Value of information is one of the key features of decision analysis. This work deals with providing a consistent and functional methodology to determine VOI on proposed well tests in the presence of uncertainties. This method strives to show that VOI analysis with the help of discretized versions of continuous probability distributions with conventional decision trees can be very accurate if the optimal method of discrete approximation is chosen rather than opting for methods such as Monte Carlo simulation to determine the VOI. This need not necessarily mean loss of accuracy at the cost of simplifying probability calculations. Both the prior and posterior probability distributions are assumed to be continuous and are discretized to find the VOI. This results in two steps of discretizations in the decision tree. Another interesting feature is that there lies a level of decision making between the two discrete approximations in the decision tree. This sets it apart from conventional discretized models since the accuracy in this case does not follow the rules and conventions that normal discrete models follow because of the decision between the two discrete approximations. The initial part of the work deals with varying the number of points chosen in the discrete model to test their accuracy against different correlation coefficients between the information and the actual values. The latter part deals more with comparing different methods of existing discretization methods and establishing conditions under which each is optimal. The problem is comprehensively dealt with in the cases of both a risk neutral and a risk averse decision maker. / text

Computer models or simulators are becoming increasingly common in many fields in science and engineering, powered by the phenomenal growth in computer hardware over the past decades. Many of these simulators implement a particular mathematical model as a deterministic computer code, meaning that running the simulator again with the same input gives the same output. Often running the code involves some computationally expensive tasks, such as solving complex systems of partial differential equations numerically. When simulator runs become too long, it may limit their usefulness. In order to overcome time or budget constraints by making the most out of limited computational resources, a statistical methodology has been proposed, known as the "Design and Analysis of Computer Experiments". The main idea is to run the expensive simulator only at a relatively few, carefully chosen design points in the input space, and based on the outputs construct an emulator (statistical model) that can emulate (predict) the output at new, untried locations at a fraction of the cost. This approach is useful provided that we can measure how much the predictions of the cheap emulator deviate from the real response surface of the original computer model. One way to quantify emulator error is to construct pointwise prediction bands designed to envelope the response surface and make assertions that the true response (simulator output) is enclosed by these envelopes with a certain probability. Of course, to be able to make such probabilistic statements, one needs to introduce some kind of randomness. A common strategy that we use here is to model the computer code as a random function, also known as a Gaussian stochastic process. We concern ourselves with smooth response surfaces and use the Gaussian covariance function that is ideal in cases when the response function is infinitely differentiable. In this thesis, we propose Fast Bayesian Inference (FBI) that is both computationally efficient and can be implemented as a black box. Simulation results show that it can achieve remarkably accurate prediction uncertainty assessments in terms of matching coverage probabilities of the prediction bands and the associated reparameterizations can also help parameter uncertainty assessments.

Phylogenetics is the study of the evolutionary relationship between species. Inference of phylogeny relies heavily on statistical models that have been extended and refined tremendously over the past years into very complex hierarchical models. Paper I introduces probabilistic graphical models to statistical phylogenetics and elaborates on the potential advantages a unified graphical model representation could have for the community, e.g., by facilitating communication and improving reproducibility of statistical analyses of phylogeny and evolution. Once the phylogeny is reconstructed it is possible to infer the rates of diversification (speciation and extinction). In this thesis I extend the birth-death process model, so that it can be applied to incompletely sampled phylogenies, that is, phylogenies of only a subsample of the presently living species from one group. Previous work only considered the case when every species had the same probability to be included and here I examine two alternative sampling schemes: diversified taxon sampling and cluster sampling. Paper II introduces these sampling schemes under a constant rate birth-death process and gives the probability density for reconstructed phylogenies. These models are extended in Paper IV to time-dependent diversification rates, again, under different sampling schemes and applied to empirical phylogenies. Paper III focuses on fast and unbiased simulations of reconstructed phylogenies. The efficiency is achieved by deriving the analytical distribution and density function of the speciation times in the reconstructed phylogeny. / <p>At the time of the doctoral defense, the following papers were unpublished and had a status as follows: Paper 1: Manuscript. Paper 4: Accepted.</p>

