Search

We propose imposing data-driven identification constraints to alleviate the multimodality problem arising in the estimation of poorly identified dynamic stochastic general equilibrium models under non-informative prior distributions. We also devise an iterative procedure based on the posterior density of the parameters for finding these constraints. An empirical application to the Smets and Wouters () model demonstrates the properties of the estimation method, and shows how the problem of multimodal posterior distributions caused by parameter redundancy is eliminated by identification constraints. Out-of-sample forecast comparisons as well as Bayes factors lend support to the constrained model.

Sequential Monte Carlo (SMC) methods, such as the particle filter, are by now one of the standard computational techniques for addressing the filtering problem in general state-space models. However, many applications require post-processing of data offline. In such scenarios the smoothing problem--in which all the available data is used to compute state estimates--is of central interest. We consider the smoothing problem for a class of conditionally linear Gaussian models. We present a forward-backward-type Rao-Blackwellized particle smoother (RBPS) that is able to exploit the tractable substructure present in these models. Akin to the well known Rao-Blackwellized particle filter, the proposed RBPS marginalizes out a conditionally tractable subset of state variables, effectively making use of SMC only for the "intractable part" of the model. Compared to existing RBPS, two key features of the proposed method are: (i) it does not require structural approximations of the model, and (ii) the aforementioned marginalization is done both in the forward direction and in the backward direction.

Computer codes intended for neutrons or gamma rays transport simulation based on Monte-Carlo methods the variance reduction methods are applied. The weight windows method is one of them. In this report the methods of solving problem of zero-weight windows generation using the generator of computer code using the Monte-Carlo methods are presented. These methods are applicable both for neutron transfer problems and gamma rays transfer problems. В расчетных кодах, предназначенных для моделирования процессов переноса нейтронов или гамма-квантов, основанных на методе Монте-Карло, применяются методы понижения дисперсии. Один из них – метод весовых окон. В данной работе представлены методы решения проблемы генерации нулевых весовых окон при использовании генератора расчетного кода, использующего метод Монте-Карло. Данные методы применимы как для задач с переносом нейтронов, так и для задач с переносом гамма-квантов.

Classical derivatives pricing theory assumes frictionless market and infinite liquidity. These assumptions are however easily violated in real market, especially for large trades and illiquid assets. In this imperfect market, one has to consider the super-replication price as perfect hedging becomes infeasible sometimes.The first part of this dissertation focuses on proposing a model incorporating both liquidity cost and price impact. We start by deriving continuous time trading dynamics as the limit of discrete rebalancing policies. Under the constraint of holding zero underlying stock at the inception and the maturity, we obtain a quasi-linear pricing equation in the viscosity sense. A perfect hedging strategy is provided as soons as the equation admits a smooth solution. When it comes to hedging a covered European option under gamma constraint, the dynamic programming principle employed previously is no longer valid. Using stochastic target and partial differential equation smoothing techniques, we prove the super-replication price now becomes the viscosity solution of a fully non-linear parabolic equation. We also show how ε-optimal strategies can be constructed, and propose a numerical resolution scheme.The second part is dedicated to the numerical resolution of the Backward Stochastic Differential Equation (BSDE). We propose a purely forward numerical scheme, which first approximates an arbitrary Lipschitz driver by local polynomials and then applies the Picard iteration to converge to the original solution. Each Picard iteration can be represented in terms of branching diffusion systems, thus avoiding the usual estimation of conditional expectation. We also prove the convergence on an unlimited time horizon. Numerical simulation is also provided to illustrate the performance of the algorithm.; La théorie classique de la valorisation des produits dérivés se repose sur l'absence de coûts de transaction et une liquidité infinie. Ces hypothèses sont toutefois ne plus véridiques dans le marché réel, en particulier quand la transaction est grande et les actifs non-liquides. Dans ce marché imparfait, on parle du prix de sur-réplication puisque la couverture parfaite est devenue parfois infaisable.La première partie de cette thèse se concentre sur la proposition d’un modèle qui intègre à la fois le coût de transaction et l’impact sur le prix du sous-jacent. Nous commençons par déduire la dynamique de l’actif en temps continu en tant que la limite de la dynamique en temps discret. Sous la contrainte d’une position nulle sur l’actif au début et à la maturité, nous obtenons une équation quasi-linéaire pour le prix du dérivé, au sens de viscosité. Nous offrons la stratégie de couverture parfaite lorsque l’équation admet une solution régulière. Quant à la couverture d’une option européenne “covered” sous la contrainte gamma, le principe de programme dynamique utilisé précédemment n'est plus valide. En suivant les techniques du cible stochastique et de l’équation différentielle partielle, nous démontrons que le prix de la sur-réplication est devenue une solution de viscosité d’une équation non linéaire de type parabolique. Nous construisons également la stratégie ε-optimale, et proposons un schéma numérique.La deuxième partie de cette thèse est consacrée aux études sur un nouveau schéma numérique d'EDSR, basé sur le processus de branchement. Nous rapprochons tout d’abord le générateur Lipschitzien par une suite de polynômes locaux, puis appliquons l’itération de Picard. Chaque itération de Picard peut être représentée en termes de processus de branchement. Nous démontrons la convergence de notre schéma sur l’horizon temporel infini. Un exemple concret est discuté à la fin dans l’objectif d’illustrer la performance de notre algorithme.

