Dissertations / Theses on the topic 'Modèle génératifs'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 50 dissertations / theses for your research on the topic 'Modèle génératifs.'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse dissertations / theses on a wide variety of disciplines and organise your bibliography correctly.
Azeraf, Elie. "Classification avec des modèles probabilistes génératifs et des réseaux de neurones. Applications au traitement des langues naturelles." Electronic Thesis or Diss., Institut polytechnique de Paris, 2022. https://theses.hal.science/tel-03880848.
Full textMany probabilistic models have been neglected for classification tasks with supervised learning for several years, as the Naive Bayes or the Hidden Markov Chain. These models, called generative, are criticized because the induced classifier must learn the observations' law. This problem is too complex when the number of observations' features is too large. It is especially the case with Natural Language Processing tasks, as the recent embedding algorithms convert words in large numerical vectors to achieve better scores.This thesis shows that every generative model can define its induced classifier without using the observations' law. This proposition questions the usual categorization of the probabilistic models and classifiers and allows many new applications. Therefore, Hidden Markov Chain can be efficiently applied to Chunking and Naive Bayes to sentiment analysis.We go further, as this proposition allows to define the classifier induced from a generative model with neural network functions. We "neuralize" the models mentioned above and many of their extensions. Models so obtained allow to achieve relevant scores for many Natural Language Processing tasks while being interpretable, able to require little training data, and easy to serve
Hadjeres, Gaëtan. "Modèles génératifs profonds pour la génération interactive de musique symbolique." Thesis, Sorbonne université, 2018. http://www.theses.fr/2018SORUS027/document.
Full textThis thesis discusses the use of deep generative models for symbolic music generation. We will be focused on devising interactive generative models which are able to create new creative processes through a fruitful dialogue between a human composer and a computer. Recent advances in artificial intelligence led to the development of powerful generative models able to generate musical content without the need of human intervention. I believe that this practice cannot be thriving in the future since the human experience and human appreciation are at the crux of the artistic production. However, the need of both flexible and expressive tools which could enhance content creators' creativity is patent; the development and the potential of such novel A.I.-augmented computer music tools are promising. In this manuscript, I propose novel architectures that are able to put artists back in the loop. The proposed models share the common characteristic that they are devised so that a user can control the generated musical contents in a creative way. In order to create a user-friendly interaction with these interactive deep generative models, user interfaces were developed. I believe that new compositional paradigms will emerge from the possibilities offered by these enhanced controls. This thesis ends on the presentation of genuine musical projects like concerts featuring these new creative tools
Hadjeres, Gaëtan. "Modèles génératifs profonds pour la génération interactive de musique symbolique." Electronic Thesis or Diss., Sorbonne université, 2018. http://www.theses.fr/2018SORUS027.
Full textThis thesis discusses the use of deep generative models for symbolic music generation. We will be focused on devising interactive generative models which are able to create new creative processes through a fruitful dialogue between a human composer and a computer. Recent advances in artificial intelligence led to the development of powerful generative models able to generate musical content without the need of human intervention. I believe that this practice cannot be thriving in the future since the human experience and human appreciation are at the crux of the artistic production. However, the need of both flexible and expressive tools which could enhance content creators' creativity is patent; the development and the potential of such novel A.I.-augmented computer music tools are promising. In this manuscript, I propose novel architectures that are able to put artists back in the loop. The proposed models share the common characteristic that they are devised so that a user can control the generated musical contents in a creative way. In order to create a user-friendly interaction with these interactive deep generative models, user interfaces were developed. I believe that new compositional paradigms will emerge from the possibilities offered by these enhanced controls. This thesis ends on the presentation of genuine musical projects like concerts featuring these new creative tools
Vandewalle, Vincent. "Estimation et sélection en classification semi-supervisée." Phd thesis, Université des Sciences et Technologie de Lille - Lille I, 2009. http://tel.archives-ouvertes.fr/tel-00447141.
Full textCaillon, Antoine. "Hierarchical temporal learning for multi-instrument and orchestral audio synthesis." Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS115.
Full textRecent advances in deep learning have offered new ways to build models addressing a wide variety of tasks through the optimization of a set of parameters based on minimizing a cost function. Amongst these techniques, probabilistic generative models have yielded impressive advances in text, image and sound generation. However, musical audio signal generation remains a challenging problem. This comes from the complexity of audio signals themselves, since a single second of raw audio spans tens of thousands of individual samples. Modeling musical signals is even more challenging as important information are structured across different time scales, from micro (e.g. timbre, transient, phase) to macro (e.g. genre, tempo, structure) information. Modeling every scale at once would require large architectures, precluding the use of resulting models in real time setups for computational complexity reasons.In this thesis, we study how a hierarchical approach to audio modeling can address the musical signal modeling task, while offering different levels of control to the user. Our main hypothesis is that extracting different representation levels of an audio signal allows to abstract the complexity of lower levels for each modeling stage. This would eventually allow the use of lightweight architectures, each modeling a single audio scale. We start by addressing raw audio modeling by proposing an audio model combining Variational Auto Encoders and Generative Adversarial Networks, yielding high-quality 48kHz neural audio synthesis, while being 20 times faster than real time on CPU. Then, we study how autoregressive models can be used to understand the temporal behavior of the representation yielded by this low-level audio model, using optional additional conditioning signals such as acoustic descriptors or tempo. Finally, we propose a method for using all the proposed models directly on audio streams, allowing their use in realtime applications that we developed during this thesis. We conclude by presenting various creative collaborations led in parallel of this work with several composers and musicians, directly integrating the current state of the proposed technologies inside musical pieces
Pagliarini, Silvia. "Modeling the neural network responsible for song learning." Thesis, Bordeaux, 2021. http://www.theses.fr/2021BORD0107.
