Search

The general field of the thesis is the sensitivity analysis of black-box functions. Sensitivity analysis studies how the variation of the output can be apportioned to the variation of input sources. It is an important tool in the construction, analysis, and optimization of computer experiments.The total interaction index is presented, which can be used for the screening of interactions. Several variance-based estimation methods are suggested. Their properties are analyzed theoretically as well as on simulations.A further chapter concerns the sensitivity analysis for models that can take functions as input variables and return a scalar value as output. A very economical sequential approach is presented, which not only discovers the sensitivity of those functional variables as a whole but identifies relevant regions in the functional domain.As a third concept, support index functions, functions of sensitivity indices over the input distribution support, are suggested.Finally, all three methods are successfully applied in the sensitivity analysis of sheet metal forming models., Le domaine général de la thèse est l’analyse de sensibilité de fonctions boîte noire. L’analyse de sensibilité étudie comment la variation d’une sortie peut être reliée à la variation des entrées. C’est un outil important dans la construction, l’analyse et l’optimisation des expériences numériques (computer experiments).Nous présentons tout d’abord l’indice d’interaction total, qui est utile pour le criblage d’interactions. Plusieurs méthodes d’estimation sont proposées. Leurs propriétés sont étudiées au plan théorique et par des simulations.Le chapitre suivant concerne l’analyse de sensibilité pour des modèles avec des entrées fonctionnelles et une sortie scalaire. Une approche séquentielle très économique est présentée, qui permet non seulement de retrouver la sensibilité de entrées fonctionnelles globalement, mais aussi d’identifier les régions d’intérêt dans leur domaine de définition.Un troisième concept est proposé, les support index functions, mesurant la sensibilité d’une entrée sur tout le support de sa loi de probabilité.Finalement les trois méthodes sont appliquées avec succès à l’analyse de sensibilité de modèles d’emboutissage.

The general field of the thesis is the sensitivity analysis of black-box functions. Sensitivity analysis studies how the variation of the output can be apportioned to the variation of input sources. It is an important tool in the construction, analysis, and optimization of computer experiments. The total interaction index is presented, which can be used for the screening of interactions. Several variance-based estimation methods are suggested. Their properties are analyzed theoretically as well as on simulations. A further chapter concerns the sensitivity analysis for models that can take functions as input variables and return a scalar value as output. A very economical sequential approach is presented, which not only discovers the sensitivity of those functional variables as a whole but identifies relevant regions in the functional domain. As a third concept, support index functions, functions of sensitivity indices over the input distribution support, are suggested. Finally, all three methods are successfully applied in the sensitivity analysis of sheet metal forming models.

Global sensitivity analysis is used to quantify the influence of input variables on a numerical model output. Sobol' indices are now classical sensitivity measures. However their estimation requires a large number of model evaluations, especially when interaction effects are of interest. Derivative-based global sensitivity measures (DGSM) have recently shown their efficiency for the identification of non-influential inputs. In this paper, we define crossed DGSM, based on second-order derivatives of model output. By using a L2- Poincaré inequality, we provide a crossed-DGSM based maximal bound for the superset importance (i.e. total Sobol' indices of an interaction between two inputs). In order to apply this result, we discuss how to estimate the Poincaré constant for various probability distributions. Several analytical and numerical tests show the performance of the bound and allow to develop a generic strategy for interaction screening

Sensitivity analysis aims at exploring which of a number of variables have an impact on a certain response. Not only are the individual variables of interest but also whether they interact or not. By analogy with the total sensitivity index, used to detect the most influential variables, a screening of interactions can be done efficiently with the so-called total interaction index (TII), defined as the superset importance of a pair of variables. Our aim is to investigate the TII, with a focus on statistical inference. At the theoretical level, we derive its connection to total and closed sensitivity indices. We present several estimation methods and prove the asymptotical efficiency of the Liu and Owen estimator. We also address the question of estimating the full set of TIIs, with a given budget of function evaluations. We observe that with the pick-and-freeze method the full set of TIIs can be estimated at a linear cost with respect to the problem dimension. The different estimators are then compared empirically. Finally, an application is given aiming at discovering a block-additive structure of a function, where no prior knowledge either about the interaction structure or about the blocks is available.

INGARCH models for time series of counts arising, e.g., in epidemiology assume the observations to be Poisson distributed conditionally on the past, with the conditional mean being an affinelinear function of the previous observations and the previous conditional means. We model outliers within such processes, assuming that we observe a contaminated process with additive Poisson distributed contamination, affecting each observation with a small probability. Our particular concern are additive outliers, which do not enter the dynamics of the process and can represent measurement artifacts and other singular events influencing a single observation. Such outliers are difficult to handle within a non-Bayesian framework since the uncontaminated values entering the dynamics of the process at contaminated time points are unobserved. We propose a Bayesian approach to outlier modeling in INGARCH processes, approximating the posterior distribution of the model parameters by application of a componentwise Metropolis- Hastings algorithm. Analyzing real and simulated data sets, we find Bayesian outlier detection with non-informative priors to work well if there are some outliers in the data., Discussion Paper / SFB 823;26/2012

http://mucm.ac.uk/UCM2012/Forms/Downloads/Posters/Fruth.pdf; International audience