Classical derivatives pricing theory assumes frictionless market and infinite liquidity. These assumptions are however easily violated in real market, especially for large trades and illiquid assets. In this imperfect market, one has to consider the super-replication price as perfect hedging becomes infeasible sometimes.The first part of this dissertation focuses on proposing a model incorporating both liquidity cost and price impact. We start by deriving continuous time trading dynamics as the limit of discrete rebalancing policies. Under the constraint of holding zero underlying stock at the inception and the maturity, we obtain a quasi-linear pricing equation in the viscosity sense. A perfect hedging strategy is provided as soons as the equation admits a smooth solution. When it comes to hedging a covered European option under gamma constraint, the dynamic programming principle employed previously is no longer valid. Using stochastic target and partial differential equation smoothing techniques, we prove the super-replication price now becomes the viscosity solution of a fully non-linear parabolic equation. We also show how ε-optimal strategies can be constructed, and propose a numerical resolution scheme.The second part is dedicated to the numerical resolution of the Backward Stochastic Differential Equation (BSDE). We propose a purely forward numerical scheme, which first approximates an arbitrary Lipschitz driver by local polynomials and then applies the Picard iteration to converge to the original solution. Each Picard iteration can be represented in terms of branching diffusion systems, thus avoiding the usual estimation of conditional expectation. We also prove the convergence on an unlimited time horizon. Numerical simulation is also provided to illustrate the performance of the algorithm.; La théorie classique de la valorisation des produits dérivés se repose sur l'absence de coûts de transaction et une liquidité infinie. Ces hypothèses sont toutefois ne plus véridiques dans le marché réel, en particulier quand la transaction est grande et les actifs non-liquides. Dans ce marché imparfait, on parle du prix de sur-réplication puisque la couverture parfaite est devenue parfois infaisable.La première partie de cette thèse se concentre sur la proposition d’un modèle qui intègre à la fois le coût de transaction et l’impact sur le prix du sous-jacent. Nous commençons par déduire la dynamique de l’actif en temps continu en tant que la limite de la dynamique en temps discret. Sous la contrainte d’une position nulle sur l’actif au début et à la maturité, nous obtenons une équation quasi-linéaire pour le prix du dérivé, au sens de viscosité. Nous offrons la stratégie de couverture parfaite lorsque l’équation admet une solution régulière. Quant à la couverture d’une option européenne “covered” sous la contrainte gamma, le principe de programme dynamique utilisé précédemment n'est plus valide. En suivant les techniques du cible stochastique et de l’équation différentielle partielle, nous démontrons que le prix de la sur-réplication est devenue une solution de viscosité d’une équation non linéaire de type parabolique. Nous construisons également la stratégie ε-optimale, et proposons un schéma numérique.La deuxième partie de cette thèse est consacrée aux études sur un nouveau schéma numérique d'EDSR, basé sur le processus de branchement. Nous rapprochons tout d’abord le générateur Lipschitzien par une suite de polynômes locaux, puis appliquons l’itération de Picard. Chaque itération de Picard peut être représentée en termes de processus de branchement. Nous démontrons la convergence de notre schéma sur l’horizon temporel infini. Un exemple concret est discuté à la fin dans l’objectif d’illustrer la performance de notre algorithme.

We provide a representation result of parabolic semi-linear PDEs, with polynomial nonlinearity, by branching diffusion processes. We extend the classical representation for KPP equations, introduced by Skorokhod [Theory Probab. Appl. 9 (1964) 445–449], Watanabe [J. Math. Kyoto Univ. 4 (1965) 385–398] and McKean [Comm. Pure Appl. Math. 28 (1975) 323–331], by allowing for polynomial nonlinearity in the pair $(u,Du)$, where $u$ is the solution of the PDE with space gradient $Du$. Similar to the previous literature, our result requires a non-explosion condition which restrict to “small maturity” or “small nonlinearity” of the PDE. Our main ingredient is the Malliavin automatic differentiation technique as in [Ann. Appl. Probab. 27 (2017) 3305–3341], based on the Malliavin integration by parts, which allows to account for the nonlinearities in the gradient. As a consequence, the particles of our branching diffusion are marked by the nature of the nonlinearity. This new representation has very important numerical implications as it is suitable for Monte Carlo simulation. Indeed, this provides the first numerical method for high dimensional nonlinear PDEs with error estimate induced by the dimension-free central limit theorem. The complexity is also easily seen to be of the order of the squared dimension. The final section of this paper illustrates the efficiency of the algorithm by some high dimensional numerical experiments.

Under supercoiling constraints, naked DNA, such as a large part of bacterial DNA, folds into braided structures called plectonemes. The double-helix can also undergo local structural transitions, leading to the formation of denaturation bubbles and other alternative structures. Various polymer models have been developed to capture these properties, with Monte-Carlo (MC) approaches dedicated to the inference of thermodynamic properties. In this chapter, we explain how to perform such Monte-Carlo simulations, following two objectives. On one hand, we present the self-avoiding supercoiled Worm-Like Chain (ssWLC) model, which is known to capture the folding properties of supercoiled DNA, and provide a detailed explanation of a standard MC simulation method. On the other hand, we explain how to extend this ssWLC model to include structural transitions of the helix., Book chapter to appear in The Bacterial Nucleoid, Methods and Protocols, Springer series