Full textDuring the first period of their life, babies and juvenile birds show comparable phases of vocal development: first, they listen to their parents/tutors in order to build a neural representation of the experienced auditory stimulus, then they start to produce sound and progressively get closer to reproducing their tutor song. This phase of learning is called the sensorimotor phase and is characterized by the presence of babbling, in babies, and subsong, in birds. It ends when the song crystallizes and becomes similar to the one produced by the adults.It is possible to find analogies between brain pathways responsible for sensorimotor learning in humans and birds: a vocal production pathway involves direct projections from auditory areas to motor neurons, and a vocal learning pathway is responsible for imitation and plasticity. The behavioral studies and the neuroanatomical structure of the vocal control circuit in humans and birds provide the basis for bio-inspired models of vocal learning.In particular, birds have brain circuits exclusively dedicated to song learning, making them an ideal model for exploring the representation of vocal learning by imitation of tutors.This thesis aims to build a vocal learning model underlying song learning in birds. An extensive review of the existing literature is discussed in the thesis: many previous studies have attempted to implement imitative learning in computational models and share a common structure. These learning architectures include the learning mechanisms and, eventually, exploration and evaluation strategies. A motor control function enables sound production and sensory response models either how sound is perceived or how it shapes the reward. The inputs and outputs of these functions lie (1)~in the motor space (motor parameters’ space), (2)~in the sensory space (real sounds) and (3)~either in the perceptual space (a low dimensional representation of the sound) or in the internal representation of goals (a non-perceptual representation of the target sound).The first model proposed in this thesis is a theoretical inverse model based on a simplified vocal learning model where the sensory space coincides with the motor space (i.e., there is no sound production). Such a simplification allows us to investigate how to introduce biological assumptions (e.g. non-linearity response) into a vocal learning model and which parameters influence the computational power of the model the most. The influence of the sharpness of auditory selectivity and the motor dimension are discussed.To have a complete model (which is able to perceive and produce sound), we needed a motor control function capable of reproducing sounds similar to real data (e.g. recordings of adult canaries). We analyzed the capability of WaveGAN (a Generative Adversarial Network) to provide a generator model able to produce realistic canary songs. In this generator model, the input space becomes the latent space after training and allows the representation of a high-dimensional dataset in a lower-dimensional manifold. We obtained realistic canary sounds using only three dimensions for the latent space. Among other results, quantitative and qualitative analyses demonstrate the interpolation abilities of the model, which suggests that the generator model we studied can be used as a motor function in a vocal learning model.The second version of the sensorimotor model is a complete vocal learning model with a full action-perception loop (i.e., it includes motor space, sensory space, and perceptual space). The sound production is performed by the GAN generator previously obtained. A recurrent neural network classifying syllables serves as the perceptual sensory response. Similar to the first model, the mapping between the perceptual space and the motor space is learned via an inverse model. Preliminary results show the influence of the learning rate when different sensory response functions are implemented
Kenmei, Youta Bénédicte Ramelie. "Génération de programmes modèles pour la représentation et l'analyse de profils d'exécution : Le modèle périodique-linéaire." Université Louis Pasteur (Strasbourg) (1971-2008), 2006. https://publication-theses.unistra.fr/public/theses_doctorat/2006/KENMEI_YOUTA_Benedicte_Ramelie_2006.pdf.
Full textThis work presents a novel approach for program execution traces analysis and modeling. After a short introduction where it is shown that general methods for data analysis are not well-adapted to understand the behavior of programs, we focus on the design of our model, called the Periodic Linear Model. It is based on the representation of a program behavior by another program. The program that models the behavior of the first one is made of sequences of nested loops, in which the functions of the innermost level compute the values of the input program trace from the loop indices. Each sequence of nested loops correspond to a particular definition of a program phase, that is a set of linearly linked intervals; these intervals are identifed through our periodic linear interpolation method. From this representation, we show that many analysis ans optimizations are possible in conjunction with some static analysis methods like the polytope model, and also that the information resulting from our model are more precise, adapted and exploitable than those obtained by general data analysis methods
Lucas, Thomas. "Modèles génératifs profonds : sur-généralisation et abandon de mode." Thesis, Université Grenoble Alpes, 2020. http://www.theses.fr/2020GRALM049.
Full textThis dissertation explores the topic of generative modelling of natural images,which is the task of fitting a data generating distribution.Such models can be used to generate artificial data resembling the true data, or to compress images.Latent variable models, which are at the core of our contributions, seek to capture the main factors of variations of an image into a variable that can be manipulated.In particular we build on two successful latent variable generative models, the generative adversarial network (GAN) and Variational autoencoder (VAE) models.Recently GANs significantly improved the quality of images generated by deep models, obtaining very compelling samples.Unfortunately these models struggle to capture all the modes of the original distribution, ie they do not cover the full variability of the dataset.Conversely, likelihood based models such as VAEs typically cover the full variety of the data well and provide an objective measure of coverage.However these models produce samples of inferior visual quality that are more easily distinguished from real ones.The work presented in this thesis strives for the best of both worlds: to obtain compelling samples while modelling the full support of the distribution.To achieve that, we focus on i) the optimisation problems used and ii) practical model limitations that hinder performance.The first contribution of this manuscript is a deep generative model that encodes global image structure into latent variables, built on the VAE, and autoregressively models low level detail.We propose a training procedure relying on an auxiliary loss function to control what information is captured by the latent variables and what information is left to an autoregressive decoder.Unlike previous approaches to such hybrid models, ours does not need to restrict the capacity of the autoregressive decoder to prevent degenerate models that ignore the latent variables.The second contribution builds on the standard GAN model, which trains a discriminator network to provide feedback to a generative network.The discriminator usually assesses the quality of individual samples, which makes it hard to evaluate the variability of the data.Instead we propose to feed the discriminator with emph{batches} that mix both true and fake samples, and train it to predict the ratio of true samples in the batch.These batches work as approximations of the distribution of generated images and allows the discriminator to approximate distributional statistics.We introduce an architecture that is well suited to solve this problem efficiently,and show experimentally that our approach reduces mode collapse in GANs on two synthetic datasets, and obtains good results on the CIFAR10 and CelebA datasets.The mutual shortcomings of VAEs and GANs can in principle be addressed by training hybrid models that use both types of objective.In our third contribution, we show that usual parametric assumptions made in VAEs induce a conflict between them, leading to lackluster performance of hybrid models.We propose a solution based on deep invertible transformations, that trains a feature space in which usual assumptions can be made without harm.Our approach provides likelihood computations in image space while being able to take advantage of adversarial training.It obtains GAN-like samples that are competitive with fully adversarial models while improving likelihood scores over existing hybrid models at the time of publication, which is a significant advancement
Loyauté, Gautier. "Un modèle génératif pour le développement de serveurs Internet." Phd thesis, Université Paris-Est, 2008. http://tel.archives-ouvertes.fr/tel-00470539.
Full textMaugé, Rudy. "Modèles de génération des marées internes." Brest, 2006. http://www.theses.fr/2006BRES2040.
Full textThis thesis studies the generation of internal tides and their propagation toward the deep ocean. We emphasize the major rote of interactions between the modes, and more particularly their influence on the nonlinear and nonhydrostatic internal-tide evolution and its subsequent disintegration in solitons. The sequence of internal tide generation models derived here, gradually leads to a more realistic representation of stratification and topography. A weakly nonlinear and nonhydrostatic internal tide model with three homogeneous layers is developed to study the influence of a double thermocline (e. G. Both a seasonal and a permanent one) on the generation of solitons. The results show the crucial rote of the second, deeper interface as it accelerates and intensifies the formation of solitons. For a continuous representation of stratification, a linear and hydrostaic model for the generation and propagation of internal tide is presented. It's derived using a modal approach, in which the topography is assumed to be slowly varying with respect to the horizontal internal-tide length scale. For parameters representative of the Gay of Biscay, the model shows how the internal-tide energy propagates toward the deep ocean, affecting the whole water column. This model offers the benefit of allowing a straightforward extension to a weakly nonlinear and nonhydrostatic internal-tide generation model, and is shown to be able to describe the local generation of solitons in the central Gay of Biscay, in accordance with observations
Webanck, Antoine. "Génération procédurale d'effets atmosphériques." Thesis, Lyon, 2019. http://www.theses.fr/2019LYSE1109/document.
