Thesis (Ph. D.)--University of Washington, 2006. / Vita. Includes bibliographical references (p. 103-108).

We introduce and explore an approach to estimating statisticalsignificance of classification accuracy, which is particularly usefulin scientific applications of machine learning where highdimensionality of the data and the small number of training examplesrender most standard convergence bounds too loose to yield ameaningful guarantee of the generalization ability of theclassifier. Instead, we estimate statistical significance of theobserved classification accuracy, or the likelihood of observing suchaccuracy by chance due to spurious correlations of thehigh-dimensional data patterns with the class labels in the giventraining set. We adopt permutation testing, a non-parametric techniquepreviously developed in classical statistics for hypothesis testing inthe generative setting (i.e., comparing two probabilitydistributions). We demonstrate the method on real examples fromneuroimaging studies and DNA microarray analysis and suggest atheoretical analysis of the procedure that relates the asymptoticbehavior of the test to the existing convergence bounds.

This paper bridges the gap between variable selection methods (e.g., Pearson coefficients, KS test) and dimensionality reductionalgorithms (e.g., PCA, LDA). Variable selection algorithms encounter difficulties dealing with highly correlated data,since many features are similar in quality. Dimensionality reduction algorithms tend to combine all variables and cannotselect a subset of significant variables.Our approach combines both methodologies by applying variable selection followed by dimensionality reduction. Thiscombination makes sense only when using the same utility function in both stages, which we do. The resulting algorithmbenefits from complex features as variable selection algorithms do, and at the same time enjoys the benefits of dimensionalityreduction.1

Thesis (Ph.D.)--Duquesne University, 2005. / Title from document title page. Abstract included in electronic submission form. Includes bibliographical references (p. 74-80) and index.

Orientador: Sheila Zambello de Pinho / Banca: Martha Maria Mischan / Banca: Décio Barbin / Banca: Antonio Francisco Iemma / Banca: David Ariovaldo Banzatto / Resumo: As funções de crescimento logística e de Gompertz têm sido bastante estudadas e freqüentemente utilizadas na área biológica. Diversos pesquisadores têm ajustado as funções logística ou de Gompertz a dados provenientes de experimentos com vários tratamentos onde curvas são ajustadas e o interesse é saber se há diferença entre estes tratamentos. A verificação da adequacidade de ajustes das funções não-lineares e a comparação de diferentes funções para um determinado conjunto de dados estão bem contempladas na literatura. Porém, quando o mesmo tipo de função é ajustado a várias situações (tratamentos) e o interesse é fazer a comparação das mesmas, há dificuldades de se encontrar subsídios na literatura. O objetivo deste trabalho foi a apresentação de um método de comparação de curvas logísticas e de Gompertz. Compararam-se as equações ajustadas através de testes dos parâmetros, utilizando-se métodos paramétricos e nãoparamétricos. Determinaram-se também, valores da variável independente x a partir dos quais a diferença entre a assíntota e a curva ajustada deixa de ser significativa. Estudaram-se nesta pesquisa o modelo logístico com erro aditivo na ausência e na presença de autocorrelação nos resíduos, o modelo logístico com erro multiplicativo na ausência e na presença de autocorrelação nos resíduos, o modelo de Gompertz com erro aditivo na ausência e na presença de autocorrelação nos resíduos e o modelo de Gompertz com erro multiplicativo na ausência e na presença de autocorrelação nos resíduos. Para ilustração da metodologia utilizaram-se dados de peso de matéria fresca (g) de sementes de feijão Phaseolus vulgaris L. cv. carioca 80 SH, porcentagens médias do peso de frutos de araribá, pesos de frangos de corte de aves Indian River e pesos de ratos Rattus norvergicus, aos quais ajustaram-se,...(Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The logistic and the Gompertz growth functions have been considerably studied and frequently used in biological area. Several researchers have fitted the Logistic and the Gompertz to data from experiments with many treatments where the purpose is to detect the difference among them. The verification of the adequacy of the non-linear fits and the comparison of different functions for a set of data are well studied in the literature. However, when the same function is fitted to several situations (treatments) and the purpose is to compare them, there are difficulty to find subsidy in the literature. The purpose of this work was to determine a method of comparison of the Logistic and the Gompertz curves and to verify until when the difference between the curves and their superior asymptotes are significant. In this research were studied the logistic and the Gompertz models considering additive and multiplicative error terms with and without autocorrelation. For enlightenment of the methodology were used data of fresh matter of Phaseolus vulgaris L. cv carioca 80 SH seeds, percentage of araribá fruit weight, weight of chicken for slaughter Indian River and weight of rats Rattus norvergicus, where were fitted the Logistic model with additive errors terms and without autocorrelation, the Logistic model with additive errors terms and with autocorrelation, the Gompertz model with additive errors terms and without autocorrelation, the Gompertz model with additive errors terms and with autocorrelation...(Complete abstract, click electronic access below) / Doutor