This thesis is devoted to the numerical study of two stochastic models arising in Bose-Einstein condensation physics. They constitute two generalisations of the Gross-Pitaevskii Equation. This deterministic partial differential equation model the wave function dynamics of a Bose-Einstein condensate trapped in an external confining potential. The first chapter contains a simple presentation of the Bose-Einstein condensation phenomenon and of the experimental methods used to construct such systems.The first model considered enables to model the fluctuations of the confining potential intensity, and takes the form of a stochastic partial differential equation. In practice, these fluctuations lead to heating of the condensate and possibly to its collapse. In the second chapter we propose to build a numerical scheme to solve this model. It is based on a spectral space discretisation and a Crank-Nicolson discretisation in space. We show that the proposed scheme converges strongly at order at least one in probability. We also present numerical simulations to illustrate this result. The third chapter is devoted to the numerical and theoretical study of the dynamics of a stationary solution (for the deterministic equation) of vortex type. We study the influence of random disturbances of the confining potential on the solution. We show that the disturbed solution conserves the symmetries of the stationary solution for times up to at least the square of the inverse of the fluctuations intensity. These results are illustrated with numerical simulations based on a Monte-Carlo method suited to rare events estimation.The second model can be used to model the effects of the temperature on the dynamics of a Bose-Einstein condensate. In the case of finite temperature, the Bose-Einstein condensation is not complete and the condensate interacts with the non-condensed particles. These interactions are interesting to understand the dynamics of the phase transition and analyse the phenomena of symmetry breaking associated, like the spontaneous nucleation of vortices We have studied in the fourth and the fifth chapters some questions linked to the long time simulation of this model solutions. The fourth chapter is devoted to the construction of an unbiased sampling method of measures known up to a multiplicative constant. The distinctive feature of this Markov-Chain Monte-Carlo algorithm is that it enables to perform an unbiased non-reversible sampling based on an overdamped Langevin equation. It constitutes a generalization of the Metropolis-Adjusted Langevin Algorithm (MALA). The fifth chapter is devoted to the numerical study of metastable dynamics linked to the nucleation of vortices in rotating Bose-Einstein condensates. A numerical integrator and a suited Monte-Carlo methods for the simulation of metastable dynamics are proposed. This Monte-Carlo method is based on the Adaptive Multilevel Splitting (AMS) algorithm., Cette thèse porte sur l'étude de méthodes numériques pour l'analyse de deux modèles stochastiques apparaissant dans le contexte de la condensation de Bose-Einstein. Ceux-ci constituent deux généralisations de l'équation de Gross-Pitaevskii. Cette équation aux dérivées partielles déterministe modélise la dynamique de la fonction d'onde d'un condensat de Bose-Einstein piégé par un potentiel extérieur confinant.Le premier modèle étudié permet de modéliser les fluctuations de l'intensité du potentiel confinant et prend la forme d'une équation aux dérivées partielles stochastiques. Celles-ci conduisent en pratique à un échauffement du condensat, et parfois mêmeà son effondrement. Nous proposons dans un premier chapitre la construction d'un schéma numérique pour la résolution de ce modèle. Il est fondé sur une discrétisation spectrale en espace, et une discrétisation temporelle de type Crank-Nicolson. Nous démontrons que le schéma proposé converge fortement en probabilité à l'ordre au moins 1 en temps, et nous présentons des simulations numériques illustrant ce résultat. Le deuxième chapitre est consacré à l'étude théorique et numérique de la dynamique d'une solution stationnaire (pour l'équation déterministe) de type vortex. Nous étudions l'influence des perturbations aléatoires du potentiel sur la solution, et montrons que la solution perturbée garde les symétries de la solution stationnaire pour des temps au moins de l'ordre du carré de l'inverse de l'intensité des fluctuations. Ces résultats sont illustrés par des simulations numériques exploitant une méthode de Monte-Carlo adaptée à la simulation d'événements rares.Le deuxième modèle permet de modéliser les effets de la température sur la dynamique d'un condensat. Lorsque celle-ci n'est pas nulle, la condensation n'est pas complète et le condensat interagit avec les particules non condensées. Ces interactions sont d'un grand intérêt pour comprendre la dynamique de transition de phase et analyser les phénomènes de brisure de symétrie associés, comme la formation spontanée de vortex. Nous nous sommes intéressés dans les chapitres 3 et 4 à des questions relatives à la simulation de la distribution des solutions de cette équation en temps long. Le troisième chapitre est consacré à la construction d'une méthode d’échantillonnage sans biais pour des mesures connues à une constante multiplicative près. C'est une méthode de Monte Carlo par chaînes de Markov qui a la particularité de permettre un échantillonnage non-réversible basé sur une équation de type Langevin sur-amortie. Elle constitue une extension de Metropolis-Adjusted Langevin Algorithm (MALA). Le quatrième chapitre est quant à lui consacré à l'étude numérique de dynamiques métastables liées à la nucléation de vortex dans des condensats en rotation. Un intégrateur numérique pour la dynamique étudiée est proposé, ainsi qu'une méthode de Monte-Carlo adaptée à la simulation d'événements rares correspondant aux changements de configurations métastables. Cette dernière est basée sur l'algorithme Adaptive Multilevel Splitting (AMS).

Event-based models find frequent usage in fields such as computational physics and biology as they may contain both continuous and discrete state variables and may incorporate both deterministic and stochastic state transitions. If the state transitions are stochastic, computer-generated random numbers are used to obtain the model solution. This type of event-based computations is also known as Monte-Carlo simulation. In this thesis, I study different approaches to execute event-based computations on parallel computers. This ultimately allows users to retrieve their simulation results in a fraction of the original computation time. As system sizes grow continuously or models have to be simulated at longer time scales, this is a necessary approach for current computational tasks. More specifically, I propose several ways to asynchronously simulate such models on parallel shared-memory computers, for example using parallel discrete-event simulation or task-based computing. The particular event-based models studied herein find applications in systems biology, computational epidemiology and computational neuroscience. In the presented studies, the proposed methods allow for high efficiency of the parallel simulation, typically scaling well with the number of used computer cores. As the scaling typically depends on individual model properties, the studies also investigate which quantities have the greatest impact on the simulation performance. Finally, the presented studies include other insights into event-based computations, such as methods how to estimate parameter sensitivity in stochastic models and how to simulate models that include both deterministic and stochastic state transitions. UPMARC

International audience; We extend the branching process based numerical algorithm of Bouchard et al. [3], that is dedicated to semilinear PDEs (or BSDEs) with Lipschitz nonlinearity, to the case where the nonlinearity involves the gradient of the solution. As in [3], this requires a localization procedure that uses a priori estimates on the true solution, so as to ensure the well-posedness of the involved Picard iteration scheme, and the global convergence of the algorithm. When, the nonlinearity depends on the gradient, the later needs to be controlled as well. This is done by using a face-lifting procedure. Convergence of our algorithm is proved without any limitation on the time horizon. We also provide numerical simulations to illustrate the performance of the algorithm.