In this dissertation, we explore parallel algorithms for general N-Body problems in high dimensions, and their applications in machine learning and image analysis on distributed infrastructures. In the first part of this work, we proposed and developed a set of basic tools built on top of Message Passing Interface and OpenMP for massively parallel nearest neighbors search. In particular, we present a distributed tree structure to index data in arbitrary number of dimensions, and a novel algorithm that eliminate the need for collective coordinate exchanges during tree construction. To the best of our knowledge, our nearest neighbors package is the first attempt that scales to millions of cores in up to a thousand dimensions. Based on our nearest neighbors search algorithms, we present "ASKIT", a parallel fast kernel summation tree code with a new near-far field decomposition and a new compact representation for the far field. Specially our algorithm is kernel independent. The efficiency of new near far decomposition depends only on the intrinsic dimensionality of data, and the new far field representation only relies on the rand of sub-blocks of the kernel matrix. In the second part, we developed a Bayesian inference framework and a variational formulation for a MAP estimation of the label field for medical image segmentation. In particular, we propose new representations for both likelihood probability and prior probability functions, as well as their fast calculation. Then a parallel matrix free optimization algorithm is given to solve the MAP estimation. Our new prior function is suitable for lots of spatial inverse problems. Experimental results show our framework is robust to noise, variations of shapes and artifacts.

Computer models or simulators are becoming increasingly common in many fields in science and engineering, powered by the phenomenal growth in computer hardware over the past decades. Many of these simulators implement a particular mathematical model as a deterministic computer code, meaning that running the simulator again with the same input gives the same output. Often running the code involves some computationally expensive tasks, such as solving complex systems of partial differential equations numerically. When simulator runs become too long, it may limit their usefulness. In order to overcome time or budget constraints by making the most out of limited computational resources, a statistical methodology has been proposed, known as the "Design and Analysis of Computer Experiments". The main idea is to run the expensive simulator only at a relatively few, carefully chosen design points in the input space, and based on the outputs construct an emulator (statistical model) that can emulate (predict) the output at new, untried locations at a fraction of the cost. This approach is useful provided that we can measure how much the predictions of the cheap emulator deviate from the real response surface of the original computer model. One way to quantify emulator error is to construct pointwise prediction bands designed to envelope the response surface and make assertions that the true response (simulator output) is enclosed by these envelopes with a certain probability. Of course, to be able to make such probabilistic statements, one needs to introduce some kind of randomness. A common strategy that we use here is to model the computer code as a random function, also known as a Gaussian stochastic process. We concern ourselves with smooth response surfaces and use the Gaussian covariance function that is ideal in cases when the response function is infinitely differentiable. In this thesis, we propose Fast Bayesian Inference (FBI) that is both computationally efficient and can be implemented as a black box. Simulation results show that it can achieve remarkably accurate prediction uncertainty assessments in terms of matching coverage probabilities of the prediction bands and the associated reparameterizations can also help parameter uncertainty assessments.

An important mechanism during nervous system development is the guidance of axons by chemical gradients. The structure responsible for responding to chemical cues in the embryonic environment is the axonal growth cone -- a structure combining sensory and motor functions to direct axon growth. In this thesis, we develop a series of mathematical models for the gradient-based guidance of axonal growth cones, based on the idea that growth cones might be optimised for such a task. In particular, we study axon guidance from the framework of Bayesian decision theory, an approach that has recently proved to be very successful in understanding higher level sensory processing problems. We build our models in complexity, beginning with a one-dimensional array of chemoreceptors simply trying to decide whether an external gradient points to the right or the left. Even with this highly simplified model, we can obtain a good fit of theory to experiment. Furthermore, we find that the information a growth cone can obtain about the locations of its receptors has a strong influence on the functional dependence of gradient sensing performance on average concentration. We find that the shape of the sensitivity curve is robust to changes in the precise inference strategy used to determine gradient detection, and depends only on the information the growth cone can obtain about the locations of its receptors. We then consider the optimal distribution of guidance cues for guidance over long range, and find that the same upper limit on guidance distance is reached regardless of whether only bound, or only unbound receptors signal. We also discuss how information from multiple cues ought to be combined for optimal guidance. In chapters 5 and 6, we extend our model to two-dimensions, and to explicitly include temporal dynamics. The two-dimensional case yields results which are essentially equivalent to the one dimensional model. In contrast, explicitly including temporal dynamics in our leads to some significant departures from the one-dimensional and two-dimensional models, depending on the timescales over which various processes operate. Overall, we suggest that decision theory, in addition to providing a useful normative approach to studying growth cone chemotaxis, might provide a framework for understanding some of the biochemical pathways involved in growth cone chemotaxis, and in the chemotaxis of other eukaryotic cells.