Full textThis thesis focusses on the synthetization of natural landscapes, and more particularly on their celestial part. The aspect of the sky is governed by plenty of atmospheric phenomena, among which clouds play a major role for they are recurrent and widespread. Even without directly considering the sky, the density of the clouds allows them to intensely modify the global illumination of a landscape. The work of this thesis thus focuses mainly on the editing, modelling and animation of cloud areas of landscape dimensions.Because the thermodynamic simulation of cloud formation is hard to control and its maximum resolution quickly limits the details of the simulated volume, we propose instead a procedural generation method. We build a lightweight cloudscape model as a hierarchy of functions. The finest details are obtained by composing procedural noises and reproduce the specific shapes of different kinds of clouds. The large-scale cloud presence is described at a high level and at different times by maps drawn by the user. These discrete maps are transformed into implicit static primitives and then interpolated by morphing, accounting for relief and winds in order to produce coherent trajectories. The implicit field obtained by mixing the interpolating primitives represents the spatiotemporal field of cloud density. Images are finally synthesized by rendering of the atmospheric participative medium according to our own implementation, executed in parallel on a graphic card
Thomas, Frédéric. "Contribution à la prise en compte des plates-formes logicielles d'exécution dans une ingénierie générative dirigée par les modèles." Phd thesis, Université d'Evry-Val d'Essonne, 2008. http://tel.archives-ouvertes.fr/tel-00382556.
Full textGenevaux, Jean-David. "Représentation, modélisation et génération procédurale de terrains." Thesis, Lyon 2, 2015. http://www.theses.fr/2015LYO22013/document.
Full textThis PhD (entitled "Representation, modelisation and procedural generation of terrains") is related to movie and videogames digital content creation, especially natural scenes.Our work is dedicated to handle and to generate landscapes efficently. We propose a new model based on a construction tree inside which the user can handle parts of the terrain intuitively. We also present techniques to efficently visualize such model. Finally, we present a new algorithm for generating large-scale terrains exhibiting hierarchical structures based on their hydrographic networks: elevation is generated in a broad compliance to water-tansport principles without having to resort on costly hydraulic simulations
Rabesandratana, Paul Eliot. "Health spending and labor productivity in an aging economy." Thesis, Lille 1, 2015. http://www.theses.fr/2015LIL12021/document.
Full textFrench population is aging and this demographic mutation should also occur in the coming decades. The negative economic effects of population aging are well-known but the magnitude of these effects depends partially on the evolution of labor productivity. To determine the evolution of labor productivity in France, this dissertation focuses on the economic effects of health spending. Indeed, on one side, health spending, by improving population’s health status, affect positively labor productivity. On the other side, these health expenditures foster the aging process by extending population longevity. This dissertation aims then to check if productivity gains from health spending are enough to annihilate the negative economic effects of population aging. We demonstrate theoretically that private health spending generate a positive externality affecting positively labor productivity (Chapter I). However, our empirical results underline that this positive effect is limited when we consider the out-of-pocket expenditure on health (Chapter II). Thereafter, we assess the productivity gains resulting from the health status enhancement of French population by using a generational accounting model. It appears that the productivity gains should be significant but not enough to annihilate the negative economic effects of population aging (Chapter III). The simulation outcomes provided by our applied general equilibrium model confirm this result (Chapter IV)
Bienvenu, Meghyn. "La génération de conséquences en logique modale." Toulouse 3, 2009. http://thesesups.ups-tlse.fr/923/.
Full textThe key notion in consequence finding is that of prime implicates, which are defined to be the logically strongest clausal consequences of a formula. Prime implicates have proven useful in artificial intelligence, especially in knowledge compilation and abductive reasoning. In this thesis, we extend the investigation of prime implicates from propositional logic to the basic multi-modal logic Kn. We begin by comparing the properties of several plausible definitions of prime implicates in Kn in order to isolate the most suitable definition. We next study the computational aspects of the selected definition. Specifically, we provide algorithms for prime implicate generation and recognition, and we study the complexity of these tasks. Finally, we show how our notion of prime implicates can be used to define a normal form for Kn with interesting knowledge compilation properties
Augeraud-Veron, Emmanuelle. "L'anticipation dans les modèles à générations imbriquées." La Rochelle, 2000. http://www.theses.fr/2000LAROS059.
Full textAmeur-Boulifa, Rabéa. "Génération de modèles comportementaux des applications réparties." Nice, 2004. http://www.theses.fr/2004NICE4094.
Full textFrom the formal semantics of ProActive – 100 % Java library for concurrent, distributed, and mobile computing -, we build, in a compositional way, finite models of finite abstract applications. These models are described mathematically and graphically. The procedure for building, of which we guaranty the ending, is described by semantics rules applied to an intermediate form of programs obtained by static analysis. Afterwards, these rules are extended so as to build parameterized models of infinite applications. Practically, a prototype for analysing a core of Java and ProActive library is constructed. Moreover, some realistic examples are studied
Ferdjoukh, Adel. "Une approche déclarative pour la génération de modèles." Thesis, Montpellier, 2016. http://www.theses.fr/2016MONTT325/document.
Full textOwning data is useful in many different fields. Data can be used to test and to validate approaches, algorithms and concepts. Unfortunately, data is rarely available, is cost to obtain, or is not adapted to most of cases due to a lack of quality.An automated data generator is a good way to generate quickly and easily data that are valid, in different sizes, likelihood and diverse.In this thesis, we propose a novel and complete model driven approach, based on constraint programming for automated data generation
Franceschi, Jean-Yves. "Apprentissage de représentations et modèles génératifs profonds dans les systèmes dynamiques." Electronic Thesis or Diss., Sorbonne université, 2022. http://www.theses.fr/2022SORUS014.
