Apesar da Análise de Componentes Principais (PCA) ser um dos métodos mais importantes na análise da estrutura a termo de taxa de juros, há fortes indícios de não ser adequada para estimar fatores da curva de juros quando há presença de dependência temporal e erros de medida. Para corrigir esses problemas é indicado utilizar a matriz de covariância de longo prazo, extraindo a correta estrutura de covariância presente nestes processos. Neste trabalho, mostramos que realizar a previsão fora da amostra da curva de taxa de juros com o método de Análise de Componentes Principais (PCA) utilizando como base a matriz de covarância de longo prazo (LRCM) parece ser mais acurada comparada a PCA com base na matriz de covariância amostral. / Although Principal Component Analysis (PCA) is one of the most common methods to estimate the structure of interest rate volatility, there are strong indications that it is not adequate to estimate interest rate factors when there is temporal dependence and measurement errors. To correct these problems it is necessary to use the longterm covariance matrix, to extract the correct covariance structure present in these processes. In this work, we show that performing the out-of-sample forecasting of the interest rate curve with the Principal Component Analysis (PCA) method based on the long-term covariance matrix (LRCM) seems to be more accurate compared to PCA based on sample covariance matrix.

Dans les grands réseaux de transport, certains éléments du réseau peuvent être responsables du traitement d’importants volumes de trafic. Cela rend ces réseaux vulnérables aux pannes telles que les coupures de câbles. Des mécanismes appropriés pour le recouvrement du trafic doivent être mis en oeuvre pour éviter les ruptures de service. Une des meilleures techniques pour protéger les réseaux de transport consiste à prévoir des mécanismes de restauration au niveau de la couche transport elle-même afin que chaque opérateur de transport puisse sécuriser son propre réseau et offrir un service de transport fiable aux autres acteurs tels que les opérateurs IP. D’autres mécanismes de protection pourront alors être déployés aux niveaux supérieurs sans interférences avec la restauration au niveau transport. Outre les pannes pouvant touchers ses composantes, un réseau doit aussi faire face à l’incertitude de la matrice de trafic qu’on chercher à acheminer dans le réseau. Cette incertitude est une conséquence de la multiplication des applications et services faisant appel au réseau. La mobilité des usagers ainsi que les pannes touchant le réseau contribuent également à cette incertitude. La thèse se découpe donc en deux parties. Dans la première partie, nous nous intéressons à la complexité des différents mécanismes de sécurisation des réseaux. Dans la seconde partie, nous nous intéressons à l’incertitude de la matrice de trafic et notamment au modèle polyédrique / In the thesis robust routing design problems in resilient networks are considered. In the first part computational complexity of such problems are discussed. The following cases are considered: - path protection and path restoration - failure-dependent and failure-independent restoration - cases with and without stub-release - single-link failures and multiple-link failures (shared risk link group) - non-bifurcated (unsplittable) flows and bifurcated flows For each of the related optimization cases a mixed-integer (in the non-bifurcated cases) or linear programming formulation (in all bifurcated cases) is presented, and their computational complexity is investigated. For the NP-hard cases original NP-hardness proofs are provided, while for the polynomial cases compact linear programming formulations (which prove the polynomiality in the question) are discussed. Moreover, pricing problems related to each of the considered NP-hard problems are discussed. The second part of the thesis deals with various routing strategies in networks where the uncertainty issues are modeled using the polyhedral model. In such networks two extrema are possible. The simplest in terms of implementation, and simultaneously the least effective strategy, is the robust stable routing. On the other hand, the most effective strategy, i.e., the dynamic routing, is virtually impossible to implement in real world networks. Therefore, the major aim of this part of the thesis is to present novel routing strategies that merge the simplicity of the robust stable routing with the efficiency of the dynamic routing

This thesis deals with infrastructure network vulnerability analysis in the natural disaster context. It starts from the observation that infrastructure such as water supply or power grid has significant influence on natural disasters’ indirect consequences. The aim is to model the vulnerability to take efficient actions. The scientific approach is divided into two complementary parts. The first one deals with the vulnerability assessment, while the second one focuses on the decision aiding process to be implemented for the assessed vulnerability reducing. The proper vulnerability analysis is based on the analysis objects modelling. In order to achieve this, we will adopt a graph theory representation. A literature review will allow us to identify the graph structure which best suits the context of the thesis. In a multi network analysis environment, interdependences, i.e. relationships between components of the same network or different networks - are a determining factor for any vulnerability model. We have thus proposed an approach to model interdependence compatible with the graph theory. There are two types of relationships: the one first is functional (dependence), while the second one is dysfunctional (influence). The vulnerability is assessed by a simulation-based approach. It is composed of one part relating to the system ability to resist the feared event; and the other part relative to its ability to be back on its nominal state after the disaster. When the vulnerability is determined, the next step will be to take the necessary decisions to manage it. This part on the decision aiding is itself divided into two sub parts: first of all the process to be used for the crisis management is established. Then a methodology for decision aiding is proposed and results on a Decision Support System development. In the age of the internet and social networks, it is possible to deploy the application on the internet.