This thesis is set within the context of Accordys, a knowledge engineeringproject aiming at providing a case-based reasoning system for fetopathology,i.e. the medical domain studying rare diseases and dysmorphia of fetuses. Theproject is based on a corpus of french fetal exam reports. This materialconsists in raw text reports diplaying a very specific vocabulary (onlypartially formalized in french medical terminologies), a "note taking" stylethat makes difficult to use tools analysing the grammar in the text, and alayout and formatting that shows a latent common structuration (organisation insections, sub-sections, observations). This thesis aims at testing thehypothesis that a uniformisation of the representation of cases that couldexploit this arborescent structure by mapping it with a tree-shaped case modelcan support the constitution of a case base which preserves the informationcontained in original reports and the similarity measurement between two cases.Mapping a case with the model (instanciating the case model) is done througha Monte Carlo tree matching method. We compare this with similarity measurementsobtained by representing our reports (both without further processing and aftersemantic enrichment through a semantic annotator) in a vector model.; Cette thèse se place dans le contexte d'Accordys, un projet d'ingénierie desconnaissances qui vise à fournir un système de rapprochement de cas enfœtopathologie, qui est le domaine de l'étude des maladies rares et dysmorphiesdu fœtus. Ce projet se base sur un corpus de comptes rendus d'examens fœtaux.Ce matériel consiste en des comptes rendus en texte brut présentant unvocabulaire très spécifique (qui n'est que partiellement formalisé dans desterminologies médicales en français), des économies linguistiques (un style"prise de notes" très prononcé rendant difficile l'utilisation d'outilsanalysant la grammaire du texte) et une mise en forme matérielle exhibant unestructuration commune latente (un découpage en sections, sous-sections,observations). Cette thèse vise à tester l'hypothèse qu'une uniformisation dela représentation des cas exploitant cette structure arborescente en la faisantcorrespondre à un modèle de cas (lui aussi arborescent) peut supporter laconstitution d'une base de cas qui conserve les informations contenues dans lescomptes rendus originaux et permette la mesure de similarité entre deux cas. Lamise en correspondance entre cas et modèle (instanciation du modèle) estréalisée via un mapping d'arbres ayant pour base une méthode de MonteCarlo. Nous comparerons ceci avec des mesures de similarités obtenues enreprésentant nos comptes rendus (soit tels quels, soit enrichis sémantiquementgrâce à un annotateur sémantique) dans un modèle vectoriel.

Infectious diseases exert a large and in many contexts growing burden on human health, but violate most of the assumptions of classical epidemiological statistics and hence require a mathematically sophisticated approach. Viral shedding data are collected during human studies—either where volunteers are infected with a disease or where existing cases are recruited—in which the levels of live virus produced over time are measured. These have traditionally been difficult to analyse due to strong, complex correlations between parameters. Here, we show how a Bayesian approach to the inverse problem together with modern Markov chain Monte Carlo algorithms based on information geometry can overcome these difficulties and yield insights into the disease dynamics of two of the most prevalent human pathogens—influenza and norovirus—as well as Ebola virus disease.

Bayesian inverse problems often involve sampling posterior distributions on infinite-dimensional function spaces. Traditional Markov chain Monte Carlo (MCMC) algorithms are characterized by deteriorating mixing times upon mesh-refinement, when the finite-dimensional approximations become more accurate. Such methods are typically forced to reduce step-sizes as the discretization gets finer, and thus are expensive as a function of dimension. Recently, a new class of MCMC methods with mesh-independent convergence times has emerged. However, few of them take into account the geometry of the posterior informed by the data. At the same time, recently developed geometric MCMC algorithms have been found to be powerful in exploring complicated distributions that deviate significantly from elliptic Gaussian laws, but are in general computationally intractable for models defined in infinite dimensions. In this work, we combine geometric methods on a finite-dimensional subspace with mesh-independent infinite-dimensional approaches. Our objective is to speed up MCMC mixing times, without significantly increasing the computational cost per step (for instance, in comparison with the vanilla preconditioned Crank-Nicolson (pCN) method). This is achieved by using ideas from geometric MCMC to probe the complex structure of an intrinsic finite-dimensional subspace where most data information concentrates, while retaining robust mixing times as the dimension grows by using pCN-like methods in the complementary subspace. The resulting algorithms are demonstrated in the context of three challenging inverse problems arising in subsurface flow, heat conduction and incompressible flow control. The algorithms exhibit up to two orders of magnitude improvement in sampling efficiency when compared with the pCN method., 37 pages, 15 figures

A classical requirement in the design of communication networks is that all entities must be connected. In a network where links may fail, the connectedness probability is called all-terminal reliability. The model is suitable for FTTH services, where link failures are unpredictable. In real scenarios, terminals must be connected by a limited number of hops. Therefore, we study the Diameter- Constrained Reliability (DCR). We are given a simple graph G = (V,E), a subset K V of terminals, a diameter d and independent failure probabilities q = 1 − p for each link. The goal is to find the probability Rd K,G that all terminals remain connected by paths composed by d hops or less. The general DCR computation is NP-Hard, and the target probability is a polynomial in p. In this chapter we study the DCR metric. It connects reliability with quality, and should be considered in the design of the physical layer in FTTH services together with connectivity requirements. We include a full discussion of the computational complexity of the DCR as a function of the number of terminals k = |K| and diameter d. Then, we find efficient DCR computation for Monma graphs, an outstanding family of topologies from robust network design. The computation suggests corollaries that enrich the subset of instances that accept efficient DCR computation. Given its NP-Hardness, several Monte Carlo-based algorithms algorithms are designed in order to find the DCR in general, inspired in two approaches: counting and interpolation. The results suggest that counting techniques outperform interpolation, and show scalability properties as well. Open problems and trends for future work are included in the conclusions.

