two ways.First, we study Decision Trees. This is a very popular group of Machine Learning methods for classification problems and it is interpretable by design. However, real world data is often dynamic, but few algorithms can maintain a decision tree when data can be both inserted and deleted from the training set. We propose a new algorithm called FuDyADT to solve this problem.Second, when data are represented as graphs, a very common machine learning technique called "embedding" consists in projecting them onto a vectorial space. This kind of method however is usually not interpretable. We propose a new embedding algorithm called Parfaite based on the factorization of the Personalized PageRank matrix. This algorithm is designed to provide interpretable results.We study both algorithms theoretically and experimentally. We show that FuDyADT is at least comparable to state-of-the-art algorithms in the usual setting, while also being able to handle unusual settings such as deletions of data and numerical features. Parfaite on the other hand produces embedding dimensions that align with the communities of the graph, making the embedding interpretable
Claeys, Emmanuelle. "Clusterisation incrémentale, multicritères de données hétérogènes pour la personnalisation d’expérience utilisateur". Thesis, Strasbourg, 2019. http://www.theses.fr/2019STRAD039.
Texto completo da fonteIn many activity sectors (health, online sales,...) designing from scratch an optimal solution for a defined problem (finding a protocol to increase the cure rate, designing a web page to promote the purchase of one or more products,...) is often very difficult or even impossible. In order to face this difficulty, designers (doctors, web designers, production engineers,...) often work incrementally by successive improvements of an existing solution. However, defining the most relevant changes remains a difficult problem. Therefore, a solution adopted more and more frequently is to compare constructively different alternatives (also called variations) in order to determine the best one by an A/B Test. The idea is to implement these alternatives and compare the results obtained, i.e. the respective rewards obtained by each variation. To identify the optimal variation in the shortest possible time, many test methods use an automated dynamic allocation strategy. Its allocate the tested subjects quickly and automatically to the most efficient variation, through a learning reinforcement algorithms (as one-armed bandit methods). These methods have shown their interest in practice but also limitations, including in particular a latency time (i.e. a delay between the arrival of a subject to be tested and its allocation) too long, a lack of explicitness of choices and the integration of an evolving context describing the subject's behaviour before being tested. The overall objective of this thesis is to propose a understable generic A/B test method allowing a dynamic real-time allocation which take into account the temporals static subjects’s characteristics
Maag, Maria Coralia Laura. "Apprentissage automatique de fonctions d'anonymisation pour les graphes et les graphes dynamiques". Thesis, Paris 6, 2015. http://www.theses.fr/2015PA066050/document.
Texto completo da fonteData privacy is a major problem that has to be considered before releasing datasets to the public or even to a partner company that would compute statistics or make a deep analysis of these data. Privacy is insured by performing data anonymization as required by legislation. In this context, many different anonymization techniques have been proposed in the literature. These techniques are difficult to use in a general context where attacks can be of different types, and where measures are not known to the anonymizer. Generic methods able to adapt to different situations become desirable. We are addressing the problem of privacy related to graph data which needs, for different reasons, to be publicly made available. This corresponds to the anonymized graph data publishing problem. We are placing from the perspective of an anonymizer not having access to the methods used to analyze the data. A generic methodology is proposed based on machine learning to obtain directly an anonymization function from a set of training data so as to optimize a tradeoff between privacy risk and utility loss. The method thus allows one to get a good anonymization procedure for any kind of attacks, and any characteristic in a given set. The methodology is instantiated for simple graphs and complex timestamped graphs. A tool has been developed implementing the method and has been experimented with success on real anonymized datasets coming from Twitter, Enron or Amazon. Results are compared with baseline and it is showed that the proposed method is generic and can automatically adapt itself to different anonymization contexts
Maag, Maria Coralia Laura. "Apprentissage automatique de fonctions d'anonymisation pour les graphes et les graphes dynamiques". Electronic Thesis or Diss., Paris 6, 2015. http://www.theses.fr/2015PA066050.
