▼ Structured probabilistic inference has shown to be useful in modeling complex latent structures of data. One successful way in which this technique has been applied is in the discovery of latent topical structures of text data, which is usually referred to as topic modeling. With the recent popularity of mobile devices and social networking, we can now easily acquire text data attached to meta information, such as geo-spatial coordinates and time stamps. This metadata can provide rich and accurate information that is helpful in answering many research questions related to spatial and temporal reasoning. However, such data must be treated differently from text data. For example, spatial data is usually organized in terms of a two dimensional region while temporal information can exhibit periodicities. While some work existing in the topic modeling community that utilizes some of the meta information, these models largely focused on incorporating metadata into text analysis, rather than providing models that make full use of the joint distribution of metainformation and text. In this thesis, I propose the event detection problem, which is a multidimensional latent clustering problem on spatial, temporal and topical data. I start with a simple parametric model to discover independent events using geo-tagged Twitter data. The model is then improved toward two directions. First, I augmented the model using Recurrent Chinese Restaurant Process (RCRP) to discover events that are dynamic in nature. Second, I studied a model that can detect events using data from multiple media sources. I studied the characteristics of different media in terms of reported event times and linguistic patterns. The approaches studied in this thesis are largely based on Bayesian nonparametric methods to deal with steaming data and unpredictable number of clusters. The research will not only serve the event detection problem itself but also shed light into a more general structured clustering problem in spatial, temporal and textual data.

▼ 

Os serviços de saneamento básico exercem um papel importante no bem-estar da população, pois a falta de abastecimento de água e esgotamento sanitário contribui de forma negativa para a saúde das pessoas, meio ambiente e ainda aumenta os gastos públicos. O objetivo principal deste trabalho foi estimar a disposição a pagar (DAP) por melhorias nos serviços de saneamento básico na cidade de Palmas-TO e identificar quais os fatores que afetam essa disposição. Especificamente, buscou-se: estimar a DAP utilizando os modelos tradicionais de escolha binária, assim como, modelos semi-paramétricos e não-paramétricos; comparar os valores da DAP estimados pelos métodos; e comparar os resultados obtidos da disposição agregada com o valor do investimento feito no setor de saneamento na cidade de Palmas-TO. A estimação da disposição a pagar por melhorias do serviço de saneamento básico foi obtida pelo método de avaliação contingente através da abordagem de Hanemann (1984) com a forma de eliciação fechada (close-ended). Os resultados econométricos obtidos mostraram que os votos de protesto prejudicaram as estimativas da DAP, visto que, diminuíram a precisão das mesmas em todos os modelos estimados. Portanto, os modelos que excluíram os votos de protesto apresentaram os melhores resultados. A disposição a pagar média pela cesta de serviços de saneamento básico foi estimada em R23,08 pelo modelo logit, em R19,94 pelo modelo semi-paramétrico e em R22,48 pelo modelo não- paramétrico. Logo, a medida mais conservadora para a disposição a pagar foi obtida pelo modelo semi-paramétrico. As variáveis que afetaram a probabilidade do indivíduo aceitar pagar alguma quantia a mais pelo bem em questão foram a renda familiar e o nível de escolaridade (positivamente) e o valor do lance oferecido (negativamente), o que leva a crer que o serviço de saneamento básico pode ser considerado um bem normal e comum. A partir da agregação da disposição a pagar média estimada, o ganho de bem-estar associado à melhoria na quantidade/qualidade do serviço de saneamento básico varia entre R 8.386.046,16 e R 9.706.617,12. Esses valores estimados foram menores que o total investido no setor nos últimos anos que foi de R 74.079.081,89. Assim, a DAP agregada indica que a capacidade de pagamento das pessoas é inferior ao total investido no setor, e com isso, pode-se inferir que elevados investimentos poderão ser realizados na área de infra-estrutura, principalmente na coleta e tratamento de esgotos para que uma maior parcela da população tenha acesso a estes serviços. Portanto, os benefícios diretos advindos da disposição a pagar dos usuários desses serviços e os custos evitados com a poluição hídrica e as doenças de veiculação hídrica, redutíveis através da implantação de projetos, dessa natureza, são suficientemente fortes a ponto de garantir a viabilidade econômica e social.

The services of basic sanitation have crucial importance in relation to population welfare because the absence of this kind of services contribute negatively to population healthy,…

Advisors/Committee Members: João Eustáquio de Lima, Walter Belluzzo Junior, Maurinho Luiz dos Santos, José Gustavo Feres, Leonardo Bornacki de Mattos, Wilson da Cruz Vieira.

▼ The past decades, the increasing availability of data has paved the way for a new, data-driven generation of models. This research proposes a non-parametric Bayesian network (NPBN) to model hydrologic processes. The Bayesian network (BN) is a directed, acyclic graph in which the variables are represented by the nodes, and the conditional probability distribution between variable pairs is represented by the arcs. NPBNs are computationally less expensive than many conceptual hydrologic models and are sufficiently flexible to be able to handle different continuous data sources. The goal of this thesis is to make an NPBN for a lowland catchment and test its performance. The case study concerns the catchment of the Vledder, Wapserveense and Steenwijker Aa. This catchment makes this research the first one in which an NPBN is comprehensively implemented for (1.) a single catchment in which the catchment processes are modelled, (2.) a Dutch catchment, and (3.) a lowland, partially managed, catchment. For the BN model, seven hydro-meteorological variables have been selected for the model, complemented by the target variable, which is the monthly maximum daily average discharge (MMDAD). The aim of the BN is to be able to accurately predict the MMDAD, and the Kling-Gupta efficiency (KGE) acts as a performance indicator by which to optimize the BN’s parameters. For this thesis, the Gaussian copula was selected to be implemented for all variable combinations in the BN, because this type allows for the use of the multivariate normal distribution to calculate a conditioned network. The fit of the Gaussian copula to the data is tested in this thesis. This method is far more convenient than the alternative called the vine-copula method and most likely gives a better fit than the other alternatives. Three distributions are compared to model the marginal distributions, of which the Gaussian mixture model has been selected. This function extrapolated too little, so a novel alteration function has been proposed to shift the predictions. Several other parameters in the BN have been analysed as well. A sensitivity analysis has been performed to understand what influences of artificial errors would be. In general, random errors have a low influence on the prediction of the model, whereas new systematic errors have a larger influence. Criteria for a practical, well-performing BN have been presented and a strategy to create such a model that satisfies these criteria has been assembled. This strategy left the selection of some connection implementations up for interpretation. The chosen implementation has been decided based on which implementation produced the best predictions of the relevant variable within the network. The final model gave a median, k-fold tested KGE of 0.73 when predicting the MMDAD. It is also analysed how well the MMDAD is predicted if not all other variables are fixed. Another novelty is that a BN model is benchmarked against a SOBEK model, a neural network, and a multiple linear regression model. Compared to these… Advisors/Committee Members: Morales Napoles, Oswaldo (mentor), Ragno, Elisa (mentor), Hrachowitz, Markus (mentor), Dekens, Bart (mentor), Hoch, Jannis (mentor), Delft University of Technology (degree granting institution).

▼  A common research interest in medical, biological, and engineering research is determining whether certain independent variables are correlated with the survival or failure times. Standard statistical techniques cannot usually be applied for failure-time data due to the lack of complete data or in other word, due to censoring. From a statistical perspective, the study of time to event data is even more challenging when further complexities such as high dimensionality or multivariablity is added to the model. In this dissertation, we consider the predicating patient survival from proteomic profile of patient serum using matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) data of non-small cell lung cancer patients. Due to much larger dimension of features in a mass spectrum compared to the study sample size, traditional linear regression modeling of survival times with high number of proteomic features is not feasible. Hence, we consider latent factor and regularized/penalized methods for fitting such models in order to predict patient survival from the mass spectrometry features. Extensive numerical studies involving both simulated as well as real mass spectrometry data are used to compare four popular regression methods, namely, partial least squares (PLS), sparse partial least square (SPLS), least absolute shrinkage and selection operator (LASSO) and elastic net regularization, on processed spectra. Right censoring is handled through a residual based multiple imputation. Overall, more complex methods such as the elastic net and SPLS result in better performances provided the operational parameters are chosen carefully via cross validation. For survival time prediction, we recommend using the elastic net based on a selected set of features. As a type of multivariate survival data, multistate models have a wide range of applications. Most of the existing regression approaches to analyze such data are based on parametric and semi-parametric procedures in which one should rely on specific model structures. In this dissertation, we construct non-parametric regression estimators of a number of temporal functions in a multistate system based on a univariate continuous baseline covariate. These estimators include state occupation probabilities, state entry, exit and waiting (sojourn) times distribution functions of a general progressive (e.g. acyclic) multistate model. The data are subject to right censoring and the censoring mechanism is explainable by observable covariates that could be time dependent. The resulting estimators are valid even if the multistate process is non-Markov. The performance of the estimators is studied using a detailed simulation. We illustrate our estimators using a data set on bone marrow transplant patients. Finally, some extension of the proposed methods to more general case with multivariate covariates are presented along with plans for future developments. Advisors/Committee Members: Datta, Somnath.

