Tesis sobre el tema "Optimisation par gradient"
Crea una cita precisa en los estilos APA, MLA, Chicago, Harvard y otros
Consulte los 45 mejores tesis para su investigación sobre el tema "Optimisation par gradient".
Junto a cada fuente en la lista de referencias hay un botón "Agregar a la bibliografía". Pulsa este botón, y generaremos automáticamente la referencia bibliográfica para la obra elegida en el estilo de cita que necesites: APA, MLA, Harvard, Vancouver, Chicago, etc.
También puede descargar el texto completo de la publicación académica en formato pdf y leer en línea su resumen siempre que esté disponible en los metadatos.
Explore tesis sobre una amplia variedad de disciplinas y organice su bibliografía correctamente.
Genest, Laurent. "Optimisation de forme par gradient en dynamique rapide". Thesis, Lyon, 2016. http://www.theses.fr/2016LYSEC022/document.
Texto completoIn order to face their new industrial challenges, automotive constructors wish to apply optimization methods in every step of the design process. By including shape parameters in the design space, increasing their number and their variation range, new problematics appeared. It is the case of crashworthiness. With the high computational time, the nonlinearity, the instability and the numerical dispersion of this rapid dynamics problem, metamodeling techniques become to heavy for the standardization of those optimization methods. We face this problematic: ”How can we carry out shape optimization in rapid dynamics with a high number of parameters ?”. Gradient methods are the most likely to solve this problematic. Because the number of parameters has a reduced effect on the optimization cost, they allow optimization with a high number of parameters. However, conventional methods used to calculate gradients are ineffective: the computation cost and the numerical noise prevent the use of finite differences and the calculation of a gradient by deriving the rapid dynamics equations is not currently available and would be really intrusive towards the software. Instead of determining the real gradient, we decided to estimate it. The Equivalent Static Loads Method is an optimization method based on the construction of a linear static problem equivalent to the rapid dynamic problem. By using the sensitivity of the equivalent problem as the estimated gradient, we have optimized rapid dynamic problems with thickness parameters. It is also possible to approximate the derivative with respect to the position of the nodes of the CAE model. But it is more common to use CAD parameters in shape optimization studies. So it is needed to have the sensitivity of the nodes position with these CAD parameters. It is possible to obtain it analytically by using parametric surface for the shape and its poles as parameters. With this link between nodes and CAD parameters, we can do shape optimization studies with a large number of parameters and this with a low optimization cost. The method has been developed for two kinds of crashworthiness objective functions. The first family of criterions is linked to a nodal displacement. This category contains objectives like the minimization of the intrusion inside the passenger compartment. The second one is linked to the absorbed energy. It is used to ensure a good behavior of the structure during the crash
De, Gournay Frédéric. "Optimisation de formes par la méthode des courbes de niveaux". Phd thesis, Ecole Polytechnique X, 2005. http://tel.archives-ouvertes.fr/tel-00446039.
Texto completoZhao, Zhidong. "Optimisation d'antennes et de réseaux d'antennes planaires par gradient de forme et ensembles de niveaux (Level Sets)". Thesis, Nice, 2015. http://www.theses.fr/2015NICE4097.
Texto completoThe objective of this thesis work is to find the optimal shape of planar antenna elements and arrays from imposed constraints (e.g. desired or imposed radiation patterns, gain or directivity) or to reconstruct the shape from experimental measurements. The optimization algorithm is based on the gradient-type method and an active contour reconstruction by means of the Level Set method. The forward problem is solved using an integral formulation of the EM problem with finite element discretization. The shape gradient is computed using two different methods: one is finite differential method based on nodal point mesh derivation with an infinitesimal modification of the triangular elements on the contour along the outward normal direction, another the topological shape gradient, which is computed based on a topological deformation on a contour. A narrow band level set method has been developed to evolve the contour of antennas and arrays using the deformation velocity computed from the shape gradient. Different configurations of antennas and antenna arrays are studied for investigating the performance of the optimization algorithm. Frequency hopping and multi-frequency techniques have been used for optimizing the shape within a frequency band. Shape optimization for planar antenna miniaturization has a large number of applications, particularly, for reflectarrays
Deschinkel, Karine. "Régulation du trafic aérien par optimisation dynamique des prix d'utilisation du réseau". École nationale supérieure de l'aéronautique et de l'espace (Toulouse ; 1972-2007), 2001. http://www.theses.fr/2001ESAE0009.
Texto completoLaurenceau, Julien. "SURFACES DE REPONSE PAR KRIGEAGE POUR L'OPTIMISATION DE FORMES AERODYNAMIQUES". Phd thesis, Institut National Polytechnique de Toulouse - INPT, 2008. http://tel.archives-ouvertes.fr/tel-00339863.
Texto completoUn nouvel optimiseur basé sur des surfaces de réponse construites par une méthode de Krigeage est proposé. A un surcout modéré, la solution obtenue est meilleure. De plus, cet optimiseur semble aussi capable de traiter des problèmes de grande dimension en interpolant le vecteur gradient aux points de construction du Krigeage.
En optimisation multidisciplinaire, les surfaces de réponse sont largement employées pour échanger facilement des données entre différentes disciplines. Ainsi, une approche d'optimisation bi-niveau avec couplage fluide/structure par surface de réponse est étudiée. L'application considérée traite de l'intégration d'une installation motrice (positionnement) sur un avion de transport civil.
Achard, Timothée. "Techniques de calcul de gradient aéro-structure haute-fidélité pour l'optimisation de voilures flexibles". Thesis, Paris, CNAM, 2017. http://www.theses.fr/2017CNAM1140/document.