Full textThe recent rise of deep learning has been motivated by numerous scientific breakthroughs, particularly regarding representation learning and generative modeling. However, most of these achievements have been obtained on image or text data, whose evolution through time remains challenging for existing methods. Given their importance for autonomous systems to adapt in a constantly evolving environment, these challenges have been actively investigated in a growing body of work. In this thesis, we follow this line of work and study several aspects of temporality and dynamical systems in deep unsupervised representation learning and generative modeling. Firstly, we present a general-purpose deep unsupervised representation learning method for time series tackling scalability and adaptivity issues arising in practical applications. We then further study in a second part representation learning for sequences by focusing on structured and stochastic spatiotemporal data: videos and physical phenomena. We show in this context that performant temporal generative prediction models help to uncover meaningful and disentangled representations, and conversely. We highlight to this end the crucial role of differential equations in the modeling and embedding of these natural sequences within sequential generative models. Finally, we more broadly analyze in a third part a popular class of generative models, generative adversarial networks, under the scope of dynamical systems. We study the evolution of the involved neural networks with respect to their training time by describing it with a differential equation, allowing us to gain a novel understanding of this generative model
Xia, Qing. "modèles et méthodes pour le génération de processus de fabrication reconfigurables." Thesis, Paris, ENSAM, 2017. http://www.theses.fr/2017ENAM0004/document.
Full textConventional manufacturing process planning approaches are inefficient to handle the process planning complexity induced by product variety and manufacturing dynamics. Reconfigurable process planning (RPP) is an emerging CAPP approach targeting to the generation of process plans for a product/part family. This thesis aims to give major contributions to the representation models and generation methods to support reconfigurable process planning at two granularity levels: product family and part family. The proposed approaches for RPP are compatible with an extended concept of product/part family which is defined by using the concept of “domain”. A feature-based product/part variety model is developed to represent the required information for RPP by using modular and platform-based techniques. Mathematical models and graph-based representation models are proposed to describe the reconfigurable process plan at two granularity levels. Based on the representation models, the generation methods and algorithms are then developed for RPP. In addition, a global framework is proposed to describe how the proposed RPP models and methods work together to handle the product/part variety and manufacturing dynamics. To test the feasibility of the proposed models and methods, a gear pump family and an oil pump body family are used as illustrative examples throughout this thesis
Paquette, Luc. "Un modèle pour la génération d'indices par une plateforme de tuteurs par traçage de modèle." Thèse, Université de Sherbrooke, 2013. http://hdl.handle.net/11143/6560.
Full textRoussel, Olivier. "Génération aléatoire de structures ordonnées par le modèle de Boltzmann." Paris 6, 2012. http://www.theses.fr/2012PA066282.
Full textUniform random generation is a central issue in combinatorics. Indeed, random sampling is virtually connected to all parts of combinatorics, whether to exact or asymptotic enumeration, or to the experimental verification of conjectures. Various methods have been developed in order to efficiently solve that issue. Boltzmann model is among them. This method, relaxing some constraints about the size of the object being currently generated, ensures a linear complexity in many actual cases, and can easily be automatized for various combinatorial classes. This thesis aims at enlarging the set of such admissible classes, while keeping the nice properties of linear complexity and ease of automation. The first part is devoted to the presentation of the Boltzmann model and existing Boltzmann samplers, and the study of their properties and mathematical foundations. In the second part, we introduce our idea of biasing those samplers in order to enlarge their range of validity. Firstly, we present a general extension, and then specialize it to several combinatorial operations such as the derivation, the shuffle product or the unpointing operation. Finally, we present a uniform random sampler for the Hadamard product. We highlight our algorithms through this thesis with examples and experimental results, illustrating the efficiency of our methods
Le, Riche Antoine. "Commerce international dans le modèle bi-sectoriel à générations imbriquées." Thesis, Aix-Marseille, 2014. http://www.theses.fr/2014AIXM2023/document.
Full textThis dissertation consists of three essays on the existence of endogenous cycles under dynamic efficiency and covers two different topics. The first one developed in chapter 2 addresses the relation between the structure of production and the occurrence of endogenous fluctuations. The second topic is developed in chapters 3 and 4, and tries to better understand the role of international trade on the emergence of sunspot cycles. These chapters explain that international trade may have a destabilizing effect.Chapter 2 examines how the existence of durable and non-durable goods are associated with the occurrence of endogenous cycles with dynamic efficiency. Main results indicate that sunspot fluctuations exist if young agents consume less non-durable goods than old agents. In this context, we show that a stabilization policy based on lump-sum taxes and lump-sum transfers allows to increase the welfare of agents and eliminate the existence of endogenous cycles.Chapter 3 analyzes a two-factor, two-good, two-country overlapping generations model. We assume that countries differ only with respect to their discount rate. We suppose that the factor of production are immobile across countries. Using a numerical simulation, we show that period-two cycles spread from one country to another.Chapter 4 considers a framework similar to the model of chapter 3 but with focus on asymmetric technology. We show through numerical simulations that the opening to international trade can create two-period cycles that can exist in the world economy even though the closed-economy equilibrium in each country is saddle-point stable
Jacques, Julien. "Contribution à l'apprentissage statistique à base de modèles génératifs pour données complexes." Habilitation à diriger des recherches, Université des Sciences et Technologie de Lille - Lille I, 2012. http://tel.archives-ouvertes.fr/tel-00761184.
Full textRicard, Ludovic. "Génération de modèles de réservoir sur maillage flexible." Paris 11, 2005. http://www.theses.fr/2005PA112342.
Full textThe high level geostatistic description of the subsurface are often far too detailed for use in routine flow simulators. To make flow simulations tractable, the number of grid blocks has to be reduced: an approximation, still relevant with flow description, is necessary. In this work, we place the emphasis on the scaling procedure from the fine scale model to the multi-scale reservoir model. Two main problems appear: Near wells, faults and channels, the volume of flexible cells may be less than fine ones, so we need to solve a downscaling problem; Far from these regions, the volume of cells are bigger than fine ones so we need to solve an upscaling problem. In this work, research has been done on each of these three areas: downscaling, upscaling and fluid flow simulation. For each of these subjects, a review, some news improvements and comparative study are proposed. The proposed downscaling method is build to be compatible with existing data integration methods. The comparative study shows that empirical methods are not enough accurate to solve the problem. Concerning the upscaling step, the proposed approach is based on an existing method: the perturbed boundary conditions. An extension to unstructured mesh is developed for the inter-cell permeability tensor. The comparative study shows that numerical methods are not always as accurate as expected and the empirical model can be sufficient in lot of cases. A new approach to single-phase fluid flow simulation is developed. This approach can handle with full tensorial permeability fields with source or sink terms
Grechka, Asya. "Image editing with deep neural networks." Electronic Thesis or Diss., Sorbonne université, 2023. https://accesdistant.sorbonne-universite.fr/login?url=https://theses-intra.sorbonne-universite.fr/2023SORUS683.pdf.