Due to the Markovian dependence structure, the normalizing constant of Markov random fields cannot be computed with standard analytical or numerical methods. This forms a central issue in terms of parameter inference or model selection as the computation of the likelihood is an integral part of the procedure. When the Markov random field is directly observed, we propose to estimate the posterior distribution of model parameters by replacing the likelihood with a composite likelihood, that is a product of marginal or conditional distributions of the model easy to compute. Our first contribution is to correct the posterior distribution resulting from using a misspecified likelihood function by modifying the curvature at the mode in order to avoid overly precise posterior parameters.In a second part we suggest to perform model selection between hidden Markov random fields with approximate Bayesian computation (ABC) algorithms that compare the observed data and many Monte-Carlo simulations through summary statistics. To make up for the absence of sufficient statistics with regard to this model choice, we introduce summary statistics based on the connected components of the dependency graph of each model in competition. We assess their efficiency using a novel conditional misclassification rate that evaluates their local power to discriminate between models. We set up an efficient procedure that reduces the computational cost while improving the quality of decision and using this local error rate we build up an ABC procedure that adapts the summary statistics to the observed data.In a last part, in order to circumvent the computation of the intractable likelihood in the Bayesian Information Criterion (BIC), we extend the mean field approaches by replacing the likelihood with a product of distributions of random vectors, namely blocks of the lattice. On that basis, we derive BLIC (Block Likelihood Information Criterion) that answers model choice questions of a wider scope than ABC, such as the joint selection of the dependency structure and the number of latent states. We study the performances of BLIC in terms of image segmentation.; La constante de normalisation des champs de Markov se présente sous la forme d'une intégrale hautement multidimensionnelle et ne peut être calculée par des méthodes analytiques ou numériques standard. Cela constitue une difficulté majeure pour l'estimation des paramètres ou la sélection de modèle. Pour approcher la loi a posteriori des paramètres lorsque le champ de Markov est observé, nous remplaçons la vraisemblance par une vraisemblance composite, c'est à dire un produit de lois marginales ou conditionnelles du modèle, peu coûteuses à calculer. Nous proposons une correction de la vraisemblance composite basée sur une modification de la courbure au maximum afin de ne pas sous-estimer la variance de la loi a posteriori. Ensuite, nous proposons de choisir entre différents modèles de champs de Markov cachés avec des méthodes bayésiennes approchées (ABC, Approximate Bayesian Computation), qui comparent les données observées à de nombreuses simulations de Monte-Carlo au travers de statistiques résumées. Afin de pallier l'absence de statistiques exhaustives pour ce choix de modèle, des statistiques résumées basées sur les composantes connexes des graphes de dépendance des modèles en compétition sont introduites. Leur efficacité est étudiée à l'aide d'un taux d'erreur conditionnel original mesurant la puissance locale de ces statistiques à discriminer les modèles. Nous montrons alors que nous pouvons diminuer sensiblement le nombre de simulations requises tout en améliorant la qualité de décision, et utilisons cette erreur locale pour construire une procédure ABC qui adapte le vecteur de statistiques résumés aux données observées. Enfin, pour contourner le calcul impossible de la vraisemblance dans le critère BIC (Bayesian Information Criterion) de choix de modèle, nous étendons les approches champs moyens en substituant la vraisemblance par des produits de distributions de vecteurs aléatoires, à savoir des blocs du champ. Le critère BLIC (Block Likelihood Information Criterion), que nous en déduisons, permet de répondre à des questions de choix de modèle plus large que les méthodes ABC, en particulier le choix conjoint de la structure de dépendance et du nombre d'états latents. Nous étudions donc les performances de BLIC dans une optique de segmentation d'images.

Crude Monte-Carlo or quasi Monte-Carlo methods are well suited to characterize events of which associated probabilities are not too low with respect to the simulation budget. For very seldom observed events, such as the collision probability between two aircraft in airspace, these approaches do not lead to accurate results. Indeed, the number of available samples is often insufficient to estimate such low probabilities (at least 10 6 samples are needed to estimate a probability of order 10 - 4 with 10% relative error with Monte-Carlo simulations). In this article, one reviewed different appropriate techniques to estimate rare event probabilities that require a fewer number of samples. These methods can be divided into four main categories: parameterization techniques of probability density function tails, simulation techniques such as importance sampling or importance splitting, geometric methods to approximate input failure space and finally, surrogate modeling. Each technique is detailed, its advantages and drawbacks are described and a synthesis that aims at giving some clues to the following question is given: “which technique to use for which problem?”.

In this article, we derive expressions for conditional expectations in terms of regular expectations without conditioning but involving some weights. For this purpose, we apply two approaches: the conditional density method and the Malliavin method. We use these expressions for the numerical estimation of the price of American options and their deltas in a Lévy and jump-diffusion setting. Several examples of applications to financial and energy markets are given including numerical examples.

In this thesis, we consider two problems that statistical model checking must cope. The first problem concerns heterogeneous systems, that naturally introduce complexity and non-determinism into the analysis. The second problem concerns rare properties, difficult to observe, and so to quantify. About the first point, we present original contributions for the formalism of composite systems in BIP language. We propose SBIP, a stochastic extension and define its semantics. SBIP allows the recourse to the stochastic abstraction of components and eliminate the non-determinism. This double effect has the advantage of reducing the size of the initial system by replacing it by a system whose semantics is purely stochastic, a necessary requirement for standard statistical model checking algorithms to be applicable. The second part of this thesis is devoted to the verification of rare properties in statistical model checking. We present a state-of-the-art algorithm for models described by a set of guarded commands. Lastly, we motivate the use of importance splitting for statistical model checking and set up an optimal splitting algorithm. Both methods pursue a common goal to reduce the variance of the estimator and the number of simulations. Nevertheless, they are fundamentally different, the first tackling the problem through the model and the second through the properties.; Dans cette thèse, nous considérons deux problèmes auxquels le model checking statistique doit faire face. Le premier concerne les systèmes hétérogènes qui introduisent complexité et non-déterminisme dans l'analyse. Le second problème est celui des propriétés rares, difficiles à observer et donc à quantifier. Pour le premier point, nous présentons des contributions originales pour le formalisme des systèmes composites dans le langage BIP. Nous en proposons une extension stochastique, SBIP, qui permet le recours à l'abstraction stochastique de composants et d'éliminer le non-déterminisme. Ce double effet a pour avantage de réduire la taille du système initial en le remplaçant par un système dont la sémantique est purement stochastique sur lequel les algorithmes de model checking statistique sont définis. La deuxième partie de cette thèse est consacrée à la vérification de propriétés rares. Nous avons proposé le recours à un algorithme original d'échantillonnage préférentiel pour les modèles dont le comportement est décrit à travers un ensemble de commandes. Nous avons également introduit les méthodes multi-niveaux pour la vérification de propriétés rares et nous avons justifié et mis en place l'utilisation d'un algorithme multi-niveau optimal. Ces deux méthodes poursuivent le même objectif de réduire la variance de l'estimateur et le nombre de simulations. Néanmoins, elles sont fondamentalement différentes, la première attaquant le problème au travers du modèle et la seconde au travers des propriétés.