Texto completoTo improve the structural design of flexible wings, gradient based Multidisciplinary Design Optimization (MDO) techniques are effective and widely used. However, gradients calculation is not trivial and can be costly when high-fidelity models are considered. Our objective is to study different suitable approaches to compute gradients of aeroelastic loads with respect to structural design parameters.To this end, two high-fidelity aero-structure gradient computation techniques for strongly coupled aeroelastic systems are proposed. The most intrusive technique includes the well-established direct and adjoint formulations that require substantial implementation effort. In contrast, we propose an alternative uncoupled non-intrusive approach easier to implement and yet capable of providing accurate gradients approximations. Both techniques have been implemented in the Onera elsA CFD software.Accuracy, efficiency and applicability of these methods are demonstrated on the civil transport aircraft Common Research Model (CRM) test-case. More specifically, an inverse design problem is set up with the objective of matching an in-flight target twist law distribution. These two methods prove to be comparable in terms of accuracy and cost. Thus they offer additional operational flexibility depending on the level of integration sought in the MDO process
Drullion, Frédérique. "Définition et étude de systèmes linéaires pour la simulation d'écoulements et l'optimisation de formes aérodynamiques par méthode de gradient". Bordeaux 1, 2004. http://www.theses.fr/2004BOR12898.
Texto completoL'Excellent, Jean-Yves. "Utilisation de préconditionneurs élément-par-élément pour la résolution de problèmes d'optimisation de grande taille". Toulouse, INPT, 1995. http://www.theses.fr/1995INPT091H.
Texto completoDo, Thien Tho. "Optimisation de forme en forgeage 3D". Paris, ENMP, 2006. http://www.theses.fr/2006ENMP1366.
Texto completoThis study focuses on shape optimization for 3D forging process. The problems to be solved consist in searching the optimal shape of the initial work piece or of the preform tool in order to minimize an objective function F which represents a measure of non-quality defined by the designer. These are often multi optima problems in which the necessary time for a cost function evaluation is very long (about a day or more). This work aims at developping an optimization module that permits to localize the global optimum within a reasonable cost (less than 50 calculations of objective function per optimization). The process simulation is carried out using the FORGE3® finite element software. The axisymmetric initial shape of the workpiece or die is parameterized using quadratic segments or Bspline curves. Several objective functions are considered, like the forging energy, the forging force or a surface defect criterion. The gradient of these objective functions is obtained by the adjoint-state method and semi-analytical differentiation. In this work, this gradient calculation (initiated in M. Laroussi's thesis) has been extended to another type of parameter "the parameters that control the shape of tool preform". Different optimization algorithms are tested for 3D applications: a standard BFGS algorithm, a moving asymptote algorithm, an evolution strategies algorithm enhanced with a response surface method based on Kriging and two new hybrid evolutionary algorithms proposed during this work. This hybrid approach consists in coupling a genetic algorithm to a response surface method that uses gradient information to dramatically reduce the number of problem simulations. All studied algorithms are compared for two 3D industrial tests, using rather coarse meshes. They make it possible to improve the initial design and to decrease the total forming energy and/or a surface defect criterion. Numerical results show the feasibility of such approaches, i. E. The achieving of satisfactory solutions within a limited number of 3D simulations, less than fifty
Chaffanjon, Pierre. "Optimisation de l'attenuation et de la dispersion des fibres optiques polymeres par l'utilisation de materiaux deuteries et par la realisation de preformes a gradient d'indice". Université Louis Pasteur (Strasbourg) (1971-2008), 1990. http://www.theses.fr/1990STR13035.
Texto completoYuan, Rui. "Stochastic Second Order Methods and Finite Time Analysis of Policy Gradient Methods". Electronic Thesis or Diss., Institut polytechnique de Paris, 2023. http://www.theses.fr/2023IPPAT010.
Texto completoTo solve large scale machine learning problems, first-order methods such as stochastic gradient descent and ADAM are the methods of choice because of their low cost per iteration. The issue with first order methods is that they can require extensive parameter tuning, and/or knowledge of the parameters of the problem. There is now a concerted effort to develop efficient stochastic second order methods to solve large scale machine learning problems. The motivation is that they require less parameter tuning and converge for wider variety of models and datasets. In the first part of the thesis, we presented a principled approach for designing stochastic Newton methods for solving both nonlinear equations and optimization problems in an efficient manner. Our approach has two steps. First, we can re-write the nonlinear equations or the optimization problem as desired nonlinear equations. Second, we apply new stochastic second order methods to solve this system of nonlinear equations. Through our general approach, we showcase many specific new second-order algorithms that can solve the large machine learning problems efficiently without requiring knowledge of the problem nor parameter tuning. In the second part of the thesis, we then focus on optimization algorithms applied in a specific domain: reinforcement learning (RL). This part is independent to the first part of the thesis. To achieve such high performance of RL problems, policy gradient (PG) and its variant, natural policy gradient (NPG), are the foundations of the several state of the art algorithms (e.g., TRPO and PPO) used in deep RL. In spite of the empirical success of RL and PG methods, a solid theoretical understanding of even the “vanilla” PG has long been elusive. By leveraging the RL structure of the problem together with modern optimization proof techniques, we derive new finite time analysis of both PG and NPG. Through our analysis, we also bring new insights to the methods with better hyperparameter choices
Si, Fodil David Mohand. "Commande floue et optimisation de base de règles floues pour la régulation de la réactivité et de la température moyenne dans les réacteurs nucléaires à eau pressurisée". Châtenay-Malabry, Ecole centrale de Paris, 2000. http://www.theses.fr/2000ECAP0865.
Texto completoSubrin, Kévin. "Optimisation du comportement de cellules robotiques par gestion des redondances : application à la découpe de viande et à l’Usinage Grande Vitesse". Thesis, Clermont-Ferrand 2, 2013. http://www.theses.fr/2013CLF22417/document.