Recent discoveries uncovered flaws in machine learning algorithms such as deep neural networks. Deep neural networks seem vulnerable to small amounts of non-random noise, created by exploiting the input to output mapping of the network. Applying this noise to an input image drastically decreases classication performance. Such image is referred to as an adversarial example. The purpose of this thesis is to examine how known regularization/robustness methods perform on adversarial examples. The robustness methods: dropout, low-pass filtering, denoising autoencoder, adversarial training and committees have been implemented, combined and tested. For the well-known benchmark, the MNIST (Mixed National Institute of Standards and Technology) dataset, the best combination of robustness methods has been found. Emerged from the results of the experiments, ensemble of models trained on adversarial examples is considered to be the best approach for MNIST. Harmfulness of the adversarial noise and some robustness experiments are demonstrated on CIFAR10 (The Canadian Institute for Advanced Research) dataset as well. Apart from robustness tests, the thesis describes experiments with human classification performance on noisy images and the comparison with performance of deep neural network.

Face à un contexte croissant d'incertitude, les systèmes agricoles doivent être performants, non seulement dans des conditions moyennes, mais aussi quand les perturbations sont plus importantes. Récemment, de nombreux concepts ont été développés pour étudier la durabilité des systèmes dans des environnements changeants, dont celui de robustesse. Néanmoins, son transfert aux systèmes de grandes cultures reste limité. L'objectif de la thèse est alors de proposer un cadre d'évaluation de la robustesse qui soit opérationnel en conditions de grandes cultures. Nous nous sommes focalisés sur la culture du blé tendre et avons défini la robustesse comme la capacité d'un système agricole à maintenir ses performances de rendement malgré l'apparition de perturbations. Un modèle économétrique défini à l'échelle du système de culture a été développé pour évaluer cette robustesse face à des conditions météorologiques et des niveaux de pression maladies changeants. Il a été appliqué sur 145 exploitations agricoles françaises et 2300 parcelles de blé enquêtées sur la période 2011-2014. Les résultats montrent que la robustesse aux perturbations abiotiques et biotiques varie d'un système de culture à l'autre. Les systèmes les plus robustes (respectivement, les moins robustes) ont pu être identifiés, en tenant compte des niveaux de rendement atteints en conditions moyennes et de la sensibilité de ces rendements à des variations climatiques. Alors que les situations de rendements "élevés" versus "faibles" s'expliquent surtout par des pratiques agronomiques dites d'intensification, ce sont surtout des variables dites de flexibilité qui expliquent la robustesse versus la sensibilité aux perturbations abiotiques. Au-delà de l'intérêt de cette approche pour mieux appréhender la robustesse des systèmes agricoles, la méthodologie développée présente l'avantage de pouvoir être appliquée à d'autres performances et/ou d'autres aléas. / Given increasing uncertainties surrounding the future of agriculture, farming systems need to perform well both in average conditions and in situations presenting substantial variations. Different concepts have been developed in recent years to assess the sustainability of agricultural systems within a context of global change, including the concept of robustness. But its empirical operationalization remains a challenge, particularly with regard to arable cropping systems. The aim of this thesis is to propose a methodological framework for assessing robustness which is operational for arable conditions. Focusing on wheat production, the robustness is defined as the ability of a cropping system to maintain yield performances despite the presence of perturbations. An econometric model defined at the cropping system level is used to assess yield robustness in the face of changing weather conditions and fungal disease pressures. It is then applied to data from 145 French wheat-growing farms and 2,300 wheat plots surveyed over the period 2011-2014. The results show that yield robustness varies widely from one cropping system to another. Cropping systems showing the most and the least robustness to abiotic perturbations were identified on the basis of yield performances under both normal and changing weather conditions. While several management intensification and crop rotation practices differentiate high versus low wheat yield cropping systems, it appears to be flexibility practices that distinguish robust versus sensitive cropping systems. Beyond the interest of this approach per se to understand and improve the robustness of agricultural systems, this methodological framework could also be used to assess other performances and/or risks.