In this thesis, we consider two problems that statistical model checking must cope. The first problem concerns heterogeneous systems, that naturally introduce complexity and non-determinism into the analysis. The second problem concerns rare properties, difficult to observe, and so to quantify. About the first point, we present original contributions for the formalism of composite systems in BIP language. We propose SBIP, a stochastic extension and define its semantics. SBIP allows the recourse to the stochastic abstraction of components and eliminate the non-determinism. This double effect has the advantage of reducing the size of the initial system by replacing it by a system whose semantics is purely stochastic, a necessary requirement for standard statistical model checking algorithms to be applicable. The second part of this thesis is devoted to the verification of rare properties in statistical model checking. We present a state-of-the-art algorithm for models described by a set of guarded commands. Lastly, we motivate the use of importance splitting for statistical model checking and set up an optimal splitting algorithm. Both methods pursue a common goal to reduce the variance of the estimator and the number of simulations. Nevertheless, they are fundamentally different, the first tackling the problem through the model and the second through the properties.; Dans cette thèse, nous considérons deux problèmes auxquels le model checking statistique doit faire face. Le premier concerne les systèmes hétérogènes qui introduisent complexité et non-déterminisme dans l'analyse. Le second problème est celui des propriétés rares, difficiles à observer et donc à quantifier. Pour le premier point, nous présentons des contributions originales pour le formalisme des systèmes composites dans le langage BIP. Nous en proposons une extension stochastique, SBIP, qui permet le recours à l'abstraction stochastique de composants et d'éliminer le non-déterminisme. Ce double effet a pour avantage de réduire la taille du système initial en le remplaçant par un système dont la sémantique est purement stochastique sur lequel les algorithmes de model checking statistique sont définis. La deuxième partie de cette thèse est consacrée à la vérification de propriétés rares. Nous avons proposé le recours à un algorithme original d'échantillonnage préférentiel pour les modèles dont le comportement est décrit à travers un ensemble de commandes. Nous avons également introduit les méthodes multi-niveaux pour la vérification de propriétés rares et nous avons justifié et mis en place l'utilisation d'un algorithme multi-niveau optimal. Ces deux méthodes poursuivent le même objectif de réduire la variance de l'estimateur et le nombre de simulations. Néanmoins, elles sont fondamentalement différentes, la première attaquant le problème au travers du modèle et la seconde au travers des propriétés.

In this paper, we derive moment conditions for particle filter importance weights, which ensure that the particle filter estimates of the expectations of bounded Borel functions converge in mean square and $L^4$ sense, and that the empirical measure of the particle filter converges weakly to the true filtering measure. The result extends the previously derived conditions by not requiring the boundedness of the importance weights, but only boundedness of second or fourth order moments. We show that the boundedness of the second order moments of the weights implies the convergence of the estimates bounded functions in the mean square sense, and the $L^4$ convergence as well as the empirical measure convergence are assured by the boundedness of the fourth order moments of the weights. We also present an example class of models and importance distributions where the moment conditions hold, but the boundedness does not. The unboundedness in these models is caused by point-singularities in the weights which still leave the weight moments bounded. We show by using simulated data that the particle filter for this kind of model also performs well in practice., 12 pages, 4 figures

International audience; This book constitutes the refereed proceedings of the Computer Games Workshop, CGW 2013, held in Beijing, China, in August 2013, in conjunction with the Twenty-third International Conference on Artificial Intelligence, IJCAI 2013. The 9 revised full papers presented were carefully reviewed and selected from 15 submissions. The papers cover a wide range of topics related to computer games. They discuss six games that are played by humans in practice: Chess, Domineering, Chinese Checkers, Go, Goofspiel, and Tzaar. Moreover, there are papers about the Sliding Tile Puzzle, an application, namely, Cooperative Path-Finding Problems, and on general game playing.

The aim of this paper is to provide a fatigue-life assessment of an existing composite steel-concrete bridge within Switzerland. The Venoge bridge is a 219 meters length, four spans, four girder composite bridge, situ-ated on the Geneva-Lausanne Swiss motorway. Stress response influence lines at fatigue critical details are characterized with a refined three-dimensional (3-D) finite element (FE) model that takes into account the ef-fective transverse load repartition and the cracking of the concrete slab; the model is further validated using strain gauge measurements. Real traffic simulations are run to get the effective traffic histograms on the bridge. Weigh in Motion (WIM) data from the Denges station (1km from the bridge) are used as input for the simulations. The effective slow lane position over the bridge section is taken in account and future traffic in-crease is considered. The fatigue failure probability of the bridge at a critical location under the long-term ef-fect of highway traffic is calculated by defining a limit state function for fatigue reliability analysis. The re-sults show that fatigue is not a concern for the Venoge bridge, even though particular attention has to be paid to future change in traffic lane configurations which could require re-consideration of the fatigue critical de-tails.

We consider Monte-Carlo discretizations of partial differential equations based on a combination of semi-lagrangian schemes and probabilistic representations of the solutions. The goal of this paper is two-fold. First we give rigorous convergence estimates for our algorithm: In a simple setting, we show that under an anti-Courant-Friedrichs-Lewy condition on the time step $\delta t$ and on the mesh size $\delta x$ and for a reasonably large number of independent realizations $N$, we control the Monte-Carlo error by a term of order $\mathcal {O}(\sqrt {\delta t /N})$. Then, we show various applications of the numerical method in very general situations (nonlinear, different boundary conditions, higher dimension) and numerical examples showing that the theoretical bound obtained in the simple case seems to persist in more complex situations

International audience; We present a parallel algorithm for solving backward stochastic differential equations. We improve the algorithm proposed in Gobet Labart (2010), based on an adaptive Monte Carlo method with Picard's iterations, and propose a parallel version of it. We test our algorithm on linear and non linear drivers up to dimension 8 on a cluster of 312 CPUs. We obtained very encouraging speedups greater than 0.7.

