Diaconis and Ylvisaker (1979) give necessary conditions for conjugate priors for distributions from the natural exponential family to be proper as well as to have the property of linear posterior expectation of the mean parameter of the family. Their conditions for propriety and linear posterior expectation are also sufficient if the natural parameter space is equal to the set of all d-dimensional real numbers. In this paper their results are extended to characterize when conjugate priors are proper if the natural parameter space is bounded. For the special case where the natural exponential family is through a spherical probability distribution n,we show that the proper conjugate priors can be characterized by the behavior of the moment generating function of n at the boundary of the natural parameter space, or the second-order tail behavior of n. In addition, we show that if these families are non-regular, then linear posterior expectation never holds. The results for this special case are also extended to natural exponential families through elliptical probability distributions.

En analyse multivariée de données de grande dimension, les lois de Wishart définies dans le contexte des modèles graphiques revêtent une grande importance car elles procurent parcimonie et modularité. Dans le contexte des modèles graphiques Gaussiens régis par un graphe G, les lois de Wishart peuvent être définies sur deux restrictions alternatives du cône des matrices symétriques définies positives : le cône PG des matrices symétriques définies positives x satisfaisant xij=0, pour tous sommets i et j non adjacents, et son cône dual QG. Dans cette thèse, nous proposons une construction harmonieuse de familles exponentielles de lois de Wishart sur les cônes PG et QG. Elle se focalise sur les modèles graphiques d'interactions des plus proches voisins qui présentent l'avantage d'être relativement simples tout en incluant des exemples de tous les cas particuliers intéressants: le cas univarié, un cas d'un cône symétrique, un cas d'un cône homogène non symétrique, et une infinité de cas de cônes non-homogènes. Notre méthode, simple, se fonde sur l'analyse sur les cônes convexes. Les lois de Wishart sur QAn sont définies à travers la fonction gamma sur QAn et les lois de Wishart sur PAn sont définies comme la famille de Diaconis- Ylvisaker conjuguée. Ensuite, les méthodes développées sont utilisées pour résoudre la conjecture de Letac- Massam sur l'ensemble des paramètres de la loi de Wishart sur QAn. Cette thèse étudie aussi les sousmodèles, paramétrés par un segment dans M, d'une famille exponentielle paramétrée par le domaine des moyennes M. / In the framework of Gaussian graphical models governed by a graph G, Wishart distributions can be defined on two alternative restrictions of the cone of symmetric positive definite matrices: the cone PG of symmetric positive definite matrices x satisfying xij=0 for all non-adjacent vertices i and j and its dual cone QG. In this thesis, we provide a harmonious construction of Wishart exponential families in graphical models. Our simple method is based on analysis on convex cones. The focus is on nearest neighbours interactions graphical models, governed by a graph An, which have the advantage of being relatively simple while including all particular cases of interest such as the univariate case, a symmetric cone case, a nonsymmetric homogeneous cone case and an infinite number of non-homogeneous cone cases. The Wishart distributions on QAn are constructed as the exponential family generated from the gamma function on QAn. The Wishart distributions on PAn are then constructed as the Diaconis- Ylvisaker conjugate family for the exponential family of Wishart distributions on QAn. The developed methods are then used to solve the Letac-Massam Conjecture on the set of parameters of type I Wishart distributions on QAn. Finally, we introduce and study exponential families of distributions parametrized by a segment of means with an emphasis on their Fisher information. The focus in on distributions with matrix parameters. The particular cases of Gaussian and Wishart exponential families are further examined.