abstract: Bayesian networks are powerful tools in system reliability assessment due to their flexibility in modeling the reliability structure of complex systems. This dissertation develops Bayesian network models for system reliability analysis through the use of Bayesian inference techniques. Bayesian networks generalize fault trees by allowing components and subsystems to be related by conditional probabilities instead of deterministic relationships; thus, they provide analytical advantages to the situation when the failure structure is not well understood, especially during the product design stage. In order to tackle this problem, one needs to utilize auxiliary information such as the reliability information from similar products and domain expertise. For this purpose, a Bayesian network approach is proposed to incorporate data from functional analysis and parent products. The functions with low reliability and their impact on other functions in the network are identified, so that design changes can be suggested for system reliability improvement. A complex system does not necessarily have all components being monitored at the same time, causing another challenge in the reliability assessment problem. Sometimes there are a limited number of sensors deployed in the system to monitor the states of some components or subsystems, but not all of them. Data simultaneously collected from multiple sensors on the same system are analyzed using a Bayesian network approach, and the conditional probabilities of the network are estimated by combining failure information and expert opinions at both system and component levels. Several data scenarios with discrete, continuous and hybrid data (both discrete and continuous data) are analyzed. Posterior distributions of the reliability parameters of the system and components are assessed using simultaneous data. Finally, a Bayesian framework is proposed to incorporate different sources of prior information and reconcile these different sources, including expert opinions and component information, in order to form a prior distribution for the system. Incorporating expert opinion in the form of pseudo-observations substantially simplifies statistical modeling, as opposed to the pooling techniques and supra Bayesian methods used for combining prior distributions in the literature. The methods proposed are demonstrated with several case studies. / Dissertation/Thesis / Doctoral Dissertation Industrial Engineering 2016

This thesis is divided into two parts. In the first part, we show how problems of statistical inference and combinatorial optimization may be approached within a unified framework that employs tools from fields as diverse as machine learning, statistical physics and information theory, allowing us to i) design algorithms to solve the problems, ii) analyze the performance of these algorithms both empirically and analytically, and iii) to compare the results obtained with the optimal achievable ones. In the second part, we use this framework to study two specific problems, one of inference (compressed sensing) and the other of optimization (information hiding). In both cases, we review current approaches, identify their flaws, and propose new schemes to address these flaws, building on the use of message-passing algorithms, variational inference techniques, and spin glass models from statistical physics. / Esta tese está dividida em duas partes. Na primeira delas, mostramos como problemas de inferência estatística e de otimização combinatória podem ser abordados sob um framework unificado que usa ferramentas de áreas tão diversas quanto o aprendizado de máquina, a física estatística e a teoria de informação, permitindo que i) projetemos algoritmos para resolver os problemas, ii) analisemos a performance destes algoritmos tanto empiricamente como analiticamente, e iii) comparemos os resultados obtidos com os limites teóricos. Na segunda parte, este framework é usado no estudo de dois problemas específicos, um de inferência (compressed sensing) e outro de otimização (ocultação de dados). Em ambos os casos, revisamos abordagens recentes, identificamos suas falhas, e propomos novos esquemas que visam corrigir estas falhas, baseando-nos sobretudo em algoritmos de troca de mensagens, técnicas de inferência variacional, e modelos de vidro de spin da física estatística.