▼ 

Computer and Information Science

Sparse regressions based on constraining the L1-norm of the coefficients became popular due to their ability to handle high dimensional data unlike the regular regressions which suffer from overfitting and model identifiability issues especially when sample size is small. They are often the method of choice in many fields of science and engineering for simultaneously selecting covariates and fitting parsimonious linear models that are better generalizable and easily interpretable. However, significant challenges may be posed by the need to accommodate extremes and other domain constraints such as dynamical relations among variables, spatial and temporal constraints, need to provide uncertainty estimates and feature correlations, among others. We adopted a hierarchical Bayesian version of the sparse regression framework and exploited its inherent flexibility to accommodate the constraints. We applied sparse regression for the feature selection problem of statistical downscaling of the climate variables with particular focus on their extremes. This is important for many impact studies where the climate change information is required at a spatial scale much finer than that provided by the global or regional climate models. Characterizing the dependence of extremes on covariates can help in identification of plausible causal drivers and inform extremes downscaling. We propose a general-purpose sparse Bayesian framework for covariate discovery that accommodates the non-Gaussian distribution of extremes within a hierarchical Bayesian sparse regression model. We obtain posteriors over regression coefficients, which indicate dependence of extremes on the corresponding covariates and provide uncertainty estimates, using a variational Bayes approximation. The method is applied for selecting informative atmospheric covariates at multiple spatial scales as well as indices of large scale circulation and global warming related to frequency of precipitation extremes over continental United States. Our results confirm the dependence relations that may be expected from known precipitation physics and generates novel insights which can inform physical understanding. We plan to extend our model to discover covariates for extreme intensity in future. We further extend our framework to handle the dynamic relationship among the climate variables using a nonparametric Bayesian mixture of sparse regression models based on Dirichlet Process (DP). The extended model can achieve simultaneous clustering and discovery of covariates within each cluster. Moreover, the a priori knowledge about association between pairs of data-points is incorporated in the model through must-link constraints on a Markov Random Field (MRF) prior. A scalable and efficient variational Bayes approach is developed to infer posteriors on regression coefficients and cluster variables.

Temple University – Theses

Advisors/Committee Members: Obradovic, Zoran, Ganguly, Auroop R.;, Vucetic, Slobodan, Latecki, Longin;.

▼ 

ENGLISH ABSTRACT : When trespassers target businesses and homes, outdoor spaces are typically the first point of illegal entry. Camera systems can help secure these environments, but typically many cameras are needed to cover large areas. In practice most camera systems are only used to review events after they have happened. It is however possible to do much more, and we explore the paradigm of active monitoring where cameras detect trespassers and give visual verification of the alarm. This detection needs to be resilient to weather effects and other environmental noise, while running in real-time on high resolution video sequences. Our goal is to replace multiple static cameras with a single Pan-Tilt-Zoom (PTZ) camera that can monitor expansive terrain. These cameras can survey, detect and zoom in on objects of interest. We develop and implement a robust, real-time algorithm based on an interaction framework between an illumination invariant and a color based background model. We also developed and implemented a novel technique where we use optical flow motion vectors to determine the size and shape of the spatial Gaussian kernels in non parametric models. Although computationally more expensive we demonstrate these more sophisticated models can be more robust. These ideas are adapted for PTZ cameras, exploiting their pan, tilt and zoom capabilities. We apply background modeling on panorama images that inform PTZ camera movement. By this we discuss the construction of a system that secures perimeters using zooming camera analytics.

AFRIKAANSE OPSOMMING : Wanneer oortreders besighede en huise onwettig betree, is die buite omgewing tipies die eerste punt van toegang. Kameras kan aansienlik help om hierdie omgewings te beveilig, maar mens kort groot aantal kameras as jy groot spasies wil monitor. Meeste kamera sisteme word ook net gebruik om die gebeurtenisse na die tyd te besigtig. Ons ondersoek die moontlikheid van aktiewe monitering wat oortreders intyds kan identifiseer. Hierdie deteksie moet robuus wees teen weersomstandighede en ander omgewingseffekte, en moet intyds hardloop op hoë resolusie video sekwensies. Die uiteindelike doel is om 'n groot aantal statiese kameras te vervang met een Pan-Tilt-Zoom (PTZ) kamera wat 'n groot terrein kan dek. Hierdie kameras kan rondkyk en in zoem op relevante voorwerpe. Ons begin deur intydse robuuste monitering te verbeter gebaseer op die kombinasie van illuminasie invariante en kleur gebasseerde modelle. Ons stel dan 'n nuwe tegniek voor om optiese vloei vektore te gebruik om Gaussiese verspreidings se vorm en orientasie te bepaal vir nie-parametriese modelle. Ons pas hierdie idees aan vir gebruik op PTZ kameras. Dit behels onder andere die ontwikkeling van panoramiese agtergrond modelle. Ons gebruik die PTZ eienskappe om die opgespoorde indringers in detail te monitor. Van hier kan ons die PTZ kamera se beweging beheer volgens enige deteksies.

Advisors/Committee Members: Herbst, Barend Mattheus, Crnojevic, Vladimir, Brink, Willie, Stellenbosch University. Faculty of Science. Department of Applied Mathematics.

▼ This master thesis aims to gain new empirical insights into longitudinal driving behavior by means of the enumeration of a new hybrid car-following (CF) model which combines parametric and non parametric formulation. On one hand, the model, which predicts the drivers acceleration given a set of variables, benefits from innovative machine learning techniques such as Gaussian process regression (GPR) to make predictions when there exist correlation between new input and the training dataset. On the other hand, it uses existent traditional parametric CF models to predict acceleration when no similar situations are found in the training dataset. This formulation guarantees a complete and continues model and deals with the challenges of new available types of dataset in the transport field: noisy and incomplete yet with large amount of data. Multiple models have been trained using the Optimal Velocity Model (OVM) as a basis parametric model and a dataset collected in the PPA project in Amsterdam by traffic radar detection in stop and go traffic conditions. The other main innovation of this thesis is that variables rarely included in any CF model such as the status and the distance of drivers to the traffic light are also analyzed. Results show that the GPR model formulation is robust as the model performs better than OVM alone according to the main KPI, but still collisions occasionally occur. Moreover, results depict that traffic light status actively influences driver behavior. Overall, this thesis gives insights into new powerful mathematical techniques that can be applied to describe longitudinal driving behavior or any modeled process. Advisors/Committee Members: Hoogendoorn, Serge (mentor), Knoop, Victor (graduation committee), Steenbakkers, Jeroen (graduation committee), Alonso Mora, Javier (graduation committee), Delft University of Technology (degree granting institution).

▼ Although the effect of interest rate stochasticity can safely be ignored for short-dated exchange traded derivatives, this is not the case for the kind of long-dated over-the-counter derivatives often used by insurance companies, fund managers, and other financial institutions. We therefore extend existing derivatives pricing techniques, specifically local volatility, stochastic volatility, and model-free pricing, to the case of non-deterministic interest rates. We also present empirical examples to highlight the potentially significant effect on long-term contracts.

