Métodos de agrupamento hierárquico de textos são muito úteis para analisar o conhecimento embutido em coleções textuais, organizando os documentos textuais em grupos e subgrupos para facilitar a exploração do conhecimento em diversos níveis de granularidade. Tais métodos pertencem à área de aprendizado não supervisionado de máquina, uma que vez obtêm modelos de agrupamento apenas pela observação de regularidades existentes na coleção textual, sem supervisão humana. Os métodos tradicionais de agrupamento assumem que a coleção textual é representada apenas pela informação técnica, ou seja, palavras e frases extraídas diretamente dos textos. Por outro lado, em muitas tarefas de agrupamento existe conhecimento adicional e valioso a respeito dos dados, geralmente extraído por um processo avançado com apoio de usuários especialistas do domínio do problema. Devido ao alto custo para obtenção desses dados, esta informação adicional é definida como privilegiada e usualmente está disponível para representar apenas um subconjunto dos documentos textuais. Recentemente, um novo paradigma de aprendizado de máquina denominado LUPI (Learning Using Privileged Information) foi proposto por Vapnik para incorporar informação privilegiada em métodos aprendizado supervisionado. Neste trabalho de doutorado, o paradigma LUPI foi estendido para aprendizado não supervisionado, em especial, para agrupamento hierárquico de textos. Foram propostas e avaliadas abordagens para lidar com diferentes desafios existentes em tarefas de agrupamento, envolvendo a extração e estruturação da informação privilegiada e seu uso para refinar ou corrigir modelos de agrupamento. As abordagens propostas se mostraram eficazes em (i) consenso de agrupamentos, permitindo combinar diferentes representações e soluções de agrupamento; (ii) aprendizado de métricas, em que medidas de proximidades mais robustas foram obtidas com base na informação privilegiada; e (iii) seleção de modelos, em que a informação privilegiada é explorada para identificar relevantes estruturas de agrupamento hierárquico. Todas as abordagens apresentadas foram investigadas em um cenário de agrupamento incremental, permitindo seu uso em aplicações práticas caracterizadas pela necessidade de eficiência computacional e alta frequência de publicação de novo conhecimento textual. / Hierarchical text clustering methods are very useful to analyze the implicit knowledge in textual collections, enabling the organization of textual documents into clusters and subclusters to facilitate the knowledge browsing at various levels of granularity. Such methods are classified as unsupervised machine learning, since the clustering models are obtained only by observing regularities of textual data without human supervision. Traditional clustering methods assume that the text collection is represented only by the technical information, i.e., words and phrases extracted directly from the texts. On the other hand, in many text clustering tasks there is an additional and valuable knowledge about the problem domain, usually extracted by an advanced process with support of the domain experts. Due to the high cost of obtaining such expert knowledge, this additional information is defined as privileged and is usually available to represent only a subset of the textual documents. Recently, a new machine learning paradigm called LUPI (Learning Using Privileged Information) was proposed by Vapnik to incorporate privileged information into supervised learning methods. In this thesis, the LUPI paradigm was extended to unsupervised learning setting, in particular for hierarchical text clustering. We propose and evaluate approaches to deal with different challenges for clustering tasks, involving the extraction and structuring of privileged information and using this additional information to refine or correct clustering models. The proposed approaches were effective in (i) consensus clustering, allowing to combine different clustering solutions and textual representations; (ii) metric learning, in which more robust proximity measures are obtained from privileged information; and (iii) model selection, in which the privileged information is exploited to identify the relevant structures of hierarchical clustering. All the approaches presented in this thesis were investigated in an incremental clustering scenario, allowing its use in practical applications that require computational efficiency as well as deal with high frequency of publication of new textual knowledge.