Texto completoIndustrial robots have evolved fundamentally in recent years to reach the industrial requirements. We now find more suitable anthropomorphic robots leading to the realization of more complex tasks like deformable objects cutting such as meat cutting or constrained to high stresses as machining. The behavior study of anthropomorphic robots, parallel or hybrid one highlights a kinematic and dynamic anisotropy, which impacts the expected accuracy. This thesis studied the integration of the kinematic redundancy that can partially overcome this problem by well setting the task to achieve it in a space compatible with the expected capacity. This work followed a three-step approach: analytical modeling of robotic cells by serial equivalent based on the TCS method, formalizing the constraints of meat cutting process and machining process and a multicriteria optimization.The first originality of this work focuses on the development of a 6 DoFs model to analyze the operator actions who naturally optimizes his arm behavior to ensure the task it performs. The second originality concerns the optimized placement of structural redundancy (9 DoFs robotic cell) where positioning parameters are incorporated as controllable variables (11 DoFs robotic cell). Thus, the thesis makes contributions to : - the definition of criteria adapted to the realization of complex and under high stress task for the management of the kinematic redundancy; - the structural behavior identification, under stress, by metrology tools (Laser tracker ) and the self- adaptation paths by using an industrial force control; - the behavior optimization to improve the cutting process quality (meat cutting and machining)
DRIANCOURT, XAVIER. "Optimisation par descente de gradient stochastique de systemes modulaires combinant reseaux de neurones et programmation dynamique. Application a la reconnaissance de la parole". Paris 11, 1994. http://www.theses.fr/1994PA112203.
Texto completoSaouab, Abdelghani. "Génération de maillages adaptatifs par une méthode variationnelle". Rouen, 1991. http://www.theses.fr/1991ROUES013.
Texto completoDécamps, Jérôme. "Méthodes itératives par blocs pour la résolution de problèmes linéaires et non linéaires à structures partiellement séparables". Toulouse, INPT, 1997. http://www.theses.fr/1997INPT092H.
Texto completoAssadi-Haghi, Atousa. "Contribution au développement de méthodes d'optimisation structurelle pour la conception assistée par ordinateur de composants et de circuits hyperfréquences". Limoges, 2007. https://aurore.unilim.fr/theses/nxfile/default/1a189347-19d7-4422-99d6-9019a30e6b99/blobholder:0/2007LIMO4050.pdf.
Texto completoThe thesis manuscript reports on the study of structural optimization methods for computer aided design of microwave devices. In the first part, a geometrical optimization approach is developed and applied to the design of a packaged circuit. The approach is based on model order reduction using segmentation and geometrical parameterization of the electromagnetic model. The reduced model is optimized through a gradient method, minimizing a cost function dedicated to identification of parasitic modes in the package. In the second part, a topological optimization approach, based on topology gradient evaluation, is applied for optimizing metal distribution upon the surface of a microstrip component. For solving local optimum problems, the method is hybridized with a genetic algorithm for exploring more largely the optimization domain, improving the convergence by this way
Dosne, Cyril. "Development and implementation of adjoint formulation of explicit body-force models for aero-propulsive optimizations". Electronic Thesis or Diss., Institut polytechnique de Paris, 2024. http://www.theses.fr/2024IPPAX026.
Texto completoIn civil aviation, the increasing exploration of innovative engine systems – such as ultra-high bypass ratio turbofan or open-rotor – and breakthrough engine-integration architectures – such as distributed propulsion or boundary-layer ingestion – require a coupled modeling of the aerodynamic and propulsion subsystems from the earliest design stages. Body-force models have proven capable of faithfully reproducing most of the coupling phenomena, such as the engine response to inlet flow distortions, at reduced computational cost. However, they lack an adjoint formulation to be efficiently used in gradient-based optimizations. The present PhD thesis focuses on the development of an adjoint approach for explicit body-force models. First, aero-propulsive optimizations of an academic distributed propulsion configuration are conducted using a lumped body-force model. Despite the simplicity of this model (of interest for conceptual design studies), 10.5% decrease in power consumption is achieved. Then the potential of this new methodology is investigated for the preliminary optimization of compressor stages, at first under clean inflow conditions. The Hall body-force model is considered for such purpose. The comparison of the blade shape gradients computed by the adjoint body-force with high-fidelity ones, obtained from blade-resolved computations, shows very good prediction for the rotor. This is observed over a large portion of the compressor characteristic, especially between near-design and surge operating conditions, while accuracy is reduced near the blockage. On the contrary, for stator shape gradients, only flow misalignment effects can be captured. At design conditions, the improvement of the compressor efficiency obtained by the adjoint body-force optimization has been confirmed through high-fidelity simulations. Optimization under radial inlet distortion are then investigated. Once again, the adjoint body-force approach is found capable of enhancing the compressor performances, by adapting its geometry to the off-design inflow conditions. According to high-fidelity analysis of the body-force optimized blade geometry, an increase in compressor isentropic efficiency between 1.16 and 1.47% is achieved, given the formulation of the optimization problem. Finally, an optimization of the compressor under full-annulus inlet distortion is conducted leading to very promising results, which are consistent with those found in the literature using advanced simulations
Godoy, Campbell Matias. "Sur le problème inverse de détection d'obstacles par des méthodes d'optimisation". Thesis, Toulouse 3, 2016. http://www.theses.fr/2016TOU30220/document.