In the thesis "On Boundaries of Statistical Models" problems related to a description of probability distributions with zeros, lying in the boundary of a statistical model, are treated. The distributions considered are joint distributions of finite collections of finite discrete random variables. Owing to this restriction, statistical models are subsets of finite dimensional real vector spaces. The support set problem for exponential families, the main class of models considered in the thesis, is to characterize the possible supports of distributions in the boundaries of these statistical models. It is shown that this problem is equivalent to a characterization of the face lattice of a convex polytope, called the convex support. The main tool for treating questions related to the boundary are implicit representations. Exponential families are shown to be sets of solutions of binomial equations, connected to an underlying combinatorial structure, called oriented matroid. Under an additional assumption these equations are polynomial and one is placed in the setting of commutative algebra and algebraic geometry. In this case one recovers results from algebraic statistics. The combinatorial theory of exponential families using oriented matroids makes the established connection between an exponential family and its convex support completely natural: Both are derived from the same oriented matroid. The second part of the thesis deals with hierarchical models, which are a special class of exponential families constructed from simplicial complexes. The main technical tool for their treatment in this thesis are so called elementary circuits. After their introduction, they are used to derive properties of the implicit representations of hierarchical models. Each elementary circuit gives an equation holding on the hierarchical model, and these equations are shown to be the "simplest", in the sense that the smallest degree among the equations corresponding to elementary circuits gives a lower bound on the degree of all equations characterizing the model. Translating this result back to polyhedral geometry yields a neighborliness property of marginal polytopes, the convex supports of hierarchical models. Elementary circuits of small support are related to independence statements holding between the random variables whose joint distributions the hierarchical model describes. Models for which the complete set of circuits consists of elementary circuits are shown to be described by totally unimodular matrices. The thesis also contains an analysis of the case of binary random variables. In this special situation, marginal polytopes can be represented as the convex hulls of linear codes. Among the results here is a classification of full-dimensional linear code polytopes in terms of their subgroups. If represented by polynomial equations, exponential families are the varieties of binomial prime ideals. The third part of the thesis describes tools to treat models defined by not necessarily prime binomial ideals. It follows from Eisenbud and Sturmfels' results on binomial ideals that these models are unions of exponential families, and apart from solving the support set problem for each of these, one is faced with finding the decomposition. The thesis discusses algorithms for specialized treatment of binomial ideals, exploiting their combinatorial nature. The provided software package Binomials.m2 is shown to be able to compute very large primary decompositions, yielding a counterexample to a recent conjecture in algebraic statistics.

Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2017-04-25T14:30:32Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Dissertacao_Lelio_Alejandro_Arias_Vizcaino.pdf: 1217481 bytes, checksum: 3e169ccf7afc8c3a244b8cc4a07c9cbf (MD5) / Made available in DSpace on 2017-04-25T14:30:32Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Dissertacao_Lelio_Alejandro_Arias_Vizcaino.pdf: 1217481 bytes, checksum: 3e169ccf7afc8c3a244b8cc4a07c9cbf (MD5) Previous issue date: 2016-12-05 / FACEPE / entre as principais metodologias estatísticas, a análise de regressão é uma das formas mais efetivas para modelar dados. Neste sentido, a análise de diagnóstico é imprescindível para determinar o que poder ter acontecido no processo gerador dos dados caso os pressupostos impostos a este não sejam plausíveis. Uma das ferramentas mais úteis em diagnóstico é a avaliação dos resíduos. Neste trabalho, propomos um novo resíduo para as classes de modelos de regressão linear e não linear baseados na família exponencial com dispersão variável (Smyth (1989)). A proposta permite incorporar de forma simultânea informações relativas aos submodelos da média e da dispersão sem fazer uso de matrizes de projeção para sua padronização. Resultados de simulação e de aplicações a dados reais mostram que o novo resíduo é altamente competitivo em relação ao resíduos amplamente usados e consolidados na literatura. / In statistical methodologies, regression analysis can be a very effective way to model data. In this sense, the diagnostic analysis is needed to try to determine what might happened in the data generating process if the conditions imposed to it are not true. One of the most useful techniques to detect the goodness of fit to the model is the evaluation of residuals. In this work, we propose a new residual to the class of linear and nonlinear regression models based on exponential family with variable dispersion (Smyth (1989)). The proposal incorporates simultaneously information from the sub-models of the mean and the dispersion without using projection matrices for its standardization. Simulation resultsandapplicationsinrealdatashowthatthenewresidualishighlycompetitivewith respect to residuals widely used and established in the literature.

Cette thèse examine la dispersion multivariée des modelés normales stables Tweedie. Trois estimateurs de fonction variance généralisée sont discutés. Ensuite dans le cadre de la famille exponentielle naturelle deux caractérisations du modèle normal-Poisson, qui est un cas particulier de modèles normales stables Tweedie avec composante discrète, sont indiquées : d'abord par fonction variance et ensuite par fonction variance généralisée. Le dernier fournit la solution à un problème particulier d'équation de Monge-Ampère. Enfin, pour illustrer l'application de la variance généralisée des modèles Tweedie stables normales, des exemples à partir des données réelles sont fournis. / This thesis examines the multivariate dispersion of normal stable Tweedie (NST) models. Three generalize variance estimators of some NST models are discussed. Then within the framework of natural exponential family, two characterizations of normal Poisson model, which is a special case of NST models with discrete component, are shown : first by variance function and then by generalized variance function. The latter provides a solution to a particular Monge-Ampere equation problem. Finally, to illustrate the application of generalized variance of normal stable Tweedie models, examples from real data are provided.