Os recentes avanços da ciência e tecnologia viabilizaram o crescimento de dados em quantidade e disponibilidade. Junto com essa explosão de informações geradas, surge a necessidade de analisar dados para descobrir conhecimento novo e útil. Desse modo, áreas que visam extrair conhecimento e informações úteis de grandes conjuntos de dados se tornaram grandes oportunidades para o avanço de pesquisas, tal como o Aprendizado de Máquina (AM) e a Mineração de Dados (MD). Porém, existem algumas limitações que podem prejudicar a acurácia de alguns algoritmos tradicionais dessas áreas, por exemplo o desbalanceamento das amostras das classes de um conjunto de dados. Para mitigar tal problema, algumas alternativas têm sido alvos de pesquisas nos últimos anos, tal como o desenvolvimento de técnicas para o balanceamento artificial de dados, a modificação dos algoritmos e propostas de abordagens para dados desbalanceados. Uma área pouco explorada sob a visão do desbalanceamento de dados são os problemas de classificação hierárquica, em que as classes são organizadas em hierarquias, normalmente na forma de árvore ou DAG (Direct Acyclic Graph). O objetivo deste trabalho foi investigar as limitações e maneiras de minimizar os efeitos de dados desbalanceados em problemas de classificação hierárquica. Os experimentos realizados mostram que é necessário levar em consideração as características das classes hierárquicas para a aplicação (ou não) de técnicas para tratar problemas dados desbalanceados em classificação hierárquica. / Recent advances in science and technology have made possible the data growth in quantity and availability. Along with this explosion of generated information, there is a need to analyze data to discover new and useful knowledge. Thus, areas for extracting knowledge and useful information in large datasets have become great opportunities for the advancement of research, such as Machine Learning (ML) and Data Mining (DM). However, there are some limitations that may reduce the accuracy of some traditional algorithms of these areas, for example the imbalance of classes samples in a dataset. To mitigate this drawback, some solutions have been the target of research in recent years, such as the development of techniques for artificial balancing data, algorithm modification and new approaches for imbalanced data. An area little explored in the data imbalance vision are the problems of hierarchical classification, in which the classes are organized into hierarchies, commonly in the form of tree or DAG (Direct Acyclic Graph). The goal of this work aims at investigating the limitations and approaches to minimize the effects of imbalanced data with hierarchical classification problems. The experimental results show the need to take into account the features of hierarchical classes when deciding the application of techniques for imbalanced data in hierarchical classification.

Bancos de dados relacionais são fontes de dados rigidamente estruturadas, caracterizadas por relacionamentos complexos entre um conjunto de relações (tabelas). Entender tais relacionamentos é um desafio, porque os usuários precisam considerar múltiplas relações, entender restrições de integridade, interpretar vários atributos, e construir consultas SQL para cada tentativa de exploração. Neste cenário, introduz-se uma metodologia em duas etapas; primeiro utiliza-se um grafo organizado como uma estrutura hierárquica para modelar os relacionamentos do banco de dados, e então, propõe-se uma nova técnica de visualização para exploração relacional. Os resultados demonstram que a proposta torna a exploração de bases de dados significativamente simplificada, pois o usuário pode navegar visualmente pelos dados com pouco ou nenhum conhecimento sobre a estrutura subjacente. Além disso, a navegação visual de dados remove a necessidade de consultas SQL, e de toda complexidade que elas requerem. Acredita-se que esta abordagem possa trazer um paradigma inovador no que tange à compreensão de dados relacionais / Relational databases are rigid-structured data sources characterized by complex relationships among a set of relations (tables). Making sense of such relationships is a challenging problem because users must consider multiple relations, understand their ensemble of integrity constraints, interpret dozens of attributes, and draw complex SQL queries for each desired data exploration. In this scenario, we introduce a twofold methodology; we use a hierarchical graph representation to efficiently model the database relationships and, on top of it, we designed a visualization technique for rapidly relational exploration. Our results demonstrate that the exploration of databases is deeply simplified as the user is able to visually browse the data with little or no knowledge about its structure, dismissing the need of complex SQL queries. We believe our findings will bring a novel paradigm in what concerns relational data comprehension.

