Séminaire de mathématiques appliquées (archives)

We consider the regression model\begin{equation} Y_{i}=g(x_i)+\varepsilon _{i},\,\,\,\,i=0,1,2...,n, \end{equation}where the regression function derivative has a jump point at an unknown position $\theta .$ We propose a nonparametric Kernel-based estimator of the jump location $\theta .$ Assume that $\sup_{\left| i-j\right| \geq k}\left| Cov\left( \varepsilon _{i},\varepsilon _{j}\right) \right|\leq Ck^{-\rho }$ for $0<\rho \leq 1.$ Under very general conditions, we prove the $(nh)^{\frac{-\rho}{2}}$ convergence rate of the estimator, where $h$ is the window of the kernel. This includes short-range dependent as well as long-range dependent and even non-stationary errors. Finally, we gives conditions on the windows $h$ to obtain the best rate of convergence.

Cette présentation porte sur les équations gyro-cinétiques et traite un développement rigoureux des limites de l'équation de Vlasov avec différents opérateurs de collision dans un champ magnétique fort, ainsi que du développement de méthodes numériques basées sur une décomposition micro-macro de la fonction de distribution des particules.

We present a discretization strategy for compressible fluids systems for unstrutctured grids based a Lagrange-Remap approach that does not involve any moving mesh. We present the stability properties of this solver and present a natural semi-implicit extension of the method that allows to remain stable under a CFL condition involving only the material velocity. We present another modification of the solver that allows to provide an accurate and stable solver for simulation involving low-Mach regions in the flow. This work is a collaboration with Christophe CHALONS and Mathieu GIRARDIN.

Un certain nombre d'équations aux dérivées partielles stochastiques très singulières présentent des problèmes dans leur définition même. C'est le cas, entre autres, de l'équation de KPZ, mais aussi de l'équation de quantisation stochastique en dimension 3. Les méthodes classiques d'analyse ne permettent pas de définir cette équation, et il s'avère que pour lui donner un sens, il est nécessaire de soustraire un constante infinie et de considérer formellement un nouveau problème. La théorie des distributions paracontrolées, qui combine des idées de la théorie des chemins rugueux avec la décomposition de Paley-Littlewood et le paraproduit, est un bon cadre pour donner un sens à cette renormalisation, et résoudre (localement) cette équation.

Chemotaxis is the phenomenon in which cells direct their motion according to a chemical present in their environment. Since experimental observations have shown that the motion of bacteria (e.g. Escherichia Coli) is due to the alternation of 'runs and tumbles', mathematical modelling thanks to a kinetic description has been proposed. The starting point of the study is the so-called Othmer-Dunbar-Alt model governing the dynamics of the distribution function of cells. From this system, macroscopic model can be derived after rescaling. When the taxis is small compared to the unbiased movement of cells, the scaling must be of diffusive type.

We investigate the coupling of the Multi-dimensional Optimal Order Detection (MOOD) method and the Arbitrary high order DERivatives (ADER) approach in order to design a new high order accurate, robust and computationally efficient Finite Volume (FV) scheme dedicated to solve nonlinear systems of hyperbolic conservation laws on unstructured triangular and tetrahedral meshes in two and three space dimensions, respectively.

In this presentation, we focus on a theoretical analysis and the use of statistical and optimization methods in the context of sparse linear regressions in a high-dimensional setting. The first part of this work is dedicated to the study of statistical learning methods, more precisely penalized methods and greedy algorithms. The second part concerns the application of these methods for gene regulatory networks inference. Gene regulatory networks are powerful tools to represent and analyse complex biological systems, and enable the modelling of functional relationships between elements of these systems. We thus propose to develop optimization methods to estimate relationships in such networks.

Among the numerous practical application of Statistical Process Control (SPC) is the monitoring of the number of defecting (non-conforming) items that are produced from a manufacturing company or the number of the incidents of a disease in a specic area. The basic aim is to detect any kind of change (e.g., an increase in the number of non-conforming items) quickly and accurately. In such cases, the available data are usually discrete (counts) and for process monitoring, the ordinary control charts for attributes (e.g., np-, p-, c- and u-charts) are used in practice. The main assumption is that the data come from the Poisson or the binomial distribution.

Il est bien connu que les lois marginales d'un vecteur aléatoire ne suffisent pas à caractériser sa distribution. Lorsque les lois marginales du vecteur aléatoire sont continues, le théorème de Sklar garantit l'existence et l'unicité d'une fonction appelée copule, caractérisant la dépendance entre les composantes du vecteur. La loi du vecteur aléatoire est parfaitement définie par la donnée des lois marginales et de la copule. Dans cette présentation, il sera proposé deux tests non paramétriques de détection de ruptures dans la distribution d'observations multivariées, particulièrement sensibles à des changements dans la copule des observations.