▼ 

La radiothérapie externe (EBRT en anglais pour External Beam Radiotherapy) est l'un des traitements référence du cancer de prostate. Les objectifs de la radiothérapie sont, premièrement, de délivrer une haute dose de radiations dans la cible tumorale (prostate et vésicules séminales) afin d'assurer un contrôle local de la maladie et, deuxièmement, d'épargner les organes à risque voisins (principalement le rectum et la vessie) afin de limiter les effets secondaires. Des modèles de probabilité de complication des tissus sains (NTCP en anglais pour Normal Tissue Complication Probability) sont nécessaires pour estimer sur les risques de présenter des effets secondaires au traitement. Dans le contexte de la radiothérapie externe, les objectifs de cette thèse étaient d'identifier des paramètres prédictifs de complications rectales et vésicales secondaires au traitement; de développer de nouveaux modèles NTCP permettant l'intégration de paramètres dosimétriques et de paramètres propres aux patients; de comparer les capacités prédictives de ces nouveaux modèles à celles des modèles classiques et de développer de nouvelles méthodologies d'identification de motifs de dose corrélés à l'apparition de complications. Une importante base de données de patients traités par radiothérapie conformationnelle, construite à partir de plusieurs études cliniques prospectives françaises, a été utilisée pour ces travaux. Dans un premier temps, la fréquence des symptômes gastro-Intestinaux et génito-Urinaires a été décrite par une estimation non paramétrique de Kaplan-Meier. Des prédicteurs de complications gastro-Intestinales et génito-Urinaires ont été identifiés via une autre approche classique : la régression logistique. Les modèles de régression logistique ont ensuite été utilisés dans la construction de nomogrammes, outils graphiques permettant aux cliniciens d'évaluer rapidement le risque de complication associé à un traitement et d'informer les patients. Nous avons proposé l'utilisation de la méthode d'apprentissage de machine des forêts aléatoires (RF en anglais pour Random Forests) pour estimer le risque de complications. Les performances de ce modèle incluant des paramètres cliniques et patients, surpassent celles des modèle NTCP de Lyman-Kutcher-Burman (LKB) et de la régression logistique. Enfin, la dose 3D a été étudiée. Une méthode de décomposition en valeurs populationnelles (PVD en anglais pour Population Value Decomposition) en 2D a été généralisée au cas tensoriel et appliquée à l'analyse d'image 3D. L'application de cette méthode à une analyse de population a été menée afin d'extraire un motif de dose corrélée à l'apparition de complication après EBRT. Nous avons également développé un modèle non paramétrique d'effets mixtes spatio-Temporels pour l'analyse de population d'images tridimensionnelles afin d'identifier une région anatomique dans laquelle la dose pourrait être corrélée à l'apparition d'effets secondaires.

External beam radiotherapy (EBRT) is one of the cornerstones of prostate cancer treatment. The objectives…

Advisors/Committee Members: Crevoisier, Renaud de (thesis director), Correa Morales, Juan Carlos (thesis director).

▼ 

Daktaro disertacijos tyrimo objektai yra rezervuotos sistemos ir degradaciniai modeliai. Norint užtikrinti svarbių sistemos elementų aukštą patikimumą, naudojami jų rezerviniai elementai, kurie gali būti įjungiami sugedus šiems pagrindiniams elementams. Rezerviniai elementai gali funkcionuoti skirtinguose režimuose: „karštame“, „šaltame“ arba „šiltame“. Disertacijoje yra nagrinėjamos sistemos su „šiltai“ rezervuotais elementais. Darbe suformuluojama rezervinio elemento „sklandaus įjungimo“ hipotezė ir konstruojami statistiniai kriterijai šiai hipotezei tikrinti. Nagrinėjami neparametrinio ir parametrinio taškinio bei intervalinio vertinimo uždaviniai. Disertacijoje nagrinėjami pakankamai bendri degradacijos modeliai, kurie aprašo elementų gedimų intensyvumą kaip funkciją kiek naudojamų apkrovų, tiek ir degradacijos lygio, kuri savo ruožtu modeliuojama naudojant stochastinius procesus.

In doctoral thesis redundant systems and degradation models are considered. To ensure high reliability of important elements of the system, the stand-by units can be used. These units are commuted and operate instead of the main failed unit. The stand-by units can function in the different conditions: “hot”, “cold” or “warm” reserving. In the thesis systems with “warm” stand-by units are analyzed. Hypotheses of smooth commuting are formulated and goodness-of-fit tests for these hypotheses are constructed. Nonparametric and parametric point and interval estimation procedures are given. Modeling and statistical estimation of reliability of systems from failure time and degradation data are considered.

Advisors/Committee Members: Bagdonavičius, Vilijandas (Doctoral dissertation supervisor), Leipus, Remigijus (Doctoral dissertation committee chair), Bikelis, Algimantas (Doctoral dissertation committee member), Grigelionis, Bronius (Doctoral dissertation committee member), Saulis, Leonas (Doctoral dissertation committee member), Levulienė, Rūta (Doctoral dissertation committee member), Radavičius, Marijus (Doctoral dissertation opponent), Kazakevičius, Vytautas (Doctoral dissertation opponent).

▼ 

In doctoral thesis redundant systems and degradation models are considered. To ensure high reliability of important elements of the system, the stand-by units can be used. These units are commuted and operate instead of the main failed unit. The stand-by units can function in the different conditions: “hot”, “cold” or “warm” reserving. In the thesis systems with “warm” stand-by units are analyzed. Hypotheses of smooth commuting are formulated and goodness-of-fit tests for these hypotheses are constructed. Nonparametric and parametric point and interval estimation procedures are given. Modeling and statistical estimation of reliability of systems from failure time and degradation data are considered.

Daktaro disertacijos tyrimo objektai yra rezervuotos sistemos ir degradaciniai modeliai. Norint užtikrinti svarbių sistemos elementų aukštą patikimumą, naudojami jų rezerviniai elementai, kurie gali būti įjungiami sugedus šiems pagrindiniams elementams. Rezerviniai elementai gali funkcionuoti skirtinguose režimuose: „karštame“, „šaltame“ arba „šiltame“. Disertacijoje yra nagrinėjamos sistemos su „šiltai“ rezervuotais elementais. Darbe suformuluojama rezervinio elemento „sklandaus įjungimo“ hipotezė ir konstruojami statistiniai kriterijai šiai hipotezei tikrinti. Nagrinėjami neparametrinio ir parametrinio taškinio bei intervalinio vertinimo uždaviniai. Disertacijoje nagrinėjami pakankamai bendri degradacijos modeliai, kurie aprašo elementų gedimų intensyvumą kaip funkciją kiek naudojamų apkrovų, tiek ir degradacijos lygio, kuri savo ruožtu modeliuojama naudojant stochastinius procesus.

Advisors/Committee Members: Bagdonavičius, Vilijandas (Doctoral dissertation supervisor), Leipus, Remigijus (Doctoral dissertation committee chair), Bikelis, Algimantas (Doctoral dissertation committee member), Grigelionis, Bronius (Doctoral dissertation committee member), Saulis, Leonas (Doctoral dissertation committee member), Levulienė, Rūta (Doctoral dissertation committee member), Radavičius, Marijus (Doctoral dissertation opponent), Kazakevičius, Vytautas (Doctoral dissertation opponent).

▼ 

Este trabalho apresenta as principais causas para a mortalidade das empresas paulistas criadas entre 2003 e 2007 a partir de base de dados cedida pelo SEBRAE-SP para o desenvolvimento dessa pesquisa. A amostra final, construída a partir de dados disponibilizados pela primeira vez para estudos desta natureza, contou com 662 empresas e 33 variáveis coletadas por meio de questionário aplicado diretamente às próprias empresas. A análise consistiu no teste de modelos econométricos, baseados na literatura dos modelos de duração, de forma a traduzir quais fatores são mais críticos para a sobrevivência das empresas a ponto de distingui-las em dois grupos: o das empresas vencedoras, cuja longevidade está pautada em ações que promovem ganhos de produtividade e eficiência, e aquelas desprovidas dessas ações e que muito provavelmente deixarão o mercado. Os três tipos de modelos abordados neste trabalho - não paramétrico, semi-paramétrico (riscos proporcionais) e paramétrico - apresentaram resultados similares, sendo que na abordagem de riscos proporcionais os resultados foram segmentados por tamanho e setor de atuação das empresas. Para as micro empresas, a idade do empreendedor e a iniciativa em investir na qualificação da mão de obra dos funcionários mostraram-se importantes mitigadores do risco de falha desse grupo de empresa, enquanto que para as pequenas empresas, a inovação em processos e a elaboração de um plano de negócios se destacaram dentre o conjunto de variáveis. Entre empresas dos setores de comércio e serviços, as empresas do primeiro grupo que faziam o acompanhamento das finanças (fluxo de caixa) apresentaram menor risco de falhar. Para aquelas do setor de serviços, a idade do empreendedor, o investimento em qualificação dos funcionários e o tamanho da empresa ao nascer foram importantes para reduzir o risco de falha no tempo. Outro resultado encontrado, por meio do modelo paramétrico utilizando distribuição Weibull, foi que o risco de a empresa deixar o mercado mostrou-se crescente, pelo menos nos cinco primeiros anos de existência da empresa. Entretanto, esse resultado não deve ser generalizado para períodos de tempo maiores que cinco anos.