The major concern of this work is the development of hierarchically porous structured zeolite composites for ion-exchange applications by deposition of a thin layer of zeolite on inexpensive porous supports which offers better efficiency in separation processes. The merits of utilization of zeolite composites in industrial applications are generally reducing mass transfer resistance and pressure drop. In addition to this they have advantages in the removal of metal ions from wastewater such as increasing the metals uptake and minimizing the volume of waste disposed especially after vitrification. This thesis presents results from a combination of experimental work and simulation study of experimental data to give isotherm and kinetic models. The experimental work shed light on the preparation of zeolite composites using zeolite X (Si/Al ~ 1.35) and clinoptilolite (Si/Al ~ 4.3), studying the performance of these composites on the removal of the Sr2+ and Mn2+ ions and then stabilization of waste materials resulting from the ion-exchange process. Clinoptilolite was hydrothermally synthesized to show the effect of non-framework cations on the removal process. The porous supports were diatomite which is naturally occurring silica and carbon which is obtained from Iraqi date stones by a thermal treatment conducted at 900°C. Coating the support surface with zeolites crystals was conducted in two different ways. The layer by layer approach, which has not previously been used, was used to prepare monolithic carbon clinoptilolite composite using a combination of sucrose/citric acid and zeolite. The other approach was modifying the support surface by ultrasonication in the presence of nanoparticles suspension prepared using ball mill to create nucleation sites and enhance the crystal attachment during hydrothermal treatment. Characterisation was implemented in each case using XRD, SEM, EDAX, TGA and BET method. Ion-exchange experimental results showed higher ion-exchange capacity obtained when the composites were used in comparison to pure zeolites, when a comparison is based on actual weight of zeolite used for removal of Sr2+ and Mn2+ ions. A study of encapsulation of ions showed that it is feasible to solidify the waste materials by vitrification and/or geopolymerization to eliminate leaching of ions to the environment. The simulation studies showed that the ion-exchange kinetic followed the pseudo second order kinetic model. This fitting indicates that the rate of ion-exchange process is controlled by a chemical reaction related to valence forces. The overall ion-exchange process is controlled by a combination of ion-exchange reaction, film diffusion and intra-particle diffusion. Moreover, the thermodynamic studies which were conducted under different temperatures revealed that the ion-exchange of Sr2+ and Mn2+ ions is practicable, spontaneous and endothermic.

This doctoral thesis outlines several methodological advances in network science aimed towards uncovering rapid, complex interdependencies of electromagnetic brain activity recorded from the Electroencephalogram (EEG). This entails both new analyses and modelling of EEG brain network topologies and a novel approach to analyse rapid dynamics of connectivity. Importantly, we implement these advances to provide novel insights into pathological brain function in Alzheimer's disease. We introduce the concept of hierarchical complexity of network topology, providing both an index to measure it and a model to simulate it. We then show that the topology of functional connectivity estimated from EEG recordings is hierarchically complex, existing in a scale between random and star-like topologies, this is a paradigm shift from the established understanding that complexity arises between random and regular topologies. We go on to consider the density appropriate for binarisation of EEG functional connectivity, a methodological step recommended to produce compact and unbiased networks, in light of its new-found hierarchical complexity. Through simulations and real EEG data, we show the benefit of going beyond often recommended sparse representations to account for a broader range of hierarchy level interactions. After this, we turn our attention to assessing dynamic changes in connectivity. By constructing a unified framework for multivariate signals and graphs, inspired by network science and graph signal processing, we introduce graph-variate signal analysis which allows us to capture rapid fluctuations in connectivity robust to spurious short-term correlations. We define this for three pertinent brain connectivity estimates - Pearson's correlation coefficient, coherence and phase-lag index - and show its benefit over standard dynamic connectivity measures in a range of simulations and real data. Applying these novel methods to EEG datasets of the performance of visual short-term memory binding tasks by familial and sporadic Alzheimer's disease patients, we uncover disorganisation of the topological hierarchy of EEG brain function and abnormalities of transient phase-based activity which paves the way for new interpretations of the disease's affect on brain function. Hierarchical complexity and graph-variate dynamic connectivity are entirely new methods for analysing EEG brain networks. The former provides new interpretations of complexity in static connectivity patterns while the latter enables robust analysis of transient temporal connectivity patterns, both at the frontiers of analysis. Although designed with EEG functional connectivity in mind, we hope these techniques will be picked up in the broader field, having consequences for research into complex networks in general.