No description available.

No description available.

Systems composed of dynamical networks - such as the human body with its biological networks or the global economic network consisting of regional clusters - often exhibit complicated collective dynamics. Three fundamental processes that are typically present are failure, damage spread, and recovery. Here we develop a model for such systems and find phase diagrams for single and interacting networks. By investigating networks with a small number of nodes, where finite-size effects are pronounced, we describe the spontaneous recovery phenomenon present in these systems. In the case of interacting networks the phase diagram is very rich and becomes increasingly more complex as the number of interacting networks increases. In the simplest example of two interacting networks we find two critical points, four triple points, ten allowed transitions, and two forbidden transitions, as well as complex hysteresis loops. Remarkably, we find that triple points play the dominant role in constructing the optimal repairing strategy in damaged interacting systems. To test our model, we analyze an example of real interacting financial networks and find evidence of rapid dynamical transitions between well-defined states, in agreement with the predictions of our model.

Spatial structure in a population creates distinctive patterns in genetic data. There are two reasons to model this process. First, since the genetic structure of a population is induced by its historical spatial structure, it can be used to make inference about history and demography. Second, these models provide corrections to other analyses that are confounded by spatial structure. Since is it is now common to collect genome-wide data on many thousands of samples, a major challenge is to develop fast, scalable, approximate algorithms that can analyse these datasets. A practical approach is to focus on subsets of the data that are most informative, for example rare variants. First we look at the problem of estimating selection coefficients in spatially structured populations. We demonstrate this approach using classical datasets of moth colour morph frequencies, and then use it in a model incorporating both ancient and modern DNA to estimate the selective advantage of one of the best known examples of local adaptation in humans, lactase persistence in Europeans. Next, we turn to the problem of association studies in spatially structured populations. We demonstrate that rare variants are more confounded by non-genetic risk than common variants. Excess confounding is a consequence of the fact that rare variants are highly in- formative about recent ancestry and therefore, in a spatially explicit model, about location. Finally, we use this insight into rare variants to develop methods for inference about population history using rare variant and haplotype sharing as simple summary statistics. These approaches are extremely fast and can be applied to genome-wide data on thousands of samples, yet they provide an accurate description of the history of a population, both identifying recent ancestry and estimating migration rates between subpopulations.

We investigate mechanisms to improve efficiency for line and point transect surveys of clustered populations by combining the distance methods with adaptive sampling. In adaptive sampling, survey effort is increased when areas of high animal density are located, thereby increasing the number of observations. We begin by building on existing adaptive sampling techniques, to create both point and line transect adaptive estimators, these are then extended to allow the inclusion of covariates in the detection function estimator. However, the methods are limited, as the total effort required cannot be forecast at the start of a survey, and so a new fixed total effort adaptive approach is developed. A key difference in the new method is that it does not require the calculation of the inclusion probabilities typically used by existing adaptive estimators. The fixed effort method is primarily aimed at line transect sampling, but point transect derivations are also provided. We evaluate the new methodology by computer simulation, and report on surveys of harbour porpoise in the Gulf of Maine, in which the approach was compared with conventional line transect sampling. Line transect simulation results for a clustered population showed up to a 6% improvement in the adaptive density variance estimate over the conventional, whilst when there was no clustering the adaptive estimate was 1% less efficient than the conventional. For the harbour porpoise survey, the adaptive density estimate cvs showed improvements of 8% for individual porpoise density and 14% for school density over the conventional estimates. The primary benefit of the fixed effort method is the potential to improve survey coverage, allowing a survey to complete within a fixed time and effort; an important feature if expensive survey resources are involved, such as an aircraft, crew and observers.