Texto completoThis PhD thesis is dedicated to the study of the inverse problem of obstacle/object detection using optimization methods. This problem consists in localizing an unknown object omega inside a known bounded domain omega by means of boundary measurements and more precisely by a given Cauchy pair on a part Gammaobs of thetaOmega. We cover the scalar and vector scenarios for this problem considering both the Laplace and the Stokes equations. For both cases, we rely on identifiability result which ensures that there is a unique obstacle/object which corresponds to the considered boundary measurements. The strategy used in this work is to reduce the inverse problem into the minimization of a cost-type functional: the Kohn-Vogelius functional. This kind of approach is widely used and permits to use optimization tools for numerical implementations. However, in order to well-define the functional, this approach needs to assume the knowledge of a measurement on the whole exterior boundary thetaOmega. This last point leads us to first study the data completion problem which consists in recovering the boundary conditions on an inaccessible region, i.e. on thetaOmega\Gammaobs, from the Cauchy data on the accessible region Gammaobs. This inverse problem is also studied through the minimization of a Kohn-Vogelius type functional. The ill-posedness of this problem enforces us to regularize the functional via a Tikhonov regularization. We obtain several theoretical properties as convergence properties, in particular when data is corrupted by noise. Based on these theoretical results, we reconstruct numerically the boundary data by implementing a gradient algorithm in order to minimize the regularized functional. Then we study the obstacle detection problem when only partial boundary measurements are available. We consider the inaccessible boundary conditions and the unknown object as the variables of the functional and then, using geometrical shape optimization tools, in particular the shape gradient of the Kohn-Vogelius functional, we perform the numerical reconstruction of the unknown inclusion. Finally, we consider, into the two dimensional vector case, a new degree of freedom by studying the case when the number of objects is unknown. Hence, we use the topological shape optimization in order to minimize the Kohn-Vogelius functional. We obtain the topological asymptotic expansion of the solution of the 2D Stokes equations and characterize the topological gradient for this functional. Then we determine numerically the number and location of the obstacles. Additionally, we propose a blending algorithm which combines the topological and geometrical shape optimization methods in order to determine numerically the number, location and shape of the objects
Subrin, Kévin. "Optimisation du comportement de cellules robotiques par gestion des redondances : application à la découpe de viande et à l'Usinage Grande Vitesse". Phd thesis, Université Blaise Pascal - Clermont-Ferrand II, 2013. http://tel.archives-ouvertes.fr/tel-00999471.
Texto completoLaborde, Maxime. "Systèmes de particules en interaction, approche par flot de gradient dans l'espace de Wasserstein". Thesis, Paris Sciences et Lettres (ComUE), 2016. http://www.theses.fr/2016PSLED014/document.
Texto completoSince 1998 and the seminal work of Jordan, Kinderlehrer and Otto, it is well known that a large class of parabolic equations can be seen as gradient flows in the Wasserstein space. This thesis is devoted to extensions of this theory to equations and systems which do not have exactly a gradient flow structure. We study different kind of couplings. First, we treat the case of nonlocal interactions in the drift. Then, we study cross diffusion systems which model congestion for several species. We are also interested in reaction-diffusion systems as diffusive prey-predator systems or tumor growth models. Finally, we introduce a new class of systems where the interaction is given by a multi-marginal transport problem. In many cases, we give numerical simulations to illustrate our theorical results
Jolibois, Alexandre. "A study on the acoustic performance of tramway low height noise barriers: gradient-based numerical optimization and experimental approaches ( Étude de la performance acoustique des écrans antibruit de faible hauteur pour le tramway : optimisation numérique par méthode de gradient et approches expérimentales)". Thesis, Paris Est, 2013. http://www.theses.fr/2013PEST1116/document.
Texto completoNoise has become a main nuisance in urban areas to the point that according to the World Health Organization 40% of the European population is exposed to excessive noise levels, mainly due to ground transportation. There is therefore a need to find new ways to mitigate noise in urban areas. In this work, a possible device to achieve this goal is studied: a low-height noise barrier. It consists of a barrier typically less than one meter high placed close to a source, designed to decrease the noise level for nearby pedestrians and cyclists. This type of device is studied both numerically and experimentally. Tramway noise barriers are especially studied since the noise sources are in this case very close to the ground and can therefore be attenuated efficiently. The shape and the surface treatment of the barrier are optimized using a gradient-based method coupled to a 2D boundary element method (BEM). The optimization variables are the node coordinates of a control mesh and the parameters describing the surface impedance. Sensitivities are calculated efficiently using the adjoint state approach. Numerical results show that the shapes generated by the optimization algorithm tend to be quite irregular but provide a significant improvement of more than 5 dB (A) compared to simpler shapes. Utilizing an absorbing treatment on the source side of the barrier is shown to be efficient as well. This second point has been confirmed by scale model measurements. In addition, a full scale low height noise barrier prototype has been built and tested in situ close to a tramway track in Grenoble. Measurements show that the device provides more than 10 dB (A) of attenuation for a close receiver located at the typical height of human ears. These results therefore seem to confirm the applicability of such protections to efficiently decrease noise exposure in urban areas
Cousturier, Richard. "Amélioration par la gestion de redondance du comportement des robots à structure hybride sous sollicitations d’usinage". Thesis, Université Clermont Auvergne (2017-2020), 2017. http://www.theses.fr/2017CLFAC090/document.
Texto completoIndustrial robots have evolved fundamentally in recent years to reach the industrial requirements. We now find more suitable anthropomorphic robots leading to the realization of more complex tasks like deformable objects cutting such as meat cutting or constrained to high loading like during machining. The behavior study of anthropomorphic robots, parallel or hybrid one highlights a kinematic and dynamic anisotropy, which impacts the expected accuracy.This thesis studied the integration of the kinematic redundancy that can partially overcome this problem by well setting the task to achieve it in a space compatible with the expected capacity.This work helped us to improve our optimization tool and to try it on both FE model of the robot and real robot.Thus, the thesis makes contributions to: - the definition of criteria adapted to the realization of complex and under high loading task for the management of the kinematic redundancy; - the structural behavior identification, under loading, by metrology tools (Laser tracker) ; - the behavior optimization to improve the cutting process quality during machining ; - robots finite elements modeling using stiffness identification for both bodies and joints
Ouriemchi, Mohammed. "Résolution de problèmes non linéaires par les méthodes de points intérieurs : théorie et algorithmes". Phd thesis, Université du Havre, 2005. http://tel.archives-ouvertes.fr/tel-00011376.