<p>Automotive positioning systems relying exclusively on the input from a GPS receiver, which is a line of sight sensor, tend to be sensitive to situations with limited sky visibility. Such situations include: urban environments with tall buildings; inside parking structures; underneath trees; in tunnels and under bridges. In these situations, the system has to rely on integration of relative motion sensors to estimate vehicle position. However, these sensor measurements are generally affected by errors such as offsets and scale factors, that will cause the resulting position accuracy to deteriorate rapidly once GPS input is lost. </p><p>The approach in this thesis is to use a GPS receiver in combination with low cost sensor equipment to produce a robust positioning module. The module should be capable of handling situations where GPS input is corrupted or unavailable. The working principle is to calibrate the relative motion sensors when GPS is available to improve the accuracy during GPS intermission. To fuse the GPS information with the sensor outputs, different models have been proposed and evaluated on real data sets. These models tend to be nonlinear, and have therefore been processed in an Extended Kalman Filter structure. </p><p>Experiments show that the proposed solutions can compensate for most of the errors associated with the relative motion sensors, and that the resulting positioning accuracy is improved accordingly.</p>

<p>Quantile regression have its advantage properties comparing to the OLS model regression which are full measurement of the effects of a covariate on response, robustness and Equivariance property. In this paper, I use a survey data in Belgium and apply a linear model to see the advantage properites of quantile regression. And I use a quantile regression model with the raw data to analyze the different cost of family on different numbers of children and apply a Wald test. The result shows that for most of the family types and living standard, from the lower quantile to the upper quantile the family cost on children increases along with the increasing number of children and the cost of each child is the same. And we found a common behavior that the cost of the second child is significantly more than the cost of the first child for a nonworking type of family and all living standard families, at the upper quantile (from 0.75 quantile to 0.9 quantile) of the conditional distribution.</p>

The goal of this thesis is to explore different control strategies to execute rhythmic movements. This issue is covered both with design perspectives (implementation in a robot) and analysis perspectives. Indeed we aim at analyzing both our robot behavior, and the behavior of human subjects executing the same task. Interesting parallels between these data sets are raised, illustrating for instance the ubiquitous trade-off of control theory between performance and robustness. / L'objectif de cette thèse est d'explorer différentes stratégies d'exécution des mouvements rythmiques. Cet objectif est couvert à la fois dans des perspectives d'implémentation (sur un robot) et d'analyse. En effet, nous souhaitons analyser à la fois le comportement de notre robot, et le comportement de sujets humains exécutant la même tâche. Des parallèles intéressants, entre les deux ensemble de données, sont proposés et illustrent, par exemple, le compromis entre la robustesse et la performance, souvent utilisé dans la thèorie du contrôle.

See appended files.

Ce travail vise à mettre en évidence la robustesse du procédé de thixoformage des aciers via létude qualitative et quantitative de linfluence des différents paramètres impliqués dans le processus. Le thixoformage est la mise en forme des métaux à létat semi-solide et à microstructure globulaire. Dans ces conditions, il a été montré que la consistance du matériau lors de la mise en forme diminue quand la vitesse de cisaillement augmente et que les paramètres régissant son comportement sont dépendants du temps. Etape-clé de cette technologie, le chauffage est étudié au même titre que la déformation. Linfluence de la stratégie de chauffage, de la géométrie de linducteur et des conditions extérieures est déterminée à laide de modélisations et dessais de réchauffage de lopins. Le transport entre le four et la presse est également abordé. Pour étudier la sensibilité de létape de déformation, trois géométries sont utilisées : le filage direct, une pièce détude rhéologique et une pièce industrielle. Ici encore, les essais pratiques et la modélisation sont employés de manière complémentaire. Les effets de la température du lopin et de celle des outils, de la vitesse de déformation, de la lubrification et des effets thermiques sont ainsi analysés. Comme pour le chauffage, la répétabilité de la déformation est mise en évidence. Par ailleurs, les défauts et structures obtenus dans les pièces sont présentés avant de discuter de la thixoformabilité des différentes nuances dacier. En particulier, la qualification dalliages spécialement dessinés pour le thixoformage ou au contraire réputés difficiles à mettre en oeuvre par cette technique est discutée. Finalement, le problème de la durée de vie des outillages est abordé puisque celui-ci est le dernier obstacle au développement de la technologie. / This work aims to assess the robustness of the steel thixoforming process by a qualitative and quantitative study of the influence of the parameters involved in the process. Thixoforming is the shaping of metals in the semi-solid state and with a globular microstructure. In this case, one shows that consistency decreases when shear rate increases and that the process parameters leading the behaviour are time dependent. As a key-step of the technology, heating is studied as well as forming. Influences of heating strategy, inductors geometry and external conditions are determined by the help of modelling and heating experiments. Handling system between heating furnace and press is also tackled. To study the formings relability and sensitivity, three geometries are used : direct extrusion, a rheological study part and an industrial part. Again, practical experiments and modelling are used in a complementary way. Effects of slugs temperature, forming speed, lubrication and tools temperature are analysed. Repeatability of heating and forming are highlighted. The defects and structures obtained in the parts are presented before discussing the thixoformability of steel grades. In particular, qualification of especially designed grades as well as non adapted grades is discussed. Finally, the tooling wear and lifetime issues are detailed as it is one of the mean locks of the technology industrialisation.