This thesis presents the main results that determined the bankruptcy of enterprises located in the São Paulo State from 2003 to 2007. The models used in this work were possible due to the partnership with SEBRAE, Small Business Service Supporting, located in the State of São Paulo. This institution provided the data basis for this research and its final version was compound by 662 enterprises and 33 variables, which were collected from a survey done by SEBRAE and the related enterprise. For first time available for research like this The research was supported by econometrics models, more precisely duration models, which identified the most important factors regarding enterprises survival. Two enterprise groups were distinguished: that one that will survive and grow and another will fail. In this work, three models were used: parametric, non-parametric and…

Advisors/Committee Members: Helio Nogueira da Cruz, Carlos Jose Caetano Bacha, Milton de Abreu Campanario, Sérgio Kannebley Júnior, Carlos Antonio Luque.

▼ Traditionally visual cells have been characterized using their responses to artificial stimuli by simple parametric models. However, recent investigations show that visual cells adapt to the statistical properties of the stimuli used to probe them. Thus, to characterize visual cells in their natural operating conditions, it is important to use naturalistic stimuli. Simple parametric models are designed for specific classes of cells, making assumptions about their response properties. But, if these assumptions do not match the cell response properties, the interpretation of the estimated model parameters is questionable. An alternative is to use generic non-parametric models that can characterize a broad range of cell classes.; This thesis contains technical and scientific contributions. Technically, we develop methods to estimate generic non-parametric models of visual cells from their responses to arbitrary, including natural, stimuli. In the first part of this thesis, we introduce the Volterra Relevant Space Technique (VRST), that allows the estimation of spatial Volterra models of visual cells from their responses to natural stimuli. Disregarding temporal properties of the response generation mechanism for the estimation of spatial Volterra models is a good first approximation. However, in most conditions responses of visual cells are not spatial, but spatio temporal. So, in the second part of this dissertation we build the extended Projection Pursuit Regression (ePPR) algorithm, that estimates a very general model for the characterization of visual cells in space and time. The generality of the ePPR model reveals differences in response properties of cortical cells to natural and random stimuli that had not been observed with existing models. Thus, scientifically this thesis shows that using natural stimuli for the characterization of visual cells is relevant to understand natural vision. Advisors/Committee Members: Grzywacz, Norberto M.Mendel, Jerry M. (Committee Chair), Marmarelis, Vasilis Z. (Committee Member), Hirsch, Judith. A. (Committee Member).

▼ 

Cette thèse porte sur l’apprentissage statistique et l’analyse de données multi-dimensionnelles. Elle se focalise particulièrement sur l’apprentissage non supervisé de modèles génératifs pour la classification automatique. Nous étudions les modèles de mélanges Gaussians, aussi bien dans le contexte d’estimation par maximum de vraisemblance via l’algorithme EM, que dans le contexte Bayésien d’estimation par Maximum A Posteriori via des techniques d’échantillonnage par Monte Carlo. Nous considérons principalement les modèles de mélange parcimonieux qui reposent sur une décomposition spectrale de la matrice de covariance et qui offre un cadre flexible notamment pour les problèmes de classification en grande dimension. Ensuite, nous investiguons les mélanges Bayésiens non-paramétriques qui se basent sur des processus généraux flexibles comme le processus de Dirichlet et le Processus du Restaurant Chinois. Cette formulation non-paramétrique des modèles est pertinente aussi bien pour l’apprentissage du modèle, que pour la question difficile du choix de modèle. Nous proposons de nouveaux modèles de mélanges Bayésiens non-paramétriques parcimonieux et dérivons une technique d’échantillonnage par Monte Carlo dans laquelle le modèle de mélange et son nombre de composantes sont appris simultanément à partir des données. La sélection de la structure du modèle est effectuée en utilisant le facteur de Bayes. Ces modèles, par leur formulation non-paramétrique et parcimonieuse, sont utiles pour les problèmes d’analyse de masses de données lorsque le nombre de classe est indéterminé et augmente avec les données, et lorsque la dimension est grande. Les modèles proposés validés sur des données simulées et des jeux de données réelles standard. Ensuite, ils sont appliqués sur un problème réel difficile de structuration automatique de données bioacoustiques complexes issues de signaux de chant de baleine. Enfin, nous ouvrons des perspectives Markoviennes via les processus de Dirichlet hiérarchiques pour les modèles Markov cachés.

This thesis focuses on statistical learning and multi-dimensional data analysis. It particularly focuses on unsupervised learning of generative models for model-based clustering. We study the Gaussians mixture models, in the context of maximum likelihood estimation via the EM algorithm, as well as in the Bayesian estimation context by maximum a posteriori via Markov Chain Monte Carlo (MCMC) sampling techniques. We mainly consider the parsimonious mixture models which are based on a spectral decomposition of the covariance matrix and provide a flexible framework particularly for the analysis of high-dimensional data. Then, we investigate non-parametric Bayesian mixtures which are based on general flexible processes such as the Dirichlet process and the Chinese Restaurant Process. This non-parametric model formulation is relevant for both learning the model, as well for dealing with the issue of model selection. We propose new Bayesian non-parametric parsimonious mixtures and derive a MCMC sampling technique where the…

Advisors/Committee Members: Glotin, Hervé (thesis director).

▼ 

A estimação e previsão da volatilidade de ativos são de suma importância para os mercados financeiros. Temas como risco e incerteza na teoria econômica incentivaram a procura por métodos capazes de modelar a variância condicional que evolui ao longo do tempo. O objetivo central desta dissertação foi modelar via modelos ARCH – GARCH e modelos aditivos o índice do IBOVESPA e o ativo PETR4 para analisar a existência de correlação entre as volatilidades estimadas. A estimação da volatilidade dos ativos no método paramétrico foi realizada via modelos EGARCH; já para o método não paramétrico, utilizouse os modelos aditivos com 5 defasagens.

Volatility estimation and forecasting are very important matters for the financial markets. Themes like risk and uncertainty in modern economic theory have encouraged the search for methods that allow for modeling of time varying variances. The main objective of this dissertation was estimate through GARCH models and additive models of IBOVESPA and PETR4 assets; and analyzes the existence of correlation between volatilities estimated. We use EGARCH models to estimate through parametric methods and use additive models 5 to estimate non parametric methods.

Advisors/Committee Members: Kloeckner, Gilberto de Oliveira.

▼ 

Cette thèse aborde le problème de la recherche d'une structure (ou clustering) dans lesnoeuds d'un graphe. Dans le cadre des modèles aléatoires à variables latentes, on attribue à chaque noeud i une variable aléatoire non observée (latente) Zi, et la probabilité de connexion des noeuds i et j dépend conditionnellement de Zi et Zj . Contrairement au modèle d'Erdos-Rényi, les connexions ne sont pas indépendantes identiquement distribuées; les variables latentes régissent la loi des connexions des noeuds. Ces modèles sont donc hétérogènes, et leur structure est décrite par les variables latentes et leur loi; ce pourquoi on s'attache à en faire l'inférence à partir du graphe, seule variable observée.La volonté commune des deux travaux originaux de cette thèse est de proposer des méthodes d'inférence de ces modèles, consistentes et de complexité algorithmique au plus linéaire en le nombre de noeuds ou d'arêtes, de sorte à pouvoir traiter de grands graphes en temps raisonnable. Ils sont aussi tous deux fondés sur une étude fine de la distribution des degrés, normalisés de façon convenable selon le modèle.Le premier travail concerne le Stochastic Blockmodel. Nous y montrons la consistence d'un algorithme de classiffcation non supervisée à l'aide d'inégalités de concentration. Nous en déduisons une méthode d'estimation des paramètres, de sélection de modèles pour le nombre de classes latentes, et un test de la présence d'une ou plusieurs classes latentes (absence ou présence de clustering), et nous montrons leur consistence.Dans le deuxième travail, les variables latentes sont des positions dans l'espace ℝd, admettant une densité f, et la probabilité de connexion dépend de la distance entre les positions des noeuds. Les clusters sont définis comme les composantes connexes de l'ensemble de niveau t > 0 fixé de f, et l'objectif est d'en estimer le nombre à partir du graphe. Nous estimons la densité en les positions latentes des noeuds grâce à leur degré, ce qui permet d'établir une correspondance entre les clusters et les composantes connexes de certains sous-graphes du graphe observé, obtenus en retirant les nœuds de faible degré. En particulier, nous en déduisons un estimateur du nombre de clusters et montrons saconsistence en un certain sens

.This thesis addresses the clustering of the nodes of a graph, in the framework of randommodels with latent variables. To each node i is allocated an unobserved (latent) variable Zi and the probability of nodes i and j being connected depends conditionally on Zi and Zj . Unlike Erdos-Renyi's model, connections are not independent identically distributed; the latent variables rule the connection distribution of the nodes. These models are thus heterogeneous and their structure is fully described by the latent variables and their distribution. Hence we aim at infering them from the graph, which the only observed data.In both original works of this thesis, we propose consistent inference methods with a computational cost no more than linear with respect to the number of nodes or edges,…

Advisors/Committee Members: Daudin, Jean-Jacques (thesis director), Robin, Stéphane (thesis director).