Texto completoDans cette thèse, nous avons utilisé une fonction barrière logarithmique. A chaque itération externe, la technique SQP se charge de produire une série de sous-problèmes quadratiques dont les solutions forment une suite, dite interne, de directions de descente pour résoudre le problème non linéaire pénalisé.
Nous avons introduit un changement de variable sur le pas de déplacement ce qui a permis d'obtenir des conditions d'optimalité plus stable numériquement.
Nous avons réalisé des simulations numériques pour comparer les performances de la méthode des gradients conjugués à celle de la méthode D.C., appliquées pour résoudre des problèmes quadratiques de région de confiance.
Nous avons adapté la méthode D.C. pour résoudre les sous-problèmes verticaux, ce qui nous a permis de ramener leurs dimensions de $n+m$ à $m+p$ ($ p < n $).
L'évolution de l'algorithme est contrôlée par la fonction de mérite. Des tests numériques permettent de comparer les avantages de différentes formes de la fonction de mérite. Nous avons introduit de nouvelles règles pour améliorer cette évolution.
Les expériences numériques montrent un gain concernant le nombre de problèmes résolus. L'étude de la convergence de notre méthode SDC, clôt ce travail.
Weymann, Jacques. "Commande du trafic par guidage des véhicules avec prise en compte du comportement humain et de la saturation". Toulouse, ENSAE, 1994. http://www.theses.fr/1994ESAE0010.
Texto completoPouilly-Cathelain, Maxime. "Synthèse de correcteurs s’adaptant à des critères multiples de haut niveau par la commande prédictive et les réseaux de neurones". Electronic Thesis or Diss., université Paris-Saclay, 2020. http://www.theses.fr/2020UPASG019.
Texto completoThis PhD thesis deals with the control of nonlinear systems subject to nondifferentiable or nonconvex constraints. The objective is to design a control law considering any type of constraints that can be online evaluated.To achieve this goal, model predictive control has been used in addition to barrier functions included in the cost function. A gradient-free optimization algorithm has been used to solve this optimization problem. Besides, a cost function formulation has been proposed to ensure stability and robustness against disturbances for linear systems. The proof of stability is based on invariant sets and the Lyapunov theory.In the case of nonlinear systems, dynamic neural networks have been used as a predictor for model predictive control. Machine learning algorithms and the nonlinear observers required for the use of neural networks have been studied. Finally, our study has focused on improving neural network prediction in the presence of disturbances.The synthesis method presented in this work has been applied to obstacle avoidance by an autonomous vehicle
Barakat, Anas. "Contributions to non-convex stochastic optimization and reinforcement learning". Electronic Thesis or Diss., Institut polytechnique de Paris, 2021. http://www.theses.fr/2021IPPAT030.
Texto completoThis thesis is focused on the convergence analysis of some popular stochastic approximation methods in use in the machine learning community with applications to optimization and reinforcement learning.The first part of the thesis is devoted to a popular algorithm in deep learning called ADAM used for training neural networks. This variant of stochastic gradient descent is more generally useful for finding a local minimizer of a function. Assuming that the objective function is differentiable and non-convex, we establish the convergence of the iterates in the long run to the set of critical points under a stability condition in the constant stepsize regime. Then, we introduce a novel decreasing stepsize version of ADAM. Under mild assumptions, it is shown that the iterates are almost surely bounded and converge almost surely to critical points of the objective function. Finally, we analyze the fluctuations of the algorithm by means of a conditional central limit theorem.In the second part of the thesis, in the vanishing stepsizes regime, we generalize our convergence and fluctuations results to a stochastic optimization procedure unifying several variants of the stochastic gradient descent such as, among others, the stochastic heavy ball method, the Stochastic Nesterov Accelerated Gradient algorithm, and the widely used ADAM algorithm. We conclude this second part by an avoidance of traps result establishing the non-convergence of the general algorithm to undesired critical points, such as local maxima or saddle points. Here, the main ingredient is a new avoidance of traps result for non-autonomous settings, which is of independent interest.Finally, the last part of this thesis which is independent from the two previous parts, is concerned with the analysis of a stochastic approximation algorithm for reinforcement learning. In this last part, we propose an analysis of an online target-based actor-critic algorithm with linear function approximation in the discounted reward setting. Our algorithm uses three different timescales: one for the actor and two for the critic. Instead of using the standard single timescale temporal difference (TD) learning algorithm as a critic, we use a two timescales target-based version of TD learning closely inspired from practical actor-critic algorithms implementing target networks. First, we establish asymptotic convergence results for both the critic and the actor under Markovian sampling. Then, we provide a finite-time analysis showing the impact of incorporating a target network into actor-critic methods
Pham, Chi-Tuân. "Linéarisation du flux visqueux des équations de navier-stokes et de modèles de turbulence pour l'optimisation aérodynamique en turbomachines". Phd thesis, Paris, ENSAM, 2006. http://pastel.archives-ouvertes.fr/pastel-00002058.
Texto completoPham, Chi-Tuân. "Linéarisation du flux visqueux des équations de navier-stokes et de modèles de turbulence pour l'optimisation aérodynamique en turbomachines". Phd thesis, Paris, ENSAM, 2006. http://www.theses.fr/2006ENAM0030.