Hybrid Moments/PDF methods have shown to be well suitable for the description of polydispersed turbulent two-phase flows in non-equilibrium which are encountered in some industrial situations involving chemical reactions, combustion or sprays. They allow to obtain a fine enough physical description of the polydispersity, non-linear source terms and convection phenomena. However, their approximations are noised with the statistical error, which in several situations may be a source of a bias. An alternative hybrid Moments-Moments/PDF approach examined in this work consists in coupling the Moments and the PDF descriptions, within the description of the dispersed phase itself. This hybrid method could reduce the statistical error and remove the bias. However, such a coupling is not straightforward in practice and requires the development of accurate and stable numerical schemes. The approaches introduced in this work rely on the combined use of the upwinding and relaxation-type techniques. They allow to obtain stable unsteady approximations for a system of partial differential equations containing non-smooth external data which are provided by the PDF part of the model. A comparison of the results obtained using the present method with those of the "classical" hybrid approach is presented in terms of the numerical errors for a case of a co-current gas-particle wall jet.; Les méthodes hybrides Moments/PDF sont bien adaptées pour la description des écoulements diphasiques turbulents, polydispersés, hors équilibre thermodynamique. Ces méthodes permettent d'avoir une description assez fine de la polydispersion, de la convection et des termes sources non-linéaires. Cependant, les approximations issues de telles simulations sont bruitées ce qui, dans certaines situations, occasionne un biais. L'approche alternative étudiée dans ce travail consiste à coupler une description Eulerienne des moments avec une description stochastique Lagrangienne à l'intérieur de la phase dispersée, permettant de réduire l'erreur statistique et d'éliminer le biais. La mise en oeuvre de cette méthode nécessite le développement de schémas numériques robustes. Les approches proposées sont basées sur l'utilisation simultanée des techniques de relaxation et de décentrement, et permettent d'obtenir des approximations stables des solutions instationnaires du système d'équations aux dérivées partielles, avec des données peu régulières provenant du couplage avec le modèle stochastique. Une comparaison des résultats de la méthode hybride Moments-Moments/PDF avec ceux issus de la méthode hybride ''classique'' est présentée en termes d'analyse des erreurs numériques sur un cas de jet co-courant gaz-particules.

Microscopic models of flocking and swarming take into account large numbers of interacting individuals. Numerical resolution of large flocks implies huge computational costs. Typically for $N$ interacting individuals we have a cost of $O(N^2)$. We tackle the problem numerically by considering approximated binary interaction dynamics described by kinetic equations and simulating such equations by suitable stochastic methods. This approach permits us to compute approximate solutions as functions of a small scaling parameter $\varepsilon$ at a reduced complexity of $O(N)$ operations. Several numerical results show the efficiency of the algorithms proposed.

In a first part, we are interested in the behavior of a system of Stochastic PDEs with two time-scales- more precisely, we focus on the approximation of the slow component thanks to an efficient numerical scheme. We first prove an averaging principle, which states that the slow component converges to the solution of the so-called averaged equation. We then show that a numerical scheme of Euler type provides a good approximation of an unknown coefficient appearing in the averaged equation. Finally, we build and we analyze a discretization scheme based on the previous results, according to the HMM methodology (Heterogeneous Multiscale Method). We precise the orders of convergence with respect to the time-scale parameter and to the parameters of the numerical discretization- we study the convergence in a strong sense - approximation of the trajectories - and in a weak sense - approximation of the laws. In a second part, we study a method for approximating solutions of parabolic PDEs, which combines a semi-lagrangian approach and a Monte-Carlo discretization. We first show in a simplified situation that the variance depends on the discretization steps. We then provide numerical simulations of solutions, in order to show some possible applications of such a method., Dans une première partie, on s'intéresse à un système d'EDP stochastiques variant selon deux échelles de temps, et plus particulièrement à l'approximation de la composante lente à l'aide d'un schéma numérique efficace. On commence par montrer un principe de moyennisation, à savoir la convergence de la composante lente du système vers la solution d'une équation dite moyennée. Ensuite on prouve qu'un schéma numérique de type Euler fournit une bonne approximation d'un coefficient inconnu apparaissant dans cette équation moyennée. Finalement, on construit et on analyse un schéma de discrétisation du système à partir des résultats précédents, selon la méthodologie dite HMM (Heterogeneous Multiscale Method). On met en évidence l'ordre de convergence par rapport au paramètre d'échelle temporelle et aux différents paramètres du schéma numérique- on étudie les convergences au sens fort (approximation des trajectoires) et au sens faible (approximation des lois). Dans une seconde partie, on étudie une méthode d'approximation de solutions d'EDP paraboliques, en combinant une approche semi-lagrangienne et une discrétisation de type Monte-Carlo. On montre d'abord dans un cas simplifié que la variance dépend des pas de discrétisation- enfin on fournit des simulations numériques de solutions, afin de mettre en avant les applications possibles d'une telle méthode.

Декларируется проблема построения безопасной “песочницы” для выполнения внешнего кода в гридах на основе языка программирования Ruby. Анализируются пути решения этой проблемы., Problem of creation of safe sandbox for execution of external code inside Ruby grids was stated. Conducted analysis of possible solution of the problem., Декларується проблема побудови безпечної “пісочниці” для виконання зовнішнього коду в грідах на основі мови програмування Ruby. Аналізуються шляхи вирішення цієї проблеми.

Phylodynamics - the field aiming to quantitatively integrate the ecological and evolutionary dynamics of rapidly evolving populations like those of RNA viruses – increasingly relies upon coalescent approaches to infer past population dynamics from reconstructed genealogies. As sequence data have become more abundant, these approaches are beginning to be used on populations undergoing rapid and rather complex dynamics. In such cases, the simple demographic models that current phylodynamic methods employ can be limiting. First, these models are not ideal for yielding biological insight into the processes that drive the dynamics of the populations of interest. Second, these models differ in form from mechanistic and often stochastic population dynamic models that are currently widely used when fitting models to time series data. As such, their use does not allow for both genealogical data and time series data to be considered in tandem when conducting inference. Here, we present a flexible statistical framework for phylodynamic inference that goes beyond these current limitations. The framework we present employs a recently developed method known as particle MCMC to fit stochastic, nonlinear mechanistic models for complex population dynamics to gene genealogies and time series data in a Bayesian framework. We demonstrate our approach using a nonlinear Susceptible-Infected-Recovered (SIR) model for the transmission dynamics of an infectious disease and show through simulations that it provides accurate estimates of past disease dynamics and key epidemiological parameters from genealogies with or without accompanying time series data., Author Summary Reliable information about the demographic history of populations is important to both population biologists and epidemiologists, but is often absent or unreliable. There has therefore been great interest in developing statistical methods for inferring past population dynamics from gene genealogies reconstructed from molecular sequences. These “phylodynamic” methods take advantage of the fact that changes in population size can dramatically affect the shape of genealogies, making it possible to infer past changes in population size from a genealogy. However, in order for past population dynamics to be inferred, a demographic model must be specified. Current methods use demographic models that are often too simple for populations undergoing complex dynamics and generally do not allow for the parameters influencing the population dynamics to be estimated. We show how current phylodynamic methods can be extended to allow a much wider class of models to be fit to genealogies and illustrate our approach using an epidemiological model for the transmission of an infectious disease.