▼ 

Neste trabalho, apresentamos os modelos simétricos parcialmente lineares AR(1), que generalizam os modelos parcialmente lineares para a presença de erros autocorrelacionados seguindo uma estrutura de autocorrelação AR(1) e erros seguindo uma distribuição simétrica ao invés da distribuição normal. Dentre as distribuições simétricas, podemos considerar distribuições com caudas mais pesadas do que a normal, controlando a curtose e ponderando as observações aberrantes no processo de estimação. A estimação dos parâmetros do modelo é realizada por meio do critério de verossimilhança penalizada, que utiliza as funções escore e a matriz de informação de Fisher, sendo todas essas quantidades derivadas neste trabalho. O número efetivo de graus de liberdade e resultados assintóticos também são apresentados, assim como procedimentos de diagnóstico, destacando-se a obtenção da curvatura normal de influência local sob diferentes esquemas de perturbação e análise de resíduos. Uma aplicação com dados reais é apresentada como ilustração.

In this master dissertation, we present the symmetric partially linear models with AR(1) errors that generalize the normal partially linear models to contain autocorrelated errors AR(1) following a symmetric distribution instead of the normal distribution. Among the symmetric distributions, we can consider heavier tails than the normal ones, controlling the kurtosis and down-weighting outlying observations in the estimation process. The parameter estimation is made through the penalized likelihood by using score functions and the expected Fisher information. We derive these functions in this work. The effective degrees of freedom and asymptotic results are also presented as well as the residual analysis, highlighting the normal curvature of local influence under different perturbation schemes. An application with real data is given for illustration.

Advisors/Committee Members: Gilberto Alvarenga Paula, Francisco José de Azevêdo Cysneiros, Germán Mauricio Ibacache Pulgar.

▼ 

This study investigates whether Gaussian Process Regression can be applied to evaluate buffer queue times in large scale data processing systems. It is additionally considered whether high-frequency data stream rates can be generalized into a small subset of the sample space. With the aim of providing basis for dynamic software optimization, a promising foundation for continued research is introduced. The study is intended to contribute to Direct Market Access financial trading systems which processes immense amounts of market data daily. Due to certain limitations, we shoulder a naïve approach and model latencies as a function of only data throughput in eight small historical intervals. The training and test sets are represented from raw market data, and we resort to pruning operations to shrink the datasets by a factor of approximately 0.0005 in order to achieve computational feasibility. We further consider four different implementations of Gaussian Process Regression. The resulting algorithms perform well on pruned datasets, with an average R² statistic of 0.8399 over six test sets of approximately equal size as the training set. Testing on non-pruned datasets indicate shortcomings from the generalization procedure, where input vectors corresponding to low-latency target values are associated with less accuracy. We conclude that depending on application, the shortcomings may be make the model intractable. However for the purposes of this study it is found that buffer queue times can indeed be modelled by regression algorithms. We discuss several methods for improvements, both in regards to pruning procedures and Gaussian Processes, and open up for promising continued research. 

Denna studie undersöker huruvida Gaussian Process Regression kan appliceras för att utvärdera buffer-kötider i storskaliga dataprocesseringssystem. Dessutom utforskas ifall dataströmsfrekvenser kan generaliseras till en liten delmängd av utfallsrymden. Medmålet att erhålla en grund för dynamisk mjukvaruoptimering introduceras en lovandestartpunkt för fortsatt forskning. Studien riktas mot Direct Market Access system för handel på finansiella marknader, somprocesserar enorma mängder marknadsdata dagligen. På grund av vissa begränsningar axlas ett naivt tillvägagångssätt och väntetider modelleras som en funktion av enbartdatagenomströmning i åtta små historiska tidsinterval. Tränings- och testdataset representeras från ren marknadsdata och pruning-tekniker används för att krympa dataseten med en ungefärlig faktor om 0.0005, för att uppnå beräkningsmässig genomförbarhet. Vidare tas fyra olika implementationer av Gaussian Process Regression i beaktning. De resulterande algorithmerna presterar bra på krympta dataset, med en medel R2 statisticpå 0.8399 över sex testdataset, alla av ungefär samma storlek som träningsdatasetet. Tester på icke krympta dataset indikerar vissa brister från pruning, där input vektorermotsvararande låga latenstider är associerade med mindre…

▼ 

A interação genótipos ambientes (G E) foi definido por Shelbourne (1972) como sendo a variação entre genótipos em resposta a diferentes condições ambientais. Sua magnitude na expressão fenotípica do caráter pode reduzir a correlação entre fenótipo e genótipo, in acionando a variância genética e, por sua vez, parâmetros dependentes desta, como herdabilidade e ganho genético com a seleção. Estudos sobre a adaptabilidade e a estabilidade fenotípica permitem particularizar os efeitos da interação GE ao nível de genótipo e ambiente, identificando a contribuição relativa de cada um para a interação total. Varias metodologias estatísticas têm sido propostas para a interpretação da interação G E proveniente de um grupo de cultivares testados em vários ambientes. Entre essas metodologias destaca-se os modelos AMMI (Additive Main Eects and Multiplicative Interaction Model), que vem ganhando grande aplicabilidade nos últimos anos. O modelo AMMI e um método uni-multivariado, que engloba uma analise de variância para os efeitos principais, que são os efeitos dos genótipos (G) e os ambientes (E) e para os efeitos multiplicativos (interação genótipo ambiente), para a qual utiliza-se a decomposição em valor singular (DVS). Essa técnica multivariada baseia-se no uso dos autovalores e autovetores provenientes da matriz de interação G E. Araujo e Dias (2005) verificaram o problema de superestimação e subestimação de autovalores estimados da maneira convencional. Efron(1979) propôs uma técnica de simulação numérica chamada Bootstrap para avaliar tais incertezas. O método Bootstrap consiste em uma técnica de reamostragem que permite aproximar a distribuição de uma função das observações a partir da distribuição empírica dos dados. Por meio desse método, podem ser estimados o erro-padrão da referida estimativa e os intervalos de confiança, com o intuito de fazer inferência sobre os parâmetros em questão. O objetivo deste trabalho será estudar o efeito da interação G E, avaliar a adaptabilidade e estabilidade de genótipos em diferentes ambientes através do modelo AMMI, com as analises através dos gráficos Biplot, encontrar a distribuição empírica dos autovalores e calcular o intervalo de confiança através o método Bootstrap não-paramétrico. Com o estudo da distribuição empírica dos autovalores poder-se-a validar os testes de hipóteses propostos na literatura para identificar o numero de IPCA (Incremental Principal Component Analysis) para seleção dos modelos AMMI, e propor um teste para seleção dos modelos.

The genotype environment interaction (G E) was dened by Shelbourne (1972) as the variation among genotypes in response to dierent environmental conditions. Its magnitude in phenotypic expression of the character can reduce the correlation between genotype and phenotype, in ating the genetic variance and, in turn, dependent on the parameters, as heritability and genetic gain with selection. Studies on the phenotypic adaptability and stability allow particularize the eects of interaction G E at the level of genotype and…

Advisors/Committee Members: Carlos Tadeu dos Santos Dias, Lúcio Borges de Araújo, Sonia Maria de Stefano Piedade.