Texto completoThe computation of derivatives of aerodynamic functions, with respect to design parameters of the solid shape is now a branch of computational fluid dynamics. This differentiation with respect to two dependent variables, the flow field and the mesh, bound by the discrete fluid dynamics equations, needs the resolution of a linear system. Its matrix is the Jacobian matrix of the fluid dynamics equations with respect to the flow field (direct differentiation method) or the tranposate of this Jacobian matrix (adjoint method). The accuracy of the computation of this Jacobian matrix is discussed when the RANS equations are used. The aim of this PhD thesis is to determine the level of accuracy for the linearization of a discrete viscous flux and the discrete equations of some turbulence models, needed to reach accurate gradients of functions used by the conceptors of turbomachineries, with respect to some design parameters of a blade. Approximate viscous flux linearizations (with or without a thin layer approach) and the linearization of two turbulence models (algebraic Michel et al. Model and Launder-Sharma k-e two-equation model) are described. Several approximations for the linearization of Michel et al. 's model are tested and compared. Results (values of gradients of aerodynamic functions, flow sensibilities for the direct differentiation method) are shown for Délery's C nozzle, ONERA M6 wing and two turbine isolated blades. Recommendations for the computation of derivatives are given for turbomachinery flows with RANS equations
Demirel, Mustafa. "Contribution à l'optimisation des mesures de température et de déformations par capteur à fibre optique à réseau de Bragg : application au procédé de fabrication des composites par infusion de résine". Phd thesis, Ecole Nationale Supérieure des Mines de Saint-Etienne, 2009. http://tel.archives-ouvertes.fr/tel-00440938.
Texto completoLabat, Christian. "Algorithmes d'optimisation de critères pénalisés pour la restauration d'images : application à la déconvolution de trains d'impulsions en imagerie ultrasonore". Phd thesis, Ecole centrale de nantes - ECN, 2006. http://tel.archives-ouvertes.fr/tel-00132861.
Texto completo- Démontrer la convergence des algorithmes SQ approchés et GCNL+SQ1D.
- Etablir des liens forts entre les algorithmes SQ approchés et GCNL+SQ1D.
- Illustrer expérimentalement en déconvolution d'images le fait que les algorithmes SQ approchés et GCNL+SQ1D sont préférables aux algorithmes SQ exacts.
- Appliquer l'approche pénalisée à un problème de déconvolution d'images en contrôle non destructif par ultrasons.
Flammarion, Nicolas. "Stochastic approximation and least-squares regression, with applications to machine learning". Thesis, Paris Sciences et Lettres (ComUE), 2017. http://www.theses.fr/2017PSLEE056/document.
Texto completoMany problems in machine learning are naturally cast as the minimization of a smooth function defined on a Euclidean space. For supervised learning, this includes least-squares regression and logistic regression. While small problems are efficiently solved by classical optimization algorithms, large-scale problems are typically solved with first-order techniques based on gradient descent. In this manuscript, we consider the particular case of the quadratic loss. In the first part, we are interestedin its minimization when its gradients are only accessible through a stochastic oracle. In the second part, we consider two applications of the quadratic loss in machine learning: clustering and estimation with shape constraints. In the first main contribution, we provided a unified framework for optimizing non-strongly convex quadratic functions, which encompasses accelerated gradient descent and averaged gradient descent. This new framework suggests an alternative algorithm that exhibits the positive behavior of both averaging and acceleration. The second main contribution aims at obtaining the optimal prediction error rates for least-squares regression, both in terms of dependence on the noise of the problem and of forgetting the initial conditions. Our new algorithm rests upon averaged accelerated gradient descent. The third main contribution deals with minimization of composite objective functions composed of the expectation of quadratic functions and a convex function. Weextend earlier results on least-squares regression to any regularizer and any geometry represented by a Bregman divergence. As a fourth contribution, we consider the the discriminative clustering framework. We propose its first theoretical analysis, a novel sparse extension, a natural extension for the multi-label scenario and an efficient iterative algorithm with better running-time complexity than existing methods. The fifth main contribution deals with the seriation problem. We propose a statistical approach to this problem where the matrix is observed with noise and study the corresponding minimax rate of estimation. We also suggest a computationally efficient estimator whose performance is studied both theoretically and experimentally
Mussard, Bastien. "Modélisation quantochimiques des forces de dispersion de London par la méthode des phases aléatoires (RPA) : développements méthodologiques". Thesis, Université de Lorraine, 2013. http://www.theses.fr/2013LORR0292/document.
Texto completoIn this thesis are shown developments in the random phase approximation (RPA) in the context of range-separated theories. We present advances in the formalism of the RPA in general, and particularly in the "dielectric matrix" formulation of RPA, which is explored in details. We show a summary of a work on the RPA equations with localized orbitals, especially developments of the virtual localized orbitals that are the "projected oscillatory orbitals" (POO). A program has been written to calculate functions such as the exchange hole, the response function, etc... on real space grid (parallelepipedic or of the "DFT" type) ; some of those visualizations are shown here. In the real space, we offer an adaptation of the effective energy denominator approximation (EED), originally developed in the reciprocal space in solid physics. The analytical gradients of the RPA correlation energies in the context of range separation has been derived. The formalism developed here with a Lagrangian allows an all-in-one derivation of the short- and long-range terms that emerge in the expressions of the gradient. These terms show interesting parallels. Geometry optimizations at the RSH-dRPA-I and RSH-SOSEX levels on a set of 16 molecules are shown, as well as calculations and visualizations of correlated densities at the RSH-dRPA-I level
Ailleres, Norbert. "Etude comparative des séquences d'échos de spins, d'échos stimulés et d'échos de gradients en I. R. M : optimisation du contraste". Toulouse 3, 1992. http://www.theses.fr/1992TOU30254.
Texto completoHaberer, Isabelle. "Photoinscription de gradients d'indice dans des hydrogels par formation de réseaux interpénétrés : optimisation du procédé, caractérisation des produits et application à l'optique de contact". Mulhouse, 1994. http://www.theses.fr/1994MULH0343.
Texto completoTakouda, Pawoumodom Ledogada. "Problèmes d'approximation matricielle linéaires coniques : approches par projections et via optimisation sous contraintes de semidéfinie positivité". Toulouse 3, 2003. http://www.theses.fr/2003TOU30129.
Texto completoPham, Tan Hung. "Problèmes couplés inverses : optimisation des systèmes couplés magnéto-thermiques avec la méthode des éléments finis et les algorithmes numériques d’optimisation de type gradient". Grenoble INPG, 1997. http://www.theses.fr/1997INPG0151.