Este trabalho apresenta três principais contribuições: i. a proposta de uma nova lógica de descrição probabilística; ii. um novo algoritmo de inferência em primeira ordem a ser utilizado em terminologias representadas nessa lógica; e iii. aplicações práticas em problemas reais. A lógica aqui proposta, crALC (credal ALC), adiciona inclusões probabilísticas na popular lógica ALC combinando as terminologias com condições de aciclicidade, de Markov, e adotando uma semântica baseada em interpretações. Como os métodos de inferência exata tradicionalmente apresentam problemas de escalabilidade devido à presença de quantificadores (restrições universal e existencial), apresentamos um algoritmo de loopy propagation em primeira-ordem que se comporta bem para terminologias com domínios não triviais. Uma série de testes foi feita com o algoritmo proposto em comparação com algoritmos tradicionais da literatura; os resultados apresentados mostram uma clara vantagem em relação aos outros algoritmos. São apresentadas ainda duas aplicações da lógica e do algoritmo para resolver problemas reais da área de robótica móvel. Embora os problemas tratados sejam relativamente simples, eles constituem a base de muitos outros problemas da área, sendo um passo importante na representação de conhecimento de agentes/robôs autônomos e no raciocínio sobre esse conhecimento. / This work presents two major contributions: i. a new probabilistic description logic; ii. a new algorithm for inference in terminologies expressed in this logic; iii. practical applications in real tasks. The proposed logic, referred to as crALC (credal ALC), adds probabilistic inclusions to the popular logic ALC, combining the usual acyclicity and Markov conditions, and adopting interpretation-based semantics. As exact inference does not seem scalable due to the presence of quantifiers (existential and universal), we present a first-order loopy propagation algorithm that behaves appropriately for non-trivial domain sizes. A series of tests were done comparing the performance of the proposed algorithm against traditional ones; the presented results are favorable to the first-order algorithm. Two applications in the field of mobile robotics are presented, using the new probabilistic logic and the inference algorithm. Though the problems can be considered simple, they constitute the basis for many other tasks in mobile robotics, being a important step in knowledge representation and in reasoning about it.

Este trabalho tem o objetivo de divulgar a metodologia de ponderação de modelos ou Bayesian Model Averaging (BMA) entre os pesquisadores da área agronômica e discutir suas vantagens e limitações. Com o BMA é possível combinar resultados de diferentes modelos acerca de determinada quantidade de interesse, com isso, o BMA apresenta-se como sendo uma metodologia alternativa de análise de dados frente os usuais métodos de seleção de modelos tais como o Coeficiente de Determinação Múltipla (R2 ), Coeficiente de Determinação Múltipla Ajustado (R2), Estatística de Mallows ( Cp) e Soma de Quadrados de Predição (PRESS). Vários trabalhos foram, recentemente, realizados com o objetivo de comparar o desempenho do BMA em relação aos métodos de seleção de modelos, porém, há ainda muitas situações para serem exploradas até que se possa chegar a uma conclusão geral acerca desta metodologia. Neste trabalho, o BMA foi aplicado a um conjunto de dados proveniente de um experimento agronômico. A seguir, o desempenho preditivo do BMA foi comparado com o desempenho dos métodos de seleção acima citados por meio de um estudo de simulação variando o grau de multicolinearidade e o tamanho amostral. Em cada uma dessas situações, foram utilizadas 1000 amostras geradas a partir de medidas descritivas de conjuntos de dados reais da área agronômica. O desempenho preditivo das metodologias em comparação foi medido pelo Logaritmo do Escore Preditivo (LEP). Os resultados empíricos obtidos indicaram que o BMA apresenta desempenho semelhante aos métodos usuais de seleção de modelos nas situações de multicolinearidade exploradas neste trabalho. / The objective of this work was divulge to Bayesian Model Averaging (BMA) between the researchers of the agronomy area and discuss its advantages and limitations. With the BMA is possible combine results of difeerent models about determined quantity of interest, with that, the BMA presents as being a metodology alternative of data analysis front the usual models selection approaches, for example the Coefficient of Multiple Determination (R2), Coefficient of Multiple Determination Adjusted (R2), Mallows (Cp Statistics) and Prediction Error Sum Squares (PRESS). Several works recently were carried out with the objective of compare the performance of the BMA regarding the approaches of models selection, however, there is still many situations for will be exploited to that can arrive to a general conclusion about this metodology. In this work, the BMA was applied to data originating from an agronomy experiment. It follow, the predictive performance of the BMA was compared with the performance of the approaches of selection above cited by means of a study of simulation varying the degree of multicollinearity, measured by the number of condition of the matrix standardized X'X and the number of observations in the sample. In each one of those situations, were utilized 1000 samples generated from the descriptive information of agronomy data. The predictive performance of the metodologies in comparison was measured by the Logarithm of the Score Predictive (LEP). The empirical results obtained indicated that the BMA presents similar performance to the usual approaches of selection of models in the situations of multicollinearity exploited.