Dans les applications modernes des statistiques et de l'apprentissage, il est courant que les données récoltées présentent un certain nombre d'imperfections. En particulier, les données sont souvent hétérogènes, c'est-à-dires qu'elles contiennent à la fois des informations quantitatives et qualitatives, incomplètes, lorsque certaines informations sont inaccessibles ou corrompues, et multi-sources, c'est-à-dire qu'elles résultent de l'agrégation de plusieurs jeux de données indépendant. Dans cette thèse, nous développons plusieurs méthodes pour l'analyse de données hétérogènes, incomplètes et multi-source. Nous nous attachons à étudier tous les aspects de ces méthodes, en fournissant des études théoriques précises, ainsi que des implémentations disponibles au public, et des évaluations empiriques. En particulier, nous considérons en détail deux applications issues de l'écologie pour la première et de la médecine pour la seconde. / In modern applications of statistics and machine learning, one often encounters many data imperfections. In particular, data are often heterogeneous, i.e. combine quantitative and qualitative information, incomplete, with missing values caused by machine failure or nonresponse phenomenons, and multi-source, when the data result from the compounding of diverse sources. In this dissertation, we develop several methods for the analysis of multi-source, heterogeneous and incomplete data. We provide a complete framework, and study all the aspects of the different methods, with thorough theoretical studies, open source implementations, and empirical evaluations. We study in details two particular applications from ecology and medical sciences.

The subject of this thesis is the maximization of the information divergence from an exponential family on a finite set, a problem first formulated by Nihat Ay. A special case is the maximization of the mutual information or the multiinformation between different parts of a composite system. My thesis contributes mainly to the mathematical aspects of the optimization problem. A reformulation is found that relates the maximization of the information divergence with the maximization of an entropic quantity, defined on the normal space of the exponential family. This reformulation simplifies calculations in concrete cases and gives theoretical insight about the general problem. A second emphasis of the thesis is on examples that demonstrate how the theoretical results can be applied in particular cases. Third, my thesis contain first results on the characterization of exponential families with a small maximum value of the information divergence.:1. Introduction 2. Exponential families 2.1. Exponential families, the convex support and the moment map 2.2. The closure of an exponential family 2.3. Algebraic exponential families 2.4. Hierarchical models 3. Maximizing the information divergence from an exponential family 3.1. The directional derivatives of D(*|E ) 3.2. Projection points and kernel distributions 3.3. The function DE 3.4. The first order optimality conditions of DE 3.5. The relation between D(*|E) and DE 3.6. Computing the critical points 3.7. Computing the projection points 4. Examples 4.1. Low-dimensional exponential families 4.1.1. Zero-dimensional exponential families 4.1.2. One-dimensional exponential families 4.1.3. One-dimensional exponential families on four states 4.1.4. Other low-dimensional exponential families 4.2. Partition models 4.3. Exponential families with max D(*|E ) = log(2) 4.4. Binary i.i.d. models and binomial models 5. Applications and Outlook 5.1. Principles of learning, complexity measures and constraints 5.2. Optimally approximating exponential families 5.3. Asymptotic behaviour of the empirical information divergence A. Polytopes and oriented matroids A.1. Polytopes A.2. Oriented matroids Bibliography Index Glossary of notations