Texto completoCheng, Jianqiang. "Stochastic Combinatorial Optimization". Thesis, Paris 11, 2013. http://www.theses.fr/2013PA112261.
Texto completoIn this thesis, we studied three types of stochastic problems: chance constrained problems, distributionally robust problems as well as the simple recourse problems. For the stochastic programming problems, there are two main difficulties. One is that feasible sets of stochastic problems is not convex in general. The other main challenge arises from the need to calculate conditional expectation or probability both of which are involving multi-dimensional integrations. Due to the two major difficulties, for all three studied problems, we solved them with approximation approaches.We first study two types of chance constrained problems: linear program with joint chance constraints problem (LPPC) as well as maximum probability problem (MPP). For both problems, we assume that the random matrix is normally distributed and its vector rows are independent. We first dealt with LPPC which is generally not convex. We approximate it with two second-order cone programming (SOCP) problems. Furthermore under mild conditions, the optimal values of the two SOCP problems are a lower and upper bounds of the original problem respectively. For the second problem, we studied a variant of stochastic resource constrained shortest path problem (called SRCSP for short), which is to maximize probability of resource constraints. To solve the problem, we proposed to use a branch-and-bound framework to come up with the optimal solution. As its corresponding linear relaxation is generally not convex, we give a convex approximation. Finally, numerical tests on the random instances were conducted for both problems. With respect to LPPC, the numerical results showed that the approach we proposed outperforms Bonferroni and Jagannathan approximations. While for the MPP, the numerical results on generated instances substantiated that the convex approximation outperforms the individual approximation method.Then we study a distributionally robust stochastic quadratic knapsack problems, where we only know part of information about the random variables, such as its first and second moments. We proved that the single knapsack problem (SKP) is a semedefinite problem (SDP) after applying the SDP relaxation scheme to the binary constraints. Despite the fact that it is not the case for the multidimensional knapsack problem (MKP), two good approximations of the relaxed version of the problem are provided which obtain upper and lower bounds that appear numerically close to each other for a range of problem instances. Our numerical experiments also indicated that our proposed lower bounding approximation outperforms the approximations that are based on Bonferroni's inequality and the work by Zymler et al.. Besides, an extensive set of experiments were conducted to illustrate how the conservativeness of the robust solutions does pay off in terms of ensuring the chance constraint is satisfied (or nearly satisfied) under a wide range of distribution fluctuations. Moreover, our approach can be applied to a large number of stochastic optimization problems with binary variables.Finally, a stochastic version of the shortest path problem is studied. We proved that in some cases the stochastic shortest path problem can be greatly simplified by reformulating it as the classic shortest path problem, which can be solved in polynomial time. To solve the general problem, we proposed to use a branch-and-bound framework to search the set of feasible paths. Lower bounds are obtained by solving the corresponding linear relaxation which in turn is done using a Stochastic Projected Gradient algorithm involving an active set method. Meanwhile, numerical examples were conducted to illustrate the effectiveness of the obtained algorithm. Concerning the resolution of the continuous relaxation, our Stochastic Projected Gradient algorithm clearly outperforms Matlab optimization toolbox on large graphs
TAKOUDA, Pawoumodom Ledogada. "Problèmes d'approximation matricielle linéaires coniques: Approches par Projections et via Optimisation sous contraintes de semi-définie positivité". Phd thesis, Université Paul Sabatier - Toulouse III, 2003. http://tel.archives-ouvertes.fr/tel-00005469.
Texto completoArlery, Fabien. "Formes d’ondes MSPSR, traitements et performances associés". Thesis, Evry, Institut national des télécommunications, 2017. http://www.theses.fr/2017TELE0005/document.
Texto completoNowadays, MSPSR (Multi-Static Primary Surveillance Radar) systems are sustainably settled in air surveillance program [1]. Compared to mono-static radar currently in use, an MSPSR system is based on a sparse network of transmitters (Tx) and receivers (Rx) interconnected to a Central Unit and offers advantages in terms of reliability, cost and performance.Two kinds of MSPSR systems exist: the Passive form and the Active one. While the Passive MSPSR uses transmitters of opportunity such as radio Frequency Modulation (FM) transmitters and/or Digital Video Broadcasting-Terrestrial (DVB-T) transmitters [2], the Active MSPSR uses dedicated transmitters, which emit a waveform that is controlled and designed for a radar application. Each receiver processes the signal coming from all transmitters and reflected on the targets; and the Central Unit restores the target location by intersecting “ellipsoids” from all (transmitter, receiver) pairs. Compared to passive MSPSR, the main advantages of the active MSPSR are the use of dedicated waveforms that allow reaching better performances (like a better association of the transmitters’ contributions at the receiver level); more flexibility in the deployment of transmitters and receivers station (in order to meet the requirements in localisation accuracy and in horizontal and altitude coverages); and the guarantee of having a service continuity. On this purpose, this thesis analyses the differents codes criteria such as the ambiguity function behaviour, the PAPR (Peak to Average Power Ratio), the spectrum efficiency, etc... . Then, in order to find dedicated waveforms for MSPSR systems, one solution is to find easily-constructed families of sequences. Thus building on the works carried out by the Telecommunication field for solving multi-user issues, this document investigates the application of spreading codes and OFDM signals in MSPSR concept. Besides, another solution is to directly generate a set of sequences. Based on cyclic algorithms in [3] we derive a new algorithm that allows to optimize sets of sequences. Similarly, using a gradient descent approach, we develop a more efficient algorithm than the cyclic one. Finally, in order to evaluate the performances of the different algorithms, this thesis generalizes the Levenshtein Bound, establishes new lower bounds on the PSLR (Peak Sidelobe Level Ratio) in mismatched filter case, and studies real data recorded during some trials
Arlery, Fabien. "Formes d’ondes MSPSR, traitements et performances associés". Electronic Thesis or Diss., Evry, Institut national des télécommunications, 2017. http://www.theses.fr/2017TELE0005.