▼ 

The Birnbaum-Saunders (BS) distribution is the most popular model used to describe lifetime process under fatigue. Throughout the years, this distribution has received a wide ranging of applications, demanding some more flexible extensions to solve more complex problems. One of the most well-known extensions of the BS distribution is the generalized Birnbaum- Saunders (GBS) family of distributions that includes the Birnbaum-Saunders special-case (BSSC) and the Birnbaum-Saunders generalized t (BSGT) models as special cases. Although the BS-SC distribution was previously developed in the literature, it was never deeply studied and hence, in this thesis, we provide a full Bayesian study and develop a tool to generate random numbers from this distribution. Further, we develop a very flexible regression model, that admits different degrees of skewness and kurtosis, based on the BSGT distribution using the generalized additive models for location, scale and shape (GAMLSS) framework. We also introduce a new extension of the BS distribution called the Birnbaum-Saunders power (BSP) family of distributions, which contains several special or limiting cases already published in the literature, including the GBS family. The main feature of the new family is that it can produce both unimodal and bimodal shapes depending on its parameter values. We also introduce this new family of distributions into the GAMLSS framework, in order to model any or all the parameters of the distribution using parametric linear and/or nonparametric smooth functions of explanatory variables. Throughout this thesis we present five different applications in real data sets in order to illustrate the developed theoretical results.

A distribuição Birnbaum-Saunders (BS) é o modelo mais popular utilizado para descrever processos de fadiga. Ao longo dos anos, essa distribuição vem recebendo aplicações nas mais diversas áreas, demandando assim algumas extensões mais flexíveis para resolver problemas mais complexos. Uma das extensões mais conhecidas na literatura é a família de distribuições Birnbaum-Saunders generalizada (GBS), que inclui as distribuições Birnbaum-Saunders casoespecial (BS-SC) e Birnbaum-Saunders t generalizada (BSGT) como modelos especiais. Embora a distribuição BS-SC tenha sido previamente desenvolvida na literatura, nunca foi estudada mais profundamente e, assim, nesta tese, um estudo bayesiano é desenvolvido acerca da mesma além de um novo gerador de números aleatórios dessa distribuição ser apresentado. Adicionalmente, um modelo de regressão baseado na distribuição BSGT é desenvolvido utilizando-se os modelos aditivos generalizados para locação, escala e forma (GAMLSS), os quais apresentam grande flexibilidade tanto para a assimetria como para a curtose. Uma nova extensão da distribuição BS também é apresentada, denominada família de distribuições Birnbaum-Saunders potência (BSP), que contém inúmeros casos especiais ou limites já publicados na literatura, incluindo a família GBS. A principal característica desta nova família é…

Advisors/Committee Members: Roseli Aparecida Leandro, Mário de Castro Andrade Filho, Cristian Marcelo Villegas Lobos.

▼ 

O principal objetivo deste trabalho é descrever e analisar do ponto de vista teórico os principais métodos não paramétricos de avaliação de eficiência e avaliar empiricamente os resultados gerados por esses diferentes métodos aplicados a um mesmo conjunto de dados. Com essa finalidade, realizou-se diversas simulações teóricas para fins comparativos dos tradicionais estimadores DEA e FDH, inclusive de seus procedimentos para inferência e correção de viés, e dos novos estimadores de ordem m e quantil. Empiricamente, utilizamos uma amostra de 184 bancos comerciais brasileiros no período de Junho/1995 à Junho/2004. Os resultados mostraram que as diferentes suposições impostas ao conjunto de produção pelos estimadores DEA e FDH afetam sensivelmente os índices de eficiência de várias firmas. Apesar disso, o uso de mais de um estimador mostrou-se um bom artifício para identificação das unidades com pior desempenho. Os procedimentos disponíveis para correção de viés e inferência, entretanto, mostraram-se deficientes principalmente para as firmas localizadas ao longo da fronteira estimada. Por outro lado, a importância da utilização dos novos estimadores não paramétricos quantil e de ordem m ficou evidente devido a presença de observações consideradas valores extremos, que distorcem os índices de eficiência estimados de outras observações. O uso de tais estimadores, mais robustos a valores extremos e outliers, gerou resultados mais confiáveis. Finalmente, procurou-se investigar se o controle de capital, o segmento de atuação e o porte dos bancos afetam sua eficiência, além de investigar o comportamento da performance do setor durante o período de análise.

The purpose of this piece of work has been to analyse and describe the main non parametric efficiency evaluation methods from a theoretical point of view and to empirically analyse the efficiency scores generated by such methods using the same data sample. For that purpose, theoretical simulations were used in order to compare traditional DEA and FDH estimators, their inference and bias correction procedures included. The new order m and quantil estimators were also used. We empirically used a sample of 184 Brazilian commercial banks spanning the June/1995 to June/2004 period. The results show that several banks’ efficiency scores were noticeably affected by the different assumptions made by the DEA and FDH estimators towards the production set. Nevertheless, the use of more than one estimator proved to be an effective way of identifying the units with the worst performance levels. The available procedures for bias correction and inference, however, proved ineffective for frontier firms. On the other hand, the importance of utilising the new order m and quantil estimators was evidenced. Several observations from our sample were detected as extreme values, which affected the estimated efficiency scores from other observations. The use of the aforementioned estimators, more robust to extreme values and outliers, generated more trustworthy results. Finally, an attempt was…

Advisors/Committee Members: Ribeiro, Eduardo Pontual, Martins Filho, Carlos.

▼ Analysing and modelling metocean variables is crucial in various scientific fields with both socio-economic and environmental impacts, such as offshore engineering and coastal morphology studies. The rapidly emerging sector of marine renewable energy is also largely based on the sound analysis and modelling of the metocean climate in a candidate area at various temporal scales (e.g. seasonal, interannual). The involved physical processes that govern the metocean environment are highly interrelated; linear metocean variables are related with either other linear or directional variables and vice versa, rendering their adequate joint description a demanding task. On the other hand, their variability aspects are of great interest for long-term planning purposes. Furthermore, such phenomena can be realized at entirely different time scales; for instance, fatigue of structures placed in the marine environment or equilibrium of a coastal zone due to erosion/accretion patterns are affected not only by the severe (extreme) environmental conditions acting for a limited time period (e.g. storm events) but also by their repeated and continuous action corresponding to a longer-term “fatigue”. The focus of this dissertation is on the analytic probabilistic modelling and assessment of linear and directional metocean random variables aiming at an integrated and feasible approach for climate modelling. This task extends to a wide spectrum of less known probabilistic approaches, such as linear-directional joint probability models, circular regression and calibration, estimation of extreme events taking into account the directional covariate, etc. The modelling approaches refer to the long-term scale, but the methods apply equally well on any time scale. Moreover, the multitude of alternative models renders the statistical decision procedure a very delicate task, since the results of this step play a decisive role in ocean and coastal engineering applications. In the first part, the main theoretical background is presented starting with conventional linear and directional univariate and bivariate models, along with some distributions that have been recently proved to be efficient for modelling metocean characteristics, while mixtures of different univariate models are also considered. The examined bivariate models that are based on the corresponding marginal distributions and an appropriate dependence structure, are described in detail. Both dimensions (one and two) are addressed by parametric and non-parametric models. Also, the current availability of multiple data sources leads to the necessity of validating and correcting (calibrating) metocean data with emphasis on regression models that take into consideration errors in both variables and the presence of outliers, while calibration techniques are described for linear and directional features, with the latter being rarely adopted in relevant applications despite its significance. As safety, performance and economic viability of marine structures are affected by directional…