Full textImage editing has a rich history which dates back two centuries. That said, "classic" image editing requires strong artistic skills as well as considerable time, often in the scale of hours, to modify an image. In recent years, considerable progress has been made in generative modeling which has allowed realistic and high-quality image synthesis. However, real image editing is still a challenge which requires a balance between novel generation all while faithfully preserving parts of the original image. In this thesis, we will explore different approaches to edit images, leveraging three families of generative networks: GANs, VAEs and diffusion models. First, we study how to use a GAN to edit a real image. While methods exist to modify generated images, they do not generalize easily to real images. We analyze the reasons for this and propose a solution to better project a real image into the GAN's latent space so as to make it editable. Then, we use variational autoencoders with vector quantification to directly obtain a compact image representation (which we could not obtain with GANs) and optimize the latent vector so as to match a desired text input. We aim to constrain this problem, which on the face could be vulnerable to adversarial attacks. We propose a method to chose the hyperparameters while optimizing simultaneously the image quality and the fidelity to the original image. We present a robust evaluation protocol and show the interest of our method. Finally, we abord the problem of image editing from the view of inpainting. Our goal is to synthesize a part of an image while preserving the rest unmodified. For this, we leverage pre-trained diffusion models and build off on their classic inpainting method while replacing, at each denoising step, the part which we do not wish to modify with the noisy real image. However, this method leads to a disharmonization between the real and generated parts. We propose an approach based on calculating a gradient of a loss which evaluates the harmonization of the two parts. We guide the denoising process with this gradient
Mounsif, Mehdi. "Exploration of Teacher-Centered and Task-Centered paradigms for efficient transfer of skills between morphologically distinct robots." Thesis, Université Clermont Auvergne (2017-2020), 2020. http://www.theses.fr/2020CLFAC065.
Full textRecently, it has been possible to observe the acceleration of robot deployment in domains beyondthe usual industrial and manufacturing framework. However, for the majority of autonomoustasks, the definition of an analytical model or the search for an optimal (or acceptable) solutionrequires resources that are seldom available in real-time, thus favoring learning-based techniques.Indeed, learned models present the advantage of being model-free as well as having a constantexecution time, consequently enabling the realization of highly complex trajectories and tasks.Data-driven techniques, however, are hindered by considerable training time, frequently requiringmillions of examples and interactions with their environment to build acceptable control policies.As such, knowledge transfer, also known as transfer learning, between models is crucial for largescaledeployment of learned policies. Although transmission strategies have been the focus ofrecent concerns, they are mainly directed towards the fields of vision or language understandingand are not directly applicable to control environments where skill transfer is likely to happenbetween robots with different kinematic structures. The works presented in this thesis manuscriptfocus precisely on this point and aims at determining to what extent understanding between twomorphologically distinct entities is possible. This question is explored through the introductionof two distinct paradigms: Task-Centered and Teacher-Centered. The Task-Centered family oftechniques is based on the idea of the separation of task-related know-how from robot control policy.Such an independent kernel can therefore be passed on to other robots of different morphology andideally make it possible for the new agent to perform the task. In this context, several blueprintsfor creating this kernel are proposed and evaluated on a wide range of simulated environments.However, despite the attractive prospects of this formulation, the "one-size-fits-all" character ofTask-Centered techniques is not free of limitations which are extensively discussed. It is in thiscontext that Teacher-Centered approaches are introduced. Pursuing the same objective, theseinnovative procedure involve an expert agent from which the knowledge related to the task mustbe distilled into the target agent. To do this, an original metric is used to circumvent the structuraldifferences between the target agent and the expert agent and allow, despite this distinction, theerror to be back-propagated in order to optimize the agent
Muraro, Alain. "Génération de mouvements optimaux pour un robot quadrupède." Nantes, 2002. http://www.theses.fr/2002NANT2091.
Full textCisneros, Iturbe Héctor Leonardo. "Modèle de génération de champs d'averses sur la région de Mexico." Montpellier 2, 2001. http://www.theses.fr/2001MON20055.
Full textChavoya, Pena Arturo. "Un modèle de développement artificiel pour la génération de structures cellulaires." Toulouse 1, 2007. http://publications.univ-tlse1.fr/681/.
Full textCell pattern formation has a crucial role in both artificial and natural development. This thesis presents an artificial development model for cell pattern generation based on the cellular automata (CA) paradigm. Cellular growth is controlled by a genome consisting of an artificial ragulatory network (ARN) and a series of structural genes. The genome was evolved by a genetic algorithm (GA) in order to produce 2D cell patterns through the selective activation and inhibition of genes. Morphogenetic gradients were used to provide cells with positional information that constrained cellular replication. After a genome was evolved, a single cell in the middle of the CA lattice was allowed to reproduce controlled by the ARN until a cell pattern was formed. The model was applied to the canonical problem of growing a french flag pattern
Boulifa, Rabea. "Génération de modèles comportemementaux des applications des applications réparties." Phd thesis, Université de Nice Sophia-Antipolis, 2004. http://tel.archives-ouvertes.fr/tel-00008927.
Full textautomatique de propriétés comportementales d'applications réparties par des
méthodes fondées sur les
modèles. En particulier, nous étudions le problème de
génération de modèles
à partir de programmes Java répartis et représentés par des systèmes de transitions
communiquants.
Pour ce faire, nous définissons une sémantique comportementale de programmes ProActive, une
librairie Java pour la programmation parallèle, distribuée et
concurrente. À partir de cette sémantique nous construisons des modèles
comportementaux pour
des abstractions finies d'applications écrites dans ce langage. Ces
modèles sont basés sur la sémantique des algèbres de
processus et peuvent donc être construits de manière compositionnelle et
hiérarchique.
La construction de modèles finis n'est pas toujours possible. Pour
pouvoir traiter des problèmes prenant en compte des données, ainsi que
des problèmes concernant des topologies non bornées d'objets
répartis, nous définissons une nouvelle notion de modèleles
hiérarchiques, à base de systèmes de transitions paramétrés et de
réseau de synchronisation paramétrés. Moyennant des abstractions ces modèles permettent de
spécifier des applications possiblement infinies par des représentations
expressives, finies, et plus proche de la structure du code.
Par ailleurs, nous définissons un système de règles sémantiques
permettant de générer automatiquement ces modèles (finis ou
paramétrés) à partir d'une forme intermédiaire, obtenue par analyse statique,
des programmes analysés.
Les modèles ainsi générés sont exploitables directement ou après
instantiation par des outils de vérification.
Job, Isabelle. "La dette publique dans les modèles à générations imbriquées." Paris 10, 1996. http://www.theses.fr/1996PA100007.