In the thesis "On Boundaries of Statistical Models" problems related to a description of probability distributions with zeros, lying in the boundary of a statistical model, are treated. The distributions considered are joint distributions of finite collections of finite discrete random variables. Owing to this restriction, statistical models are subsets of finite dimensional real vector spaces. The support set problem for exponential families, the main class of models considered in the thesis, is to characterize the possible supports of distributions in the boundaries of these statistical models. It is shown that this problem is equivalent to a characterization of the face lattice of a convex polytope, called the convex support. The main tool for treating questions related to the boundary are implicit representations. Exponential families are shown to be sets of solutions of binomial equations, connected to an underlying combinatorial structure, called oriented matroid. Under an additional assumption these equations are polynomial and one is placed in the setting of commutative algebra and algebraic geometry. In this case one recovers results from algebraic statistics. The combinatorial theory of exponential families using oriented matroids makes the established connection between an exponential family and its convex support completely natural: Both are derived from the same oriented matroid. The second part of the thesis deals with hierarchical models, which are a special class of exponential families constructed from simplicial complexes. The main technical tool for their treatment in this thesis are so called elementary circuits. After their introduction, they are used to derive properties of the implicit representations of hierarchical models. Each elementary circuit gives an equation holding on the hierarchical model, and these equations are shown to be the "simplest", in the sense that the smallest degree among the equations corresponding to elementary circuits gives a lower bound on the degree of all equations characterizing the model. Translating this result back to polyhedral geometry yields a neighborliness property of marginal polytopes, the convex supports of hierarchical models. Elementary circuits of small support are related to independence statements holding between the random variables whose joint distributions the hierarchical model describes. Models for which the complete set of circuits consists of elementary circuits are shown to be described by totally unimodular matrices. The thesis also contains an analysis of the case of binary random variables. In this special situation, marginal polytopes can be represented as the convex hulls of linear codes. Among the results here is a classification of full-dimensional linear code polytopes in terms of their subgroups. If represented by polynomial equations, exponential families are the varieties of binomial prime ideals. The third part of the thesis describes tools to treat models defined by not necessarily prime binomial ideals. It follows from Eisenbud and Sturmfels'' results on binomial ideals that these models are unions of exponential families, and apart from solving the support set problem for each of these, one is faced with finding the decomposition. The thesis discusses algorithms for specialized treatment of binomial ideals, exploiting their combinatorial nature. The provided software package Binomials.m2 is shown to be able to compute very large primary decompositions, yielding a counterexample to a recent conjecture in algebraic statistics.

Tratar a função de verossimilhança perfilada como uma verossimilhança genuína pode levar a alguns problemas, como, por exemplo, inconsistência e ineficiência dos estimadores de máxima verossimilhança. Outro problema comum refere-se à aproximação usual da distribuição da estatística da razão de verossimilhanças pela distribuição qui-quadrado, que, dependendo da quantidade de parâmetros de perturbação, pode ser muito pobre. Desta forma, torna-se importante obter ajustes para tal função. Vários pesquisadores, incluindo Barndorff-Nielsen (1983,1994), Cox e Reid (1987,1992), McCullagh e Tibshirani (1990) e Stern (1997), propuseram modificações à função de verossimilhança perfilada. Tais ajustes consistem na incorporação de um termo à verossimilhança perfilada anteriormente à estimação e têm o efeito de diminuir os vieses da função escore e da informação. Este trabalho faz uma revisão desses ajustes e das aproximações para o ajuste de Barndorff-Nielsen (1983,1994) descritas em Severini (2000a). São apresentadas suas derivações, bem como suas propriedades. Para ilustrar suas aplicações, são derivados tais ajustes no contexto da família exponencial biparamétrica. Resultados de simulações de Monte Carlo são apresentados a fim de avaliar os desempenhos dos estimadores de máxima verossimilhança e dos testes da razão de verossimilhanças baseados em tais funções. Também são apresentadas aplicações dessas funções de verossimilhança em modelos não pertencentes à família exponencial biparamétrica, mais precisamente, na família de distribuições GA0(alfa,gama,L), usada para modelar dados de imagens de radar, e no modelo de Weibull, muito usado em aplicações da área da engenharia denominada confiabilidade, considerando dados completos e censurados. Aqui também foram obtidos resultados numéricos a fim de avaliar a qualidade dos ajustes sobre a verossimilhança perfilada, analogamente às simulações realizadas para a família exponencial biparamétrica. Vale mencionar que, no caso da família de distribuições GA0(alfa,gama,L), foi avaliada a aproximação da distribuição da estatística da razão de verossimilhanças sinalizada pela distribuição normal padrão. Além disso, no caso do modelo de Weibull, vale destacar que foram derivados resultados distribucionais relativos aos estimadores de máxima verossimilhança e às estatísticas da razão de verossimilhanças para dados completos e censurados, apresentados em apêndice. / The profile likelihood function is not genuine likelihood function, and profile maximum likelihood estimators are typically inefficient and inconsistent. Additionally, the null distribution of the likelihood ratio test statistic can be poorly approximated by the asymptotic chi-squared distribution in finite samples when there are nuisance parameters. It is thus important to obtain adjustments to the likelihood function. Several authors, including Barndorff-Nielsen (1983,1994), Cox and Reid (1987,1992), McCullagh and Tibshirani (1990) and Stern (1997), have proposed modifications to the profile likelihood function. They are defined in a such a way to reduce the score and information biases. In this dissertation, we review several profile likelihood adjustments and also approximations to the adjustments proposed by Barndorff-Nielsen (1983,1994), also described in Severini (2000a). We present derivations and the main properties of the different adjustments. We also obtain adjustments for likelihood-based inference in the two-parameter exponential family. Numerical results on estimation and testing are provided. We also consider models that do not belong to the two-parameter exponential family: the GA0(alfa,gama,L) family, which is commonly used to model image radar data, and the Weibull model, which is useful for reliability studies, the latter under both noncensored and censored data. Again, extensive numerical results are provided. It is noteworthy that, in the context of the GA0(alfa,gama,L) model, we have evaluated the approximation of the null distribution of the signalized likelihood ratio statistic by the standard normal distribution. Additionally, we have obtained distributional results for the Weibull case concerning the maximum likelihood estimators and the likelihood ratio statistic both for noncensored and censored data.