▼ Le volume des données disponibles permettant de décrire l’environnement, en particulier l’atmosphère et les océans, s’est accru à un rythme exponentiel. Ces données regroupent des observations et des sorties de modèles numériques. Les observations (satellite, in situ, etc.) sont généralement précises mais sujettes à des erreurs de mesure et disponibles avec un échantillonnage spatio-temporel irrégulier qui rend leur exploitation directe difficile. L’amélioration de la compréhension des processus physiques associée à la plus grande capacité des ordinateurs ont permis des avancées importantes dans la qualité des modèles numériques. Les solutions obtenues ne sont cependant pas encore de qualité suffisante pour certaines applications et ces méthodes demeurent lourdes à mettre en œuvre. Filtrage et lissage (les méthodes d’assimilation de données séquentielles en pratique) sont développés pour abonder ces problèmes. Ils sont généralement formalisées sous la forme d’un modèle espace-état, dans lequel on distingue le modèle dynamique qui décrit l’évolution du processus physique (état), et le modèle d’observation qui décrit le lien entre le processus physique et les observations disponibles. Dans cette thèse, nous abordons trois problèmes liés à l’inférence statistique pour les modèles espace-états: reconstruction de l’état, estimation des paramètres et remplacement du modèle dynamique par un émulateur construit à partir de données. Pour le premier problème, nous introduirons tout d’abord un algorithme de lissage original qui combine les algorithmes Conditional Particle Filter (CPF) et Backward Simulation (BS). Cet algorithme CPF-BS permet une exploration efficace de l’état de la variable physique, en raffinant séquentiellement l’exploration autour des trajectoires qui respectent le mieux les contraintes du modèle dynamique et des observations. Nous montrerons sur plusieurs modèles jouets que, à temps de calcul égal, l’algorithme CPF-BS donne de meilleurs résultats que les autres CPF et l’algorithme EnKS stochastique qui est couramment utilisé dans les applications opérationnelles. Nous aborderons ensuite le problème de l’estimation des paramètres inconnus dans les modèles espace-état. L’algorithme le plus usuel en statistique pour estimer les paramètres d’un modèle espace-état est l’algorithme EM qui permet de calculer itérativement une approximation numérique des estimateurs du maximum de vraisemblance. Nous montrerons que les algorithmes EM et CPF-BS peuvent être combinés efficacement pour estimer les paramètres d’un modèle jouet. Pour certaines applications, le modèle dynamique est inconnu ou très coûteux à résoudre numériquement mais des observations ou des simulations sont disponibles. Il est alors possible de reconstruire l’état conditionnellement aux observations en utilisant des algorithmes de filtrage/lissage dans lesquels le modèle dynamique est remplacé par un émulateur statistique construit à partir des observations. Nous montrerons que les algorithmes EM et CPF-BS peuvent être adaptés dans ce cadre et permettent… Advisors/Committee Members: Monbet, Valérie (thesis director), Ailliot, Pierre (thesis director).

▼ 

La thèse se compose de deux parties : une partie consacrée à l'estimation des quantiles conditionnels et une autre à l'apprentissage supervisé. La partie "Estimation des quantiles conditionnels" est organisée en 3 chapitres : Le chapitre 1 est consacré à une introduction sur la régression linéaire locale, présentant les méthodes les plus utilisées, pour estimer le paramètre de lissage. Le chapitre 2 traite des méthodes existantes d’estimation nonparamétriques du quantile conditionnel ; Ces méthodes sont comparées, au moyen d’expériences numériques sur des données simulées et des données réelles. Le chapitre 3 est consacré à un nouvel estimateur du quantile conditionnel et que nous proposons ; Cet estimateur repose sur l'utilisation d'un noyau asymétrique en x. Sous certaines hypothèses, notre estimateur s'avère plus performant que les estimateurs usuels. La partie "Apprentissage supervisé" est, elle aussi, composée de 3 chapitres : Le chapitre 4 est une introduction à l’apprentissage statistique et les notions de base utilisées, dans cette partie. Le chapitre 5 est une revue des méthodes conventionnelles de classification supervisée. Le chapitre 6 est consacré au transfert d'un modèle d'apprentissage semi-paramétrique. La performance de cette méthode est montrée par des expériences numériques sur des données morphométriques et des données de credit-scoring.

The thesis consists of two parts: One part is about the estimation of conditional quantiles and the other is about supervised learning. The "conditional quantile estimate" part is organized into 3 chapters. Chapter 1 is devoted to an introduction to the local linear regression and then goes on to present the methods, the most used in the literature to estimate the smoothing parameter. Chapter 2 addresses the nonparametric estimation methods of conditional quantile and then gives numerical experiments on simulated data and real data. Chapter 3 is devoted to a new conditional quantile estimator, we propose. This estimator is based on the use of asymmetrical kernels w.r.t. x. We show, under some hypothesis, that this new estimator is more efficient than the other estimators already used. The "supervised learning" part is, too, with 3 chapters: Chapter 4 provides an introduction to statistical learning, remembering the basic concepts used in this part. Chapter 5 discusses the conventional methods of supervised classification. Chapter 6 is devoted to propose a method of transferring a semiparametric model. The performance of this method is shown by numerical experiments on morphometric data and credit-scoring data.

Advisors/Committee Members: Beninel, Farid (thesis director).

▼ Business cycles, the ups and downs observed somewhat simultaneously in numerous macroeconomic variables in an economy and often measured using real GDP, are important and, despite much economic research, still incom- pletely understood. Dating the business cycle has always been of interest in macroeconomic research. The dating might help to find the causes of a recession and this understanding could help to prevent or limit the duration of recessions in the future. A non-stationary, non-parametric smoothing- technique is proposed here to make business cycles simpler to analyse and interpret. The method is applied to the Euro area and to the Swedish econ- omy. For the Euro area the method finds two deeper and two milder reces- sions and one stagnation period since 1970. The dating is close to that of the CEPR. The same method is then used to date recessions in Sweden for the period 1969-2006. Four recessions were found. One research area of interest related to the dating of business cycles is forecasting of an upcoming recession. If an upcoming recession is detected, monetary policy could respond and avoid an output gap or a fall in inflation. We use a probit model to examine the in-sample performance of various financial variables as a predictor of Swedish recessions. The results show that the slope of the yield curve appears to perform better than other variables, but also that the spread is not a reliable indicator for detecting recessions in Sweden since there are many false warnings.

▼ Cette thèse couvre plusieurs sujets sur la répartition des revenus et des richesses. Dans le premier chapitre, nous développons une nouvelle méthode pour exploiter les tabulations de revenu et de richesse, telle que celle publiée par les autorités fiscales. Nous y définissons les courbes de Pareto généralisées comme la courbe des coefficients de Pareto inversés b(p), où b(p) est le rapport entre le revenu moyen ou la richesse au-dessus du rang p et le p-ième quantile Q(p) (c'est-à-dire b(p)=E[X|X>Q(p)]/Q(p)). Nous les utilisons pour caractériser des distributions entières, y compris les endroits comme le sommet où la lois de Pareto est une bonne description, et les endroits plus bas où elles ne le sont pas. Nous développons une méthode pour reconstruire de manière flexible l'ensemble de la distribution sur la base de données tabulées sur le revenu ou le patrimoine, qui produit courbes de Pareto généralisées lisses et réalistes.Dans le deuxième chapitre, nous présentons une nouvelle approche pour combiner les données d'enquête et les tabulations fiscales afin de corriger la sous-représentation des plus riches au sommet. Elle détermine de façon endogène un "point de fusion'' entre les données avant de modifier les poids tout au long de la distribution et de remplacer les nouvelles observations au-delà du support original de l'enquête. Nous fournissons des simulations de la méthode et des applications aux données réelles. Les premières démontrent que notre méthode améliore la précision et la stabilité des estimations de la distribution, par rapport à d'autres méthodes de correction d'enquêtes utilisant des données externes, et même en présence d'hypothèses extrêmes. Les applications empiriques montrent que non seulement les niveaux d'inégalité des revenus peuvent changer, mais aussi les tendances.Dans le troisième chapitre, nous estimons la distribution du revenu national dans 38 pays européens entre 1980 et 2017 en combinant enquêtes, données fiscales et comptes nationaux. Nous développons une méthodologie cohérente combinant des méthodes d'apprentissage statistique, de calage non linéaire des enquêtes et la théorie des valeurs extrêmes afin de produire des estimations de l'inégalité des revenus avant et après impôt, comparables d'un pays à l'autre et conformes aux taux de croissance macroéconomiques. Nous constatons que les inégalités se sont creusées dans une majorité de pays européens, en particulier entre 1980 et 2000. Le 1% les plus riches en Europe a augmenté plus de deux fois plus vite que les 50% les plus pauvres et a capturé 18% de la croissance des revenus régionaux.Dans le quatrième chapitre, je décompose la dynamique de la distribution de la richesse à l'aide d'un modèle stochastique dynamique simple qui sépare les effets de la consommation, du revenu du travail, des taux de rendement, de la croissance, de la démographie et du patrimoine. À partir de deux théorèmes de calcul stochastique, je montre que ce modèle est identifié de manière non paramétrique et qu'il peut être estimé à partir de données en… Advisors/Committee Members: Piketty, Thomas (thesis director).