Full textThe goal of this thesis is to examine, from a theoric point of view, the long term implications of public debt. We base our analysis on the overlapping generation models to study public indebteness which can be used for redistributive transfers or for financing public "productive" spending. In the first part, we examine the public debt without taking into account its real counterpart, that is public spending, to focus on the redistributive effects of this financial policy. We discuss the theorem of debt-neutrality and developp its critics. We are also interested in the problem of sustainability and particularly, we examine the conditions for which the government can run a ponzi game when he faces a solvability constraint. In the second part, we give up this simple way of thinking about a financial policy completely disconnected of its real counterpart. Then, we focus on the role of public debt as a political tool using for financing public spending with long term implications on economic activity. When we recognize an utility for public spending, fiscal policy can be an additional instrument for policy makers to reduce economic problems. We build two overlapping generation models where public investment has positive effects on the productivity of private factors. We examine how the level and the way of financing this kind of spending can affect the rate of growth, the unemployment level and the welfare in the economy
Uribe, Lobello Ricardo. "Génération de maillages adaptatifs à partir de données volumiques de grande taille." Thesis, Lyon 2, 2013. http://www.theses.fr/2013LYO22024.
Full textIn this document, we have been interested in the surface extraction from the volumetric representation of an object. With this objective in mind, we have studied the spatial subdivision surface extraction algorithms. This approaches divide the volume in order to build a piecewise approximation of the surface. The general idea is to combine local and simple approximations to extract a complete representation of the object's surface.The methods based on the Marching Cubes (MC) algorithm have problems to produce good quality and to handle adaptive surfaces. Even if a lot of improvements to MC have been proposed, these approaches solved one or two problems but they don't offer a complete solution to all the MC drawbacks. Dual methods are more adapted to use adaptive sampling over volumes. These methods generate surfaces that are dual to those generated by the Marching Cubes algorithm or dual grids in order to use MC methods. These solutions build adaptive meshes that represent well the features of the object. In addition, recent improvements guarantee that the produced meshes have good geometrical and topological properties.In this dissertation, we have studied the main topological and geometrical properties of volumetric objects. In a first stage, we have explored the state of the art on spatial subdivision surface extraction methods in order to identify theirs advantages, theirs drawbacks and the implications of theirs application on volumetric objects. We have concluded that a dual approach is the best option to obtain a good compromise between mesh quality and geometrical approximation. In a second stage, we have developed a general pipeline for surface extraction based on a combination of dual methods and connected components extraction to better capture the topology and geometry of the original object. In a third stage, we have presented an out-of-core extension of our surface extraction pipeline in order to extract adaptive meshes from huge volumes. Volumes are divided in smaller sub-volumes that are processed independently to produce surface patches that are later combined in an unique and topologically correct surface. This approach can be implemented in parallel to speed up its performance. Test realized in a vast set of volumes have confirmed our results and the features of our solution
Ressencourt, Hervé. "Diagnostic hors-ligne à base de modèles : approche multi-modèle pour la génération automatique de séquences de tests : application au domaine de l'automobile." Toulouse 3, 2008. http://thesesups.ups-tlse.fr/2151/.
Full textThis thesis deals with the problem of off-board diagnosis in the automotive domain. The work has consisted in proposing and implementing an operational model based approach that determines the best sequences of tests to be performed by the garage mechanic to localise a faulty component on a vehicle. A multi-model approach is proposed for the description of mechatronic systems, which allows us to handle the functional complexity of embedded systems and to match functional symptoms with a set of faults on hardware / software components. The test sequencing problem is approached along a next best test strategy based on a local heuristic. This strategy enables an interactive diagnostic session, allowing more flexibility and leaving with the human operator the initiative to accept or reject the proposed test. A software prototype has been developed and tested on the rear wiper system of real vehicles. This thesis, supported by a CIFRE grant, is the result of collaboration between the company ACTIA and the research center LAAS-CNRS in the framework of the common laboratory Autodiag (LAAS, IRIT, ACTIA) which aims at developing new methods for diagnosis in the automotive domain
Sidibe, Modibo. "Structural empirical models of spatial inequalities: housing choices, policies and generational consequences." Phd thesis, Université Lumière - Lyon II, 2011. http://pastel.archives-ouvertes.fr/pastel-00702737.
Full textGrondin, Pierre. "Génération de code ciblant un microcontrôleur MPC555 à partir d'un modèle Simulink." Mémoire, École de technologie supérieure, 2003. http://espace.etsmtl.ca/738/1/GRONDIN_Pierre.pdf.
Full textCosta, Catherine. "Harmonisation fiscale européenne : utilisation d'un modèle à générations imbriquées à deux pays." Aix-Marseille 2, 1998. http://www.theses.fr/1998AIX24019.
Full textKral, Quentin. "Etude des disques de débris avec un modèle numérique de nouvelle génération." Observatoire de Paris (1667-....), 2014. https://hal.science/tel-02095162.
Full textThe present work was dedicated to the development and use of a new generation numerical code, able to model in a self-consistent was debris discs' evolution. These dust discs, optically thin, sometimes very old, are present around at least 30 [percent] of stars. Numerous observations with Herschel, the VLT, the LBTI, ALMA and soon the EELT or the JWST give images that are always more accurate of these objects, for which, the visible structures that are observed are astonishingly different from each other, leaving often a mystery to be solved. Various models attempt to reveal the link between the observed structures, the system spectra, its photometry and the interactions with the close- environment. Indeed, the presence of planets, through its resonances or gravitational perturbations, can induce the production of structures in these discs (warp, brightness asymetry, dust clump. . . ). Like wise, the presence of gas, of a stellar companion or a massive collision could explain a part of observed asymmetries. Nonetheless, robust models have to be developed, that are able to take into account the complexity of the physics associated with these discs, especially the interactions between the collisions and dynamics, in order to give the closest outputs to observations. Firstly, our team developed a model named DyCoSS able to partially couple the collisions and dynamics. Hoewer, because of numerous limitations of this model, we decided to develop a totally new, stand alone code, called LIDT-DD : the first self-consistent model fully coupling the collisions and dynamics and taking into account the evolution of dust in debris discs. Moreover, the code has been coupled to GRaTer, a radiative transfer code, able to treat the dynamics of grains according to their size, composition, stellar type,. . . Once the model, trying to be as generic as possible, was implemented, we carried out a huge number of tests to validate it. In a second part, we analyzed massive collisions in debris discs with LIDT-DD. These collisions are planetary formation markers and deserve a special attention, to help to test the planetary formation models or even to optimize our chances to detect forming planets. With our model, we explore the observational signatures of these massive collisions and are able to follow the system evolution on most of its lifetime. Hence, we distinguish different phases and give the timescales associated. We predict that these collisisions are detectable by Spitzer/MIPS at 24 microns and show that for nearby systems (about 10 pc), we are able to resolve such events with instruments such as SPHERE/VLT in scattered light or MIRI/JWST in thermal emission
Baelde, Maxime. "Modèles génératifs pour la classification et la séparation de sources sonores en temps-réel." Thesis, Lille 1, 2019. http://www.theses.fr/2019LIL1I058/document.