Os métodos de estimação para modelos de fragilidade vêm sendo bastante discutidos na literatura estatística devido a sua grande utilização em estudos de Análise de Sobrevivência. Vários métodos de estimação de parâmetros dos modelos foram desenvolvidos: procedimentos de estimação baseados no algoritmo EM, cadeias de Markov de Monte Carlo, processos de estimação usando verossimilhança parcial, verossimilhança penalizada, quasi-verossimilhança, entro outros. Uma alternativa que vem sendo utilizada atualmente é a utilização da verossimilhança hierárquica. O objetivo principal deste trabalho foi estudar as vantagens e desvantagens da verossimilhança hierárquica para a inferência em modelos de fragilidade em relação a verossimilhança penalizada, método atualmente mais utilizado. Nós aplicamos as duas metodologias a um banco de dados real, utilizando os pacotes estatísticos disponíveis no software R, e fizemos um estudo de simulação, visando comparar o viés e o erro quadrático médio das estimativas de cada abordagem. Pelos resultados encontrados, as duas metodologias apresentaram estimativas muito próximas, principalmente para os termos fixos. Do ponto de vista prático, a maior diferença encontrada foi o tempo de execução do algoritmo de estimação, muito maior na abordagem hierárquica. / Estimation procedures for frailty models have been widely discussed in the statistical literature due its widespread use in survival studies. Several estimation methods were developed: procedures based on the EM algorithm, Monte Carlo Markov chains, estimation processes based on parcial likelihood, penalized likelihood and quasi-likelihood etc. An alternative currently used is the hierarchical likelihood. The main objective of this work was to study the hierarchical likelihood advantages and disadvantages for inference in frailty models when compared with the penalized likelihood method, which is the most used one. We applied both approaches to a real data set, using R packages available. Besides, we performed a simulation study in order to compare the methods through out the bias and the mean square error of the estimators. Both methodologies presented very similar estimates, mainly for the fixed effects. In practice, the great difference was the computational cost, much higher in the hierarchical approach.

A definição das estruturas hierárquicas envolve uma estrutura que pode ser observada em escalas de diversos comprimentos, sendo que um elemento estrutural de certa escala é formado por subestruturas periódicas de uma escala menor. Utilizando o método bidirecional de otimização topológica evolucionária BESO (Bi-directional Evolutionary Structural Optimization) em estruturas contínuas compostas por materiais únicos e por vários materiais, modeladas através do método dos elementos finitos, este trabalho implementa os procedimentos computacionais necessários com o objetivo de otimizar o comportamento dinâmico do sistema estrutural, através da maximização da frequência fundamental bem como da separação de um ou dois pares de frequências adjacentes de forma simultânea ou não, sujeita a restrições de volume estrutural nas diversas fases. Cada nível hierárquico é assumido como um meio contínuo composto por um ou mais materiais homogêneos, cada um destes com uma microestrutura associada. No projeto simultâneo com múltiplas fases, as informações foram transferidas entre a micro e a macroescala através do método da homogeneização, enquanto que as técnicas de otimização topológica visaram encontrar a melhor distribuição de fases em ambas as escalas para a maximização das propriedades desejadas Dessa forma se alcançaram uma série de topologias associadas às diversas funções objetivo utilizadas, decorrentes da maximização da frequência fundamental, do intervalo entre um par de frequências naturais consecutivas, da separação das frequências naturais a partir de uma frequência prescrita e da separação de dois pares de frequências consecutivas concomitantemente. Experimentos numéricos também foram realizados buscando o melhor leiaute na macroescala, na microescala, ou em ambas de forma acoplada, apresentando-se as discussões correspondentes. Conjunto de soluções ótimas foram gerados, baseado no método dos pesos, os quais possibilitaram por exemplo a identificação da perda de integridade estrutural em alguns casos otimizados. Foram obtidas estruturas com valores de separação entre duas frequências consecutivas muito maiores do que nas topologias não otimizadas. Por exemplo quando da otimização utilizando microestrutura única com dois materiais, a maximização do intervalo entre a terceira e a segunda frequências naturais supera em aproximadamente 520% a diferença entre as mesmas frequências na topologia não otimizada. / Hierarchical structures are structures that can be observed at different length scales, where typically a structural element at a certain scale is composed by periodic substructures at a smaller scale. Applying the Bi-directional Evolutionary Structural Optimization (BESO) method to continuous structures made of a single or multiple materials modelled via the finite element method, this work implements the computational