▼ The focus of this thesis is models for non-parametric clustering of multivariate count data. While there has been significant work in Bayesian non-parametric modelling in the last decade, in the context of mixture models for real-valued data and some forms of discrete data such as multinomial-mixtures, there has been much less work on non-parametric clustering of Multi-variate Count Data. The main challenges in clustering multivariate counts include choosing a suitable multivariate distribution that adequately captures the properties of the data, for instance handling over-dispersed data or sparse multivariate data, at the same time leveraging the inherent dependency structure between dimensions and across instances to get meaningful clusters. As the first contribution, this thesis explores extensions to the Multivariate Poisson distribution, proposing efficient algorithms for non-parametric clustering of multivariate count data. While Poisson is the most popular distribution for count modelling, the Multivariate Poisson often leads to intractable inference and a suboptimal t of the data. To address this, we introduce a family of models based on the Sparse-Multivariate Poisson, that exploit the inherent sparsity in multivariate data, reducing the number of latent variables in the formulation of Multivariate Poisson leading to a better t and more efficient inference. We explore Dirichlet process mixture model extensions and temporal non-parametric extensions to models based on the Sparse Multivariate Poisson for practical use of Poisson based models for non-parametric clustering of multivariate counts in real-world applications. As a second contribution, this thesis addresses moving beyond the limitations of Poisson based models for non-parametric clustering, for instance in handling over dispersed data or data with negative correlations. We explore, for the first time, marginal independent inference techniques based on the Gaussian Copula for multivariate count data in the Dirichlet Process mixture model setting. This enables non-parametric clustering of multivariate counts without limiting assumptions that usually restrict the marginal to belong to a particular family, such as the Poisson or the negative-binomial. This inference technique can also work for mixed data (combination of counts, binary and continuous data) enabling Bayesian non-parametric modelling to be used for a wide variety of data types. As the third contribution, this thesis addresses modelling a wide range of more complex dependencies such as asymmetric and tail dependencies during non-parametric clustering of multivariate count data with Vine Copula based Dirichlet process mixtures. While vine copula inference has been well explored for continuous data, it is still a topic of active research for multivariate counts and mixed multivariate data. Inference for multivariate counts and mixed data is a hard problem owing to ties that arise with discrete marginal. An efficient marginal independent inference approach based on extended rank likelihood,… Advisors/Committee Members: Bhattacharyya, Chiranjib (advisor).

▼ The co-operative as an economic and social institution has long been recognized for its contribution to economic development as well as its positive effect on local communities. However, over the last decade or so substantial structural changes in the agricultural sector have undermined some of the most prominent North American co-operatives. In some cases, co-ops asked for bankruptcy protection, others ceased operations while some were transformed to for-profit firms. The present study offers three essays that explore the challenges that co-operatives are facing in terms of their relationship with their members in local markets, the decision-making process of their leaders and the co-ops' role in the modern economy. These first two essays are linked by the fact that they both develop models that are about cognitions. Examining cognition offers some new insights to understanding the process behind the decline of agricultural co-ops. In the first essay the model examines consumers' cognitions, while the model in the second essay examines management's cognitions. The essays differ on the agent's ability to change the perceptions that result from those cognitions. Essay One assumes that consumers' perceptions are partially flexible and thus can change over time with some cost; on the other hand, essay Two assumes that beliefs are inflexible due to the high cost of changing them. Essay One examines the relationship between a co-operative and its membership in a local market using an economic psychological approach. More specifically, the essay presents a modified rational-choice model to investigate how cognitive dissonance can influence members' loyalty. The effect of cognitive dissonance is analyzed in a case where a local co-operative operates alongside with an investor-owned firm (IOF) in a market. The model illustrates how cognitive dissonance can give rise to switching costs for those consumers who wish to switch to the IOF. Analytical results demonstrate the effect of these switching costs on equilibrium market shares and discuss how a drop in the dissonance cost because of managerial decisions by the co-op can result in dramatic drops in its market share. Essay Two illustrates how management's mental frame can be incorporated into an economic model and develops a theoretical underpinning for the link between a strong mental frame and the financial difficulties that a firm might experience. The case of the Saskatchewan Wheat Pool with its Project Horizon plan is proposed as an example of a situation where the established mental frame gave rise to a belief regarding future member support that had a significant influence on the decision-making process of the co-op's CEO. The analysis includes a game theoretic model of a duopoly between a co-operative and an IOF, where mental framing is explicitly incorporated into the primitives of the model. Analytical results illustrate how the CEO's belief regarding member commitment can influence decision-making and therefore affect the market share and profits of the firm. … Advisors/Committee Members: Fulton, Murray E., Vercammen, James, Gray, Richard S., Cranfield, John, Zhao, Jingang.

▼ 

L’analyse dynamique des structures de grandes dimensions incluant de nombreux paramètres incertains et des non-linéarités localisées ou réparties peut être numériquement prohibitive. Afin de surmonter ce problème, des modèles d’approximation peuvent être développés pour reproduire avec précision et à faible coût de calcul la réponse de la structure.L’objectif de la première partie de ce mémoire est de développer des modèles numériques robustes vis-à-vis des modifications structurales (non-linéarités localisées, perturbations ou incertitudes paramétriques) et « légers » au sens de la réduction de la taille. Ces modèles sont construits, selon les approches de condensation directe et par synthèse modale, en enrichissant des bases de réduction tronquées, modale et de Craig-Bampton respectivement, avec des résidus statiques prenant compte des modifications structurales. Pour propager les incertitudes, l’accent est mis particulièrement sur la méthode du chaos polynomial généralisé. Sa combinaison avec les modèles réduits ainsi obtenus permet de créer des métamodèles mono et bi-niveaux, respectivement. Les deux métamodèles proposés sont comparés à d’autres métamodèles basés sur les méthodes du chaos polynomial généralisé et du Latin Hypercube appliquées sur des modèles complets et réduits. Les métamodèles proposés permettent d’approximer les comportements structuraux avec un coût de calcul raisonnable et sans perte significative de précision.La deuxième partie de ce mémoire est consacrée à l’analyse dynamique des structures périodiques non-linéaires en présence des imperfections : perturbations des paramètres structuraux ou incertitudes paramétriques. Deux études : déterministe ou stochastique, respectivement, sont donc menées. Pour ces deux configurations, un modèle analytique discret générique est proposé. Il consiste à appliquer la méthode des échelles multiples et la méthode de perturbation pour résoudre l’équation de mouvement et de projecter la solution obtenue sur des modes d’ondes stationnaires. Le modèle proposé conduit à un ensemble d’équations algébriques complexes couplées, fonctions du nombre et des positions des imperfections dans la structure. La propagation des incertitudes à travers le modèle ainsi construit est finalement assurée par les méthodes du Latin Hypercube et du chaos polynomial généralisé. La robustesse de la dynamique collective vis-à-vis des imperfections est étudiée à travers l’analyse statistique de la dispersion des réponses fréquentielles et des bassins d’attraction dans le domaine de multistabilité. L’étude numérique montre que la présence des imperfections dans une structure périodique renforce sa non-linéarité, élargit son domaine de multistabilité et génère une multiplicité de branches multimodale.

Dynamic analysis of large scale structures including several uncertain parameters and localized or distributed nonlinearities may be computationally unaffordable. In order to overcome this issue, approximation models can be developed to reproduce accurately the structural response at a low…

Advisors/Committee Members: Bouhaddi, Noureddine (thesis director), Guedri, Mohamed (thesis director), Kacem, Najib (thesis director).

▼ The overwhelming accumulation of traffic volumes and relentless changes in travel-related characteristics significantly increase vehicular emissions, and hence, seriously affect urban air quality. It is difficult, however, to accurately estimate vehicular emissions in traffic intersections, junctions, and at signalized roadways because rate models for predicting vehicular emissions are insensitive to the vehicle modes of operations, such as cruising, idling, acceleration and deceleration. The reason is that these models are usually based on the average trip speed, not vehicle dynamics. These contribute to the increased complexity of such a model and degradation of its predictive performance. This thesis advocates the feasibility of using variables such as vehicle speed, acceleration, load, power and ambient temperature to predict transport emissions to ensure that emission inventories are accurate for the sake of air quality modelling and management planning. A variety of algorithms has been developed, based on Multivariate Adaptive Regression Splines (MARS), Boosting Multivariate Adaptive Regression Splines (BMARS), Artificial Neural Networks (ANNs), as well as the non-parametric Classification and Regression Trees (CART) and a combination of them in hybrid models to improve the accuracy of the emission prediction using vehicles’ on-board measurements and chassis dynamometer testing. Several performance indices are used to evaluate: accuracy, flexibility and computational efficiency. The obtained results suggest that the CART-BMARS hybrid methodology appears to be a useful and fairly accurate tool for predicting microscale vehicle emissions and may be adopted by regulatory agencies. The significance of this thesis is in providing of feasible and effective solutions for the implementation of vehicular emissions models to address the problem of air quality modelling and control in metropoles and mega-cities.