The synthesis of four new TTF derivatives bearing phthalimides and fluorinated alkyl moieties as potential ambipolar semiconductors is described. The presence of such electron-withdrawing groups permits the stabilization of the energy of HOMO and LUMO orbitals. The solid-state structures of these novel molecules have been characterized by X-ray diffraction techniques. The potential of these materials as hole and electron conductors has been estimated under theoretical considerations by evaluating the position of the frontier energy levels as well as their charge carrier mobilities. Preparation of solution-processed single crystal organic field-effect transistors (OFETs) has resulted in hole mobilities of up to 0.33 cm2 V-1 s -1 for compound 1. On the other hand, electrical time of flight (EToF) measurements on single crystals of compound 3 demonstrated ambipolar transport, reaching very high mobility values around 2.0 cm2 V -1 s-1 for both types of charges. © 2010 American Chemical Society., The research leading to these results has received funding from the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 212311 of the ONE-P project, Marie Curie EST FuMASSEC, DGI, Spain (contracts CTQ2006-06333/BQU and CTQ2010-195011/BQU), the Generalitat de Catalunya, (2009SGR00516) and the program “Juan de la Cierva”(MICINN). We also thank Stefan T. Bromley for his advice regarding the DFT calculations and CESGA for the use of their computational resources. The work in Mons is also supported by the European ONE-P project as well as by the Belgian National Fund for Scientific Research (FNRS). Y.O. and J.C. are FNRS research fellows.

25 pages; We present a parallel algorithm for solving backward stochastic differential equations (BSDEs in short) which are very useful theoretic tools to deal with many financial problems ranging from option pricing option to risk management. Our algorithm based on Gobet and Labart (2010) exploits the link between BSDEs and non linear partial differential equations (PDEs in short) and hence enables to solve high dimensional non linear PDEs. In this work, we apply it to the pricing and hedging of American options in high dimensional local volatility models, which remains very computationally demanding. We have tested our algorithm up to dimension 10 on a cluster of 512 CPUs and we obtained linear speedups which proves the scalability of our implementation

In this paper, we use Monte Carlo methods to study the small sample properties of the classical maximum likelihood (ML) estimator in artificial samples generated by the New-Keynesian open economy DSGE model estimated by Adolfson et al. (2008) with Bayesian techniques. While asymptotic identification tests show that some of the parameters are weakly identified in the model and by the set of observable variables we consider, we document that ML is unbiased and has low MSE for many key parameters if a suitable set of observable variables are included in the estimation. These findings suggest that we can learn a lot about many of the parameters by confronting the model with data, and hence stand in sharp contrast to the conclusions drawn by Canova and Sala (2009) and Iskrev (2008). Encouraged by our results, we estimate the model using classical techniques on actual data, where we use a new simulation based approach to compute the uncertainty bands for the parameters. From a classical viewpoint, ML estimation leads to a significant improvement in fit relative to the log-likelihood computed with the Bayesian posterior median parameters, but at the expense of some the ML estimates being implausible from a microeconomic viewpoint. We interpret these results to imply that the model at hand suffers from a substantial degree of model misspecification. This interpretation is supported by the DSGE-VAR analysis in Adolfson et al. (2008). Accordingly, we conclude that problems with model misspecification, and not primarily weak identification, is the main challenge ahead in developing quantitative macromodels for policy analysis.

We propose several new methods to calculate threshold and power for Quantitative Trait Locus (QTL) detection. They are based on asymptotic theoretical results presented in Rabier et al. (2009) . The asymptotic validity is checked by simulations. The methods proposed are fast and easy to implement. A comparison of power between a multiple testing procedure and a global test has been realized, showing far better performances of the global test for the detection of a QTL.

Safety and reliability requirements on the one side and short development cycles, low costs and lightweight design on the other side are two competing aspects of truck engineering. For safety critical components essentially no failures can be tolerated within the target mileage of a truck. For other components the goals are to stay below certain predefined failure rates. Reducing weight or cost of structures often also reduces strength and reliability. The requirements on the strength, however, strongly depend on the loads in actual customer usage. Without sufficient knowledge of these loads one needs large safety factors, limiting possible weight or cost reduction potentials. There are a lot of different quantities influencing the loads acting on the vehicle in actual usage. These ‘influencing quantities’ are, for example, the road quality, the driver, traffic conditions, the mission (long haulage, distribution or construction site), and the geographic region. Thus there is a need for statistical methods to model the load distribution with all its variability, which in turn can be used for the derivation of testing specifications.

Technical report TR09-13. This article presents a survey of reinforcement learning algorithms for Markov Decision Processes (MDP). In the first half of the article, the problem of value estimation is considered. Here we start by describing the idea of bootstrapping and temporal difference learning. Next, we compare incremental and batch algorithmic variants and discuss the impact of the choice of the function approximation method on the success of learning. In the second half, we describe methods that target the problem of learning to control an MDP. Here online and active learning are discussed first, followed by a description of direct and actor-critic methods. | TRID-ID TR09-13

International audience; The fractal mobile-immobile model (MIM) is intermediate between advection-dispersion (ADE) and fractal Fokker-Planck (FFKPE) equations. It involves two time derivatives, whose orders are 1 and y (between 0 and 1) on the lefthand side, whereas all mentioned equations have identical right-hand sides. The fractal MIM model accounts for non-Fickian effects that occur when tracers spread in media because of through-flow, and can get trapped by immobile sites. The solid matrix of a porous material may contain such sites, so that non-Fickian spread is actually observed. Within the context of the fractal MIM model, we present a mapping that allows the computation of fluxes on the basis of the density of spreading particles. The mapping behaves as Fickian flux at early times, and tends to a fractional derivative at late times. By means of this mapping, we recast the fractal MIM model into conservative form, which is suitable to deal with sources and bounded domains. Mathematical proofs are illustrated by comparing the discretized fractal p.d.e. with Monte Carlo simulations