procedures needed in order to optimize the dynamic behavior of the structural system. The optimization is achieved either via maximization of the fundamental frequency, or via separation of adjacent natural frequencies, as well as via separation of one or two pairs of adjacent frequencies, simultaneously or not, all former cases subjected to volume restrictions in the different material phases. Each hierarchical level is treated as a continuous medium occupied by one or more homogeneous material, each material having an associated microstructure. In the simultaneous project with multiple phases, the information is transferred from the microstructure to the macrostructure through the homogenization method, while topology optimization techniques are employed to reach the best material distribution such that the chose objective function is maximized. In this way, a series of topologies associated to each optimization type were found, from the maximization of either the fundamental frequency, the gap between a pair of adjacent natural frequencies, the distance of all natural frequencies from a prescribed frequency, or the gap of two pairs of adjacent natural frequencies concurrently Numerical experiments were conducted in order to find the best layout for the macrostructure, microstructure, or both simultaneously via a coupled formulation. Following the results a discussion is presented. Sets of optimal solutions based on the weighed method were generated, making possible to identify the loss of structural integrity in some optimized cases. It was possible to obtain structures with separation values between two consecutive frequencies much higher than the initial values in unoptimized topologies. For example, the optimization of a single microstructure containing two materials reached a gap maximization between the third and second natural frequencies approximately 520% bigger than the difference between these same frequencies in the unoptimized topology.

L'équation intégrale du champ électrique (EFIE) et l'équation intégrale du champ combiné (CFIE) souffrent d'un mauvais conditionnement à haute discrétisation et à bassefréquence : si la taille moyenne des arrêtes du maillage est réduite ou si la fréquence est diminuée le conditionnement du système se dégrade rapidement. Cela provoque le ralentissement ou la non convergence des solveurs itératifs. Cette dissertation présente de nouveaux paradigmes permettant l'obtention de solveurs à convergence rapide pour équations intégrales; pour prévenir la dégradation du conditionnement nous avançons l'état de l'art des techniques de préconditionnement dites de Calderon et de celles reposant sur l'utilisation des bases hiérarchiques. Pour traiter l'EFIE, nous introduisons une base hiérarchique pour maillages structurés et non-structurés dérivant des pré-ondelettes primaires et duales de Haar. De plus, nous introduisons un nouveau cadre permettant de préconditionner efficacement l'EFIE dans le cas d'objets à connexion multiples. L'applicabilité à la CFIE des préconditionneurs à bases hiérarchiques fait l'objet d'une étude aboutissant à la formalisation d'une technique de préconditionnement. Nous présentons aussi un préconditionneur multiplicatif de type Calderon (RF-CMP) qui permet l'obtention d'une matrice système Hermitienne, définie positive (HDP) et bien conditionnée, sans avoir recours, contrairement aux préconditionneurs existants, au raffinement du maillage ni à l'utilisation de fonction duales. Puisque la matrice est HPD, la méthode du gradient conjugué peut servir de solveur itératif avec une convergence garantie. / The electric field integral equation (EFIE) and the combined field integral equation(CFIE) suffer from the dense-discretization and the low-frequency breakdown: if the average edgelength of the mesh is reduced, or if the frequency is decreased, then the condition number of the system matrix grows. This leads to slowly or non-converging iterative solvers. This dissertation presents new paradigms for rapidly converging integral equation solvers: to overcome the illconditioning, we advance and extend the state of the art both in hierarchical basis and in Calderón preconditioning techniques. For the EFIE, we introduce a hierarchical basis for structured and unstructured meshes based on generalized primal and dual Haar prewavelets. Furthermore, a framework is introduced which renders the hierarchical basis able to efficiently precondition the EFIE in the case that the scatterer is multiply connected. The applicability of hierarchical basis preconditioners to the CFIE is analyzed and an efficient preconditioning scheme is derived. In addition, we present a refinement-free Calderón multiplicative preconditioner (RF-CMP) that yields a system matrix which is Hermitian, positive definite (HPD), and well-conditioned. Different from existing Calderón preconditioners, no dual basis functions and thus no refinement of the mesh is required. Since the matrix is HPD—in contrast to standard discretizations of the EFIE—we can apply the conjugate gradient (CG) method as iterative solver, which guarantees convergence. Eventually, the RF-CMP is extended to the CFIE.