Full textThis thesis is part of the A-Volute company, an audio enhancement softwares editor. It offers a radar that translates multi-channel audio information into visual information in real-time. This radar, although relevant, lacks intelligence because it only analyses the audio stream in terms of energy and not in terms of separate sound sources. The purpose of this thesis is to develop algorithms for classifying and separating sound sources in real time. On the one hand, audio source classification aims to assign a label (e.g. voice) to a monophonic (one label) or polyphonic (several labels) sound. The developed method uses a specific feature, the normalized power spectrum, which is useful in both monophonic and polyphonic cases due to its additive properties of the sound sources. This method uses a generative model that allows to derive a decision rule based on a non-parametric estimation. The real-time constraint is achieved by pre-processing the prototypes with a hierarchical clustering. The results are encouraging on different databases (owned and benchmark), both in terms of accuracy and computation time, especially in the polyphonic case. On the other hand, source separation consists in estimating the sources in terms of signal in a mixture. Two approaches to this purpose were considered in this thesis. The first considers the signals to be found as missing data and estimates them through a generative process and probabilistic modelling. The other approach consists, from sound examples present in a database, in computing optimal transformations of several examples whose combination tends towards the observed mixture. The two proposals are complementary, each having advantages and drawbacks (computation time for the first, interpretability of the result for the second). The experimental results seem promising and allow us to consider interesting research perspectives for each of the proposals
Beller, Grégory. "Analyse et modèle génératif de l'expressivité : application à la Parole et à l'Interprétation musicale." Phd thesis, Université Pierre et Marie Curie - Paris VI, 2009. http://tel.archives-ouvertes.fr/tel-00431104.
Full textMendoza, Chavez Gustavo. "Approche semi-automatique de génération de modèles bielles-et-tirants." Thesis, Paris Est, 2018. http://www.theses.fr/2018PESC1030/document.
Full textWithin the field of Reinforced Concrete (RC) structures and more specifically, at the design of non-flexural elements such as corbels, nibs, and deep beams, the rational procedure of conception and justification referred as Strut-and-Tie Method (STM) has shown some advantages over classical algorithms of reinforcement computation based on FE analysis (eg. Wood-Armer or Capra-Maury).The STM remains a suitable alternative for the design of concrete structures presenting either elastic or plastic behaviour whose application framework is well defined in concrete structures’ design codes like the EuroCodes and the AASHTO-LRFD Bridge Design Specifications.Nevertheless, this method has the main inconvenient of requiring a high amount of resources investment in terms of highly experienced personal or in terms of computational capacity for, respectively, its manual application or an automatic approach through topology optimisation.The document proposes a light alternative, in terms of required iterations, to the automation of the STM, which starts from the statement that the resultant struts and ties of a suitable ST model can be distributed according to the direction of the principal stresses, $sigma_{III}$ and $sigma_{I}$ , obtained from a planar or a three-dimensional FE model
Constant, Camille. "Génération automatique de tests pour modèles avec variables ou récursivité." Rennes 1, 2008. ftp://ftp.irisa.fr/techreports/theses/2008/constant.pdf.
Full textThis thesis adresses the problem of automatic test case generation for testing the conformance of a reactive implementation. We first propose a methodology which extends the testing theory based on the ioco conformance relation by distinguishing three levels of description: properties, specification and implementation. The methogology integrates verification and conformance testing. The execution of the generated test cases on the implementation allows detecting conformance violations between implementation and specification, but also violation/satisfaction of the properties by the implementation or the specification. Secondly, we introduce a more expressive specification model: recursive interprocedural specifications. The test generation method we propose is based on coreachability analysis, which allows deciding whether and how the test purpose can still be satisfied. However, although it is possible to carry out an exact analysis, the inability of test cases to inspect their own stack prevents them from fully using the coreachability information. We discuss this partial observation problem, its consequences, and how to minimise its impact. Finally, we experiment these methods of test generation on several examples and a case study
Govokhina, Oxana. "Modèles de génération de trajectoires pour l'animation de visages parlants." Phd thesis, Grenoble INPG, 2008. http://tel.archives-ouvertes.fr/tel-00363319.
Full textGovokhina, Oxana. "Modèles de génération de trajectoires pour l'animation de visages parlants." Phd thesis, Grenoble INPG, 2008. http://www.theses.fr/2008INPG0105.
Full textThe work performed during this thesis concerns visual speech synthesis in the context of humanoid animation. Our study proposes and implements control models for facial animation that generate articulatory trajectories from text. We have used 2 audiovisual corpuses in our work. First of all, we compared objectively and subjectively the main state-of-the-art models. Then, we studied the spatial aspect of the articulatory targets generated by HMM-based synthesis and concatenation-based synthesis that combines the advantages of these methods. We have proposed a new synthesis model named TDA (Task Dynamics for Animation). The TDA system plans the geometric targets by HMM synthesis and executes the computed targets by concatenation of articulatory segments. Then, we have studied the temporal aspect of the speech synthesis and we have proposed a model named PHMM (Phased Hidden Markov Model). The PHMM manages the temporal relations between different modalities related to speech. This model calculates articulatory gestures boundaries as a function of the corresponding acoustic boundaries between allophons. It has been also applied to the automatic synthesis of Cued speech in French. Finally, a subjective evaluation of the different proposed systems (concatenation, HMM, PHMM and TDA) is presented
Leonardis, Eleonora De. "Méthodes pour l'inférence en grande dimension avec des données corrélées : application à des données génomiques." Thesis, Paris, Ecole normale supérieure, 2015. http://www.theses.fr/2015ENSU0033/document.
Full textThe availability of huge amounts of data has changed the role of physics with respect to other disciplines. Within this dissertation I will explore the innovations introduced in molecular biology thanks to statistical physics approaches. In the last 20 years the size of genome databases has exponentially increased, therefore the exploitation of raw data, in the scope of extracting information, has become a major topic in statistical physics. After the success in protein structure prediction, surprising results have been finally achieved also in the related field of RNA structure characterisation. However, recent studies have revealed that, even if databases are growing, inference is often performed in the under sampling regime and new computational schemes are needed in order to overcome this intrinsic limitation of real data. This dissertation will discuss inference methods and their application to RNA structure prediction. We will discuss some heuristic approaches that have been successfully applied in the past years, even if poorly theoretically understood. The last part of the work will focus on the development of a tool for the inference of generative models, hoping it will pave the way towards novel applications
Rapelanoro, Rabenja Hériony. "Structure temporelle et stochastique : monnaie et équilibre dans les modèles à génération." Paris, EHESS, 1996. http://www.theses.fr/1996EHES0109.