Texto completo da fonteData privacy is a major problem that has to be considered before releasing datasets to the public or even to a partner company that would compute statistics or make a deep analysis of these data. Privacy is insured by performing data anonymization as required by legislation. In this context, many different anonymization techniques have been proposed in the literature. These techniques are difficult to use in a general context where attacks can be of different types, and where measures are not known to the anonymizer. Generic methods able to adapt to different situations become desirable. We are addressing the problem of privacy related to graph data which needs, for different reasons, to be publicly made available. This corresponds to the anonymized graph data publishing problem. We are placing from the perspective of an anonymizer not having access to the methods used to analyze the data. A generic methodology is proposed based on machine learning to obtain directly an anonymization function from a set of training data so as to optimize a tradeoff between privacy risk and utility loss. The method thus allows one to get a good anonymization procedure for any kind of attacks, and any characteristic in a given set. The methodology is instantiated for simple graphs and complex timestamped graphs. A tool has been developed implementing the method and has been experimented with success on real anonymized datasets coming from Twitter, Enron or Amazon. Results are compared with baseline and it is showed that the proposed method is generic and can automatically adapt itself to different anonymization contexts
Yan, Yujin. "Μοbile data analysis : rοbust alignment and flexible clustering methοds". Electronic Thesis or Diss., Normandie, 2024. http://www.theses.fr/2024NORMIR14.
Texto completo da fonteThe widespread popularity of mobile devices in modern life has brought a revolution in communication, navigation, and daily activities. Mobile devices generate a vast amount of data, including a wide range of user behaviors from location tracking to communication patterns and sensor data. This multifaceted data not only captures the interactions between individuals and their devices, but also reveals social trends and preferences. Recognizing the importance of mobile data, our research is dedicated to exploring and mining the user daily behavior recorded by mobile devices. Initially, we focus on analyzing trajectory data, which is a component of mobile data that is typically represented as a spatio-temporal sequence. Using the semantic information embedded in trajectory points, we can gain insights into users' behaviors and activities. However, the integration of spatial and temporal dimensions increases the complexity of the analysis. We propose a semantic-based trajectory clustering (STC) architecture to analyze trajectory data, which includes data preprocessing, similarity exploration, and clustering methods. In addition, we introduce a dynamic programming-based similarity exploration (DPD) model to quantify the similarity between trajectories, thus enhancing our understanding of mobile data. A comprehensive experimental analysis is conducted on a real-world dataset to compare the DPD model with the other baseline methods. The comparison results show the adeptness of our DPD algorithm in effectively examining associations within trajectory data. To enhance experimental control, we propose a data generation model simulating daily life scenarios by generating random data based on real user data. Through quantitative comparisons between the proposed STC architecture and other approaches, our algorithm demonstrates good performance. Transitioning from trajectory data to multivariate mobile data, we are challenged to effectively utilize various sensor types to extract subtle insights into user behavior. By introducing one-dimensional multivariate sequence alignment (1D MSA) algorithm and two-dimensional multivariate sequence alignment (2D MSA) algorithm, we facilitate a comprehensive analysis of multivariate mobile data. While the 1D MSA algorithm prioritizes computational efficiency, the 2D MSA algorithm excels at extracting subtle similarities between sequences, providing a more detailed analysis. Meanwhile, we use some different clustering methods to analyze the similar subsequences obtained by the two algorithms and obtained similar or even identical clustering results. Moreover, the user states represented by each category in the clustering results are highly interpretable. This indicates that our algorithms can obtain stable and real-life consistent results. Furthermore, we compare the similar subsequences obtained by 2D MSA algorithm and baseline methods. The results show that our proposed 2D MSA algorithm has superior performance in capturing subtle similarity from the data. This robust performance makes the 2D MSA algorithm as a powerful tool for extracting meaningful subsequences in multivariate mobile data, contributing to enhanced data interpretation and practical applications
Veillon, Lise-Marie. "Apprentissage artificiel collectif ; aspects dynamiques et structurels". Thesis, Sorbonne Paris Cité, 2018. http://www.theses.fr/2018USPCD004/document.
Texto completo da fonteCollective learning in multi-agent systems considers how a community of autonomous agents sharing a learning purpose may benefit from exchanging information to learn efficiently as a community as well as individuals. The community forms a communication network where each agent may accesses observations, called learning examples. This thesis is based on a former protocol, SMILE (Sound-Multi-agent-Incremental-LEarning), which sets up parsimonious examples and hypotheses exchanges between agents. In a fully connected community, this protocol guarantees an agent’s hypothesis takes into account all the examples obtained by the community. Some sequential protocols add propagation to SMILE in order to extend this consistency guarantee to other connected networks. This thesis contribution to the artificial collective learning field is two fold.First, we investigate the influence of network structures on learning in networks when communication is limited to neighbourhood without further information propagation. Second, we present and analyze a new protocol, Waves, with SMILE’s guarantees and a more dynamic learning process thanks to its execution in parallel. The evaluation of this protocol in a simple turn-based setting gives the opportunity to improve it here in multiple ways. It is however meant to be used with online learning without any restriction on the acquisition rate of new examples, neither on speed nor number
Dromnelle, Rémi. "Architecture cognitive générique pour la coordination de stratégies d'apprentissage en robotique". Electronic Thesis or Diss., Sorbonne université, 2021. http://www.theses.fr/2021SORUS039.