One single primary study is only a little piece of a bigger puzzle. Meta-analysis is the statistical combination of results from primary studies that address a similar question. The most general case is the random-effects model, in where it is assumed that for each study the vector of outcomes T_i~N(θ_i,Σ_i ) and that the vector of true-effects for each study is θ_i~N(θ,Ψ). Since each θ_i is a nuisance parameter, inferences are based on the marginal model T_i~N(θ,Σ_i+Ψ). The main goal of a meta-analysis is to obtain estimates of θ, the sampling error of this estimate and Ψ. Standard meta-analysis techniques assume that Σ_i is known and fixed, allowing the explicit modeling of its elements and the use of Generalized Least Squares as the method of estimation. Furthermore, one can construct the variance-covariance matrix of standard errors and build confidence intervals or ellipses for the vector of pooled estimates. In practice, each Σ_i is estimated from the data using a matrix function that depends on the unknown vector θ_i. Some alternative methods have been proposed in where explicit modeling of the elements of Σ_i is not needed. However, estimation of between-studies variability Ψ depends on the within-study variance Σ_i, as well as other factors, thus not modeling explicitly the elements of Σ_i and departure of a hierarchical structure has implications on the estimation of Ψ. In this dissertation, I develop an alternative model for random-effects meta-analysis based on the theory of hierarchical models. Motivated, primarily, by Hoaglin's article "We know less than we should about methods of meta-analysis", I take into consideration that each Σ_i is unknown and estimated by using a matrix function of the corresponding unknown vector θ_i. I propose an estimation method based on the Minimum Covariance Estimator and derive formulas for the expected marginal variance for two effect sizes, namely, Pearson's moment correlation and standardized mean difference. I show through simulation studies that the proposed model and estimation method give accurate results for both univariate and bivariate meta-analyses of these effect-sizes, and compare this new approach to the standard meta-analysis method.

Thematic unity in music occurs when elements from a musical idea appear frequently, in significant places and their presence is recognized or experienced on or beneath the surface. In fugal compositions, thematic unity is evident in the opening statement of the subject and it permeates each layer of its texture. Three analytical perspectives are used to investigate the degree to which local thematic material anticipates later structural features in Johan Sebastian Bach's Fugue in G minor WTC II, and Georg Frederic Handel's Fuga II in G Major. The analytical perspectives identify: (1) cohesive relationships between motivic fragments, (2) underlying motives and their relationships to keys and harmonic progressions, and (3) voice leading reductions relative to linear and tonal prolongation. Arnold Schoenberg, Hans Keller, and Rudolph Reti provide valuable insights concerning the organic nature of thematic material. The voice leading reductions of Heinrich Schenker and William Renwick offer procedures that reveal underlying thematic relationships. The cohesive elements of the selected fugues will be explained with reference to immediate and long-range relationships.