Full textThe principal aim of this work is to develop an aspect of intertemporal model affected by uncertainity denoted "stochastic overlapping generation model. This extendes to infinite the discrete time period and hence states, identified as a "date-event" and eventualy the date of birth for some individuals. In first time, we develop the evolution process of the stochastic structure which there the "date-event" is an essential framework. This consideration is used by radner (1982), scmachtenberg (1988). To simplify the exposition, we assume that each consummer leaves in a number finite of dates or states which develop many markets contingents. For preserving the equilibrium in the differentes states, money is used, this permits to conserve values and to transfer her to a state to another. Secondly, we consider a deterministe formalisation of the intertemporel structure of the model with exchanges that money intervenes or not. This chapter permit to prove the existence of an equilibrium an adopting some assumptions wich defines this model as a version of okuno-zilicha (1981) model. Without stricte quasi-concavity of the utility fonction, but with strict monotony of preference and strict positivity of initial resource, we can assume the recouvrement of generation. Equilibrium existence can be proved with proceding on two stages. The stochastic structure of the problem is introduit by the processus of "event-date" influe the comodities, the initial endowment, the transfers and the individuals preferences or utilities. In this model, equilibrium is obtained where we suppose that exogene monetary system price do not permit no arbitrage
Chipot, Yveline. "Génération et modification de surfaces triangulées." Vandoeuvre-les-Nancy, INPL, 1991. http://www.theses.fr/1991INPL063N.
Full textMarsso, Lina. "Etude de génération de tests à partir d'un modèle pour les systèmes GALS." Thesis, Université Grenoble Alpes (ComUE), 2019. http://www.theses.fr/2019GREAM078.
Full textThis dissertation focuses on the model-based testing of GALS (GloballyAsynchronous and Locally Synchronous) systems, which are inherentlycomplex because of the combination of synchronous and asynchronous aspects.To cope with this complexity, we explore three directions:(1) techniques for synchronous components;(2) techniques for communication protocols between components; and(3) techniques for complete GALS systems, combining theresults of the two previous directions.In the first direction, we explore formal techniques for the functionaltesting of synchronous components.As a case-study, we reconsider the Message Authenticator Algorithm(MAA), a pioneering cryptographic function designed in the mid-80s, andformalize it as a synchronous dataflow.The modeling and validation of the MAA enabled us to discover variousmistakes in prior (informal and formal) specifications of the MAA, thetest vectors and code of the ISO 1987 and ISO 1990 standards, and incompilers and verification tools used by us.In the second direction, we explore the formalization and the functionaltesting of a communication protocol. As a case-study, we reconsider the formalization of the Transport Layer Security (TLS) handshake, a protocol responsible for the authentication and exchange of keys necessary to establish or resume a secure communication.Our model of the TLS version 1.3 has been validated by an approach using our new on-the-fly conformance test case generation tool, named TESTOR, developed on top of the CADP toolbox.TESTOR explores the model and generates automatically a set of controllable testcases or a complete test graph (CTG) to be executed on a physical implementation ofthe system.In the third direction, we propose a testing methodology for GALSsystems combining the two previous directions.We leverage the conformance test generation for asynchronous systems toautomatically derive realistic scenarios (inputs constraints andoracles), which are necessary ingredients for the unit testing of individual synchronouscomponents, and are difficult and error-prone to design manually.Thus our methodology integrates(1) synchronous and asynchronous concurrent models;(2) functional unit testing and behavioral conformance testing;and (3) various formal methods and their tool equipments.We illustrate our methodology on a simple, but relevant example inspiredby autonomous cars
Lacaze-Labadie, Rémi. "Planification et modèle graphique pour la génération dynamique de scénarios en environnements virtuels." Thesis, Compiègne, 2019. http://www.theses.fr/2019COMP2481/document.
Full textOur work is related to the training of crisis management in virtual environments. The specification of possible unfoldings of events in a simulation is essential for human learning in a virtual environment. This allows both to propose and orchestrate personalized learning situations and also to bring the learner toward relevant and educative scenarios. The work presented in this thesis focuses on the dynamic generation of scenarios and their execution in a virtual environment. For that, we aim at a set of objectives that are often contradictory : the freedom of action of the user, the generation of various scenarios that respect the authorial intent, the narrative control and the capacity of the system to adapt to deviations fromthe learner. The different approaches of interactive storytelling tackle more or less some of these objectives, but it is difficult to satisfy them all, and this is the challenge of our work. In addition to these objectives, we also aim at facilitating the modeling of the narrative content, which is still a real issue today when it comes to model complex environments such as the ones related to crisis management. We propose an emergent approachwhere the scenario experienced by the learner will emerge fromthe interactions between the learner, the virtual characters and our narrative system MENTA. MENTA is in charge of the narrative control by proposing a set of adjustments (over the simulation) that satisfies narrative objectives chosen by the trainer (e. g., a list of specific skills). These adjustments take the form of a prescribed scenario that is generated by MENTA via a planning engine that we have coupled with fuzzy cognitive maps through a macro-operator FRAG. A FRAG is used to model FRAGment of scenario in the form of scripted sequences of actions/events. The originality of our approach relies on a strong coupling between planning and graphical models which preserves the exploration capability and the generative power of a planning engine (which contributes to the generation of various and adaptable scenarios), while facilitating the modeling of narrative content as well as the authorial intent thanks to fragments of scenario that are scripted by the author and used during the planning process. We have worked on a concrete application example of scenarios dealing with the management of a massive influx of victims. Then, we have implemented MENTA and generated scenarios related to this example. Finally, we have tested and analyzed the performance of our system
Gombert, Baptiste. "Vers une nouvelle génération de modèles de glissements co-sismiques : analyse stochastique et approche multi-données." Thesis, Strasbourg, 2018. http://www.theses.fr/2018STRAH016/document.
Full textThe explosion in the amount and variety of available geodetic, tsunami, and seismological observations offers an outstanding opportunity to develop new seismic source models. But these data are sensitive to different sources of uncertainty and provide heterogeneous information, which makes the solution of the inverse problem non-unique.In this thesis, we use a Bayesian sampling method to propose new slip models, which benefit from an objective weighting of the various datasets by combining observational and modelling errors. These models are less affected by data overfit and allow a realistic assessment of posterior uncertainties. We apply this method to the study of slip processes occurring in three different tectonic contexts: the Landers earthquake (1992, Mw=7.3), the Ecuador-Colombia subduction zone which hosted the Pedernales earthquake (2016, Mw=7.8), and the intraslab Tehuantepec earthquake (2017, Mw=8.2). Through these analyses, we demonstrate how the study of the seismic cycle can benefit from rigorous uncertainty estimates and Bayesian sampling