Texto completoNowadays, MSPSR (Multi-Static Primary Surveillance Radar) systems are sustainably settled in air surveillance program [1]. Compared to mono-static radar currently in use, an MSPSR system is based on a sparse network of transmitters (Tx) and receivers (Rx) interconnected to a Central Unit and offers advantages in terms of reliability, cost and performance.Two kinds of MSPSR systems exist: the Passive form and the Active one. While the Passive MSPSR uses transmitters of opportunity such as radio Frequency Modulation (FM) transmitters and/or Digital Video Broadcasting-Terrestrial (DVB-T) transmitters [2], the Active MSPSR uses dedicated transmitters, which emit a waveform that is controlled and designed for a radar application. Each receiver processes the signal coming from all transmitters and reflected on the targets; and the Central Unit restores the target location by intersecting “ellipsoids” from all (transmitter, receiver) pairs. Compared to passive MSPSR, the main advantages of the active MSPSR are the use of dedicated waveforms that allow reaching better performances (like a better association of the transmitters’ contributions at the receiver level); more flexibility in the deployment of transmitters and receivers station (in order to meet the requirements in localisation accuracy and in horizontal and altitude coverages); and the guarantee of having a service continuity. On this purpose, this thesis analyses the differents codes criteria such as the ambiguity function behaviour, the PAPR (Peak to Average Power Ratio), the spectrum efficiency, etc... . Then, in order to find dedicated waveforms for MSPSR systems, one solution is to find easily-constructed families of sequences. Thus building on the works carried out by the Telecommunication field for solving multi-user issues, this document investigates the application of spreading codes and OFDM signals in MSPSR concept. Besides, another solution is to directly generate a set of sequences. Based on cyclic algorithms in [3] we derive a new algorithm that allows to optimize sets of sequences. Similarly, using a gradient descent approach, we develop a more efficient algorithm than the cyclic one. Finally, in order to evaluate the performances of the different algorithms, this thesis generalizes the Levenshtein Bound, establishes new lower bounds on the PSLR (Peak Sidelobe Level Ratio) in mismatched filter case, and studies real data recorded during some trials
Cua, Charles. "Amélioration de maillages par des méthodes de sous-gradient". Phd thesis, 1985. http://tel.archives-ouvertes.fr/tel-00318480.
Texto completoLamarre, Aldo. "Apprentissage de circuits quantiques par descente de gradient classique". Thesis, 2020. http://hdl.handle.net/1866/24322.
Texto completoWe present a new learning algorithm for quantum circuits based on gradient descent. Since this subject unifies two areas of research, we explain each field for people working in the other domain. Consequently, we begin by introducing quantum computing and quantum circuits to machine learning specialists, followed by an introduction of machine learning to quantum computing specialists. To give context and motivate our results we then give a light literature review on quantum machine learning. After this, we present our model, its algorithms and its variants, then discuss our currently achieved empirical results. Finally, we criticize our models by giving extensions and future work directions. These last two parts are our main results. They can be found in chapter 4 and 5 respectively. Our code which helped obtain these results can be found on github at this link : https://github.com/ AldoLamarre/quantumcircuitlearning.
Gendron, Pierre-Olivier. "Diffusion dans un hydrogel : applications aux biocapteurs et optimisation de la technique de spectroscopie par corrélation de fluorescence (FCS)". Thèse, 2008. http://hdl.handle.net/1866/7827.
Texto completoLaurent, César. "Advances in parameterisation, optimisation and pruning of neural networks". Thesis, 2020. http://hdl.handle.net/1866/25592.
Texto completoNeural networks are a family of Machine Learning models able to learn complex tasks directly from the data. Although already producing impressive results in many areas such as speech recognition, computer vision or machine translation, there are still a lot of challenges in both training and deployment of neural networks. In particular, training neural networks typically requires huge amounts of computational resources, and trained models are often too big or too computationally expensive to be deployed on resource-limited devices, such as smartphones or low-power chips. The articles presented in this thesis investigate solutions to these different issues. The first couple of articles focus on improving the training of Recurrent Neural Networks (RNNs), networks specially designed to process sequential data. RNNs are notoriously hard to train, so we propose to improve their parameterisation by upgrading them with Batch Normalisation (BN), a very effective parameterisation which was hitherto used only in feed-forward networks. In the first article, we apply BN to the input-to-hidden connections of the RNNs, thereby reducing internal covariate shift between layers. In the second article, we show how to apply it to both input-to-hidden and hidden-to-hidden connections of the Long Short-Term Memory (LSTM), a popular RNN architecture, thus also reducing internal covariate shift between time steps. Our experiments show that these proposed parameterisations allow for faster and better training of RNNs on several benchmarks. In the third article, we propose a new optimiser to accelerate the training of neural networks. Traditional diagonal optimisers, such as RMSProp, operate in parameters coordinates, which is not optimal when several parameters are updated at the same time. Instead, we propose to apply such optimisers in a basis in which the diagonal approximation is likely to be more effective. We leverage the same approximation used in Kronecker-factored Approximate Curvature (K-FAC) to efficiently build this Kronecker-factored Eigenbasis (KFE). Our experiments show improvements over K-FAC in training speed for several deep network architectures. The last article focuses on network pruning, the action of removing parameters from the network, in order to reduce its memory footprint and computational cost. Typical pruning methods rely on first or second order Taylor approximations of the loss landscape to identify which parameters can be discarded. We propose to study the impact of the assumptions behind such approximations. Moreover, we systematically compare methods based on first and second order approximations with Magnitude Pruning (MP), showing how they perform both before and after a fine-tuning phase. Our experiments show that better preserving the original network function does not necessarily transfer to better performing networks after fine-tuning, suggesting that only considering the impact of pruning on the loss might not be a sufficient objective to design good pruning criteria.