Texto completo da fonteThe main objective of this thesis is to propose a new method for online adaptation of robotic learning, allowing robots to dynamically and autonomously adapt their behavior according to variations in their own performance. The developed method is sufficiently general and task-independent that a robot using it can perform different dynamic tasks of various nature without any algorithm or parameter adjustment by the programmer. The algorithms underlying this method consist of a meta-control system that allows the robot to call upon two decision-making experts following a different behavioral strategy. The model-based expert builds a model of the effects of long-term actions and uses this model to decide; this strategy is computationally expensive, but quickly converges to the solution. The model-free expert is inexpensive in terms of computational resources, but takes time to converge to the optimal solution. In this work, we have developed a new criterion for the coordination of these two experts allowing the robot to dynamically change its strategy over time. We show in this work that our behavior coordination method allows the robot to maintain an optimal performance in terms of performance and computation time. We also show that the method can cope with abrupt changes in the environment, changes in goals or changes in the behavior of the human partner in the case of interaction tasks
Dupuis, Romain. "Surrogate models coupled with machine learning to approximate complex physical phenomena involving aerodynamic and aerothermal simulations". Thesis, Toulouse, INPT, 2019. http://www.theses.fr/2019INPT0017/document.
Texto completo da fonteNumerical simulations provide a key element in aircraft design process, complementing physical tests and flight tests. They could take advantage of innovative methods, such as artificial intelligence technologies spreading in aviation. Simulating the full flight mission for various disciplines pose important problems due to significant computational cost coupled to varying operating conditions. Moreover, complex physical phenomena can occur. For instance, the aerodynamic field on the wing takes different shapes and can encounter shocks, while aerothermal simulations around nacelle and pylon are sensitive to the interaction between engine flows and external flows. Surrogate models can be used to substitute expensive high-fidelitysimulations by mathematical and statistical approximations in order to reduce overall computation cost and to provide a data-driven approach. In this thesis, we propose two developments: (i) machine learning-based surrogate models capable of approximating aerodynamic experiments and (ii) integrating more classical surrogate models into industrial aerothermal process. The first approach mitigates aerodynamic issues by separating solutions with very different shapes into several subsets using machine learning algorithms. Moreover, a resampling technique takes advantage of the subdomain decomposition by adding extra information in relevant regions. The second development focuses on pylon sizing by building surrogate models substitutingaerothermal simulations. The two approaches are applied to aircraft configurations in order to bridge the gap between academic methods and real-world applications. Significant improvements are highlighted in terms of accuracy and cost gains
Dzogang, Fabon. "Représentation et apprentissage à partir de textes pour des informations émotionnelles et pour des informations dynamiques". Paris 6, 2013. http://www.theses.fr/2013PA066253.
Texto completo da fonteAutomatic knowledge extraction from texts consists in mapping lowlevel information, as carried by the words and phrases extracted fromdocuments, to higher level information. The choice of datarepresentation for describing documents is, thus, essential and thedefinition of a learning algorithm is subject to theirspecifics. This thesis addresses these two issues in the context ofemotional information on the one hand and dynamic information on theother. In the first part, we consider the task of emotion extraction forwhich the semantic gap is wider than it is with more traditionalthematic information. Therefore, we propose to study representationsaimed at modeling the many nuances of natural language used fordescribing emotional, hence subjective, information. Furthermore, wepropose to study the integration of semantic knowledge which provides,from a characterization perspective, support for extracting theemotional content of documents and, from a prediction perspective,assistance to the learning algorithm. In the second part, we study information dynamics: any corpus ofdocuments published over the Internet can be associated to sources inperpetual activity which exchange information in a continuousmovement. We explore three main lines of work: automaticallyidentified sources; the communities they form in a dynamic and verysparse description space; and the noteworthy themes they develop. Foreach we propose original extraction methods which we apply to a corpusof real data we have collected from information streams over the Internet