Tratar a função de verossimilhança perfilada como uma verossimilhança genuína pode levar a alguns problemas, como, por exemplo, inconsistência e ineficiência dos estimadores de máxima verossimilhança. Outro problema comum refere-se à aproximação usual da distribuição da estatística da razão de verossimilhanças pela distribuição qui-quadrado, que, dependendo da quantidade de parâmetros de perturbação, pode ser muito pobre. Desta forma, torna-se importante obter ajustes para tal função. Vários pesquisadores, incluindo Barndorff-Nielsen (1983,1994), Cox e Reid (1987,1992), McCullagh e Tibshirani (1990) e Stern (1997), propuseram modificações à função de verossimilhança perfilada. Tais ajustes consistem na incorporação de um termo à verossimilhança perfilada anteriormente à estimação e têm o efeito de diminuir os vieses da função escore e da informação. Este trabalho faz uma revisão desses ajustes e das aproximações para o ajuste de Barndorff-Nielsen (1983,1994) descritas em Severini (2000a). São apresentadas suas derivações, bem como suas propriedades. Para ilustrar suas aplicações, são derivados tais ajustes no contexto da família exponencial biparamétrica. Resultados de simulações de Monte Carlo são apresentados a fim de avaliar os desempenhos dos estimadores de máxima verossimilhança e dos testes da razão de verossimilhanças baseados em tais funções. Também são apresentadas aplicações dessas funções de verossimilhança em modelos não pertencentes à família exponencial biparamétrica, mais precisamente, na família de distribuições GA0(alfa,gama,L), usada para modelar dados de imagens de radar, e no modelo de Weibull, muito usado em aplicações da área da engenharia denominada confiabilidade, considerando dados completos e censurados. Aqui também foram obtidos resultados numéricos a fim de avaliar a qualidade dos ajustes sobre a verossimilhança perfilada, analogamente às simulações realizadas para a família exponencial biparamétrica. Vale mencionar que, no caso da família de distribuições GA0(alfa,gama,L), foi avaliada a aproximação da distribuição da estatística da razão de verossimilhanças sinalizada pela distribuição normal padrão. Além disso, no caso do modelo de Weibull, vale destacar que foram derivados resultados distribucionais relativos aos estimadores de máxima verossimilhança e às estatísticas da razão de verossimilhanças para dados completos e censurados, apresentados em apêndice. / The profile likelihood function is not genuine likelihood function, and profile maximum likelihood estimators are typically inefficient and inconsistent. Additionally, the null distribution of the likelihood ratio test statistic can be poorly approximated by the asymptotic chi-squared distribution in finite samples when there are nuisance parameters. It is thus important to obtain adjustments to the likelihood function. Several authors, including Barndorff-Nielsen (1983,1994), Cox and Reid (1987,1992), McCullagh and Tibshirani (1990) and Stern (1997), have proposed modifications to the profile likelihood function. They are defined in a such a way to reduce the score and information biases. In this dissertation, we review several profile likelihood adjustments and also approximations to the adjustments proposed by Barndorff-Nielsen (1983,1994), also described in Severini (2000a). We present derivations and the main properties of the different adjustments. We also obtain adjustments for likelihood-based inference in the two-parameter exponential family. Numerical results on estimation and testing are provided. We also consider models that do not belong to the two-parameter exponential family: the GA0(alfa,gama,L) family, which is commonly used to model image radar data, and the Weibull model, which is useful for reliability studies, the latter under both noncensored and censored data. Again, extensive numerical results are provided. It is noteworthy that, in the context of the GA0(alfa,gama,L) model, we have evaluated the approximation of the null distribution of the signalized likelihood ratio statistic by the standard normal distribution. Additionally, we have obtained distributional results for the Weibull case concerning the maximum likelihood estimators and the likelihood ratio statistic both for noncensored and censored data.