<p>This thesis investigates the possibility of improving black box fault diagnosis by a process called manifold learning, which simply stated is a way of finding patterns in recorded sensor data. The idea is that there is more information in the data than is exploited when using simple classification algorithms such as k-Nearest Neighbor and Support Vector Machines, and that this additional information can be found by using manifold learning methods. To test the idea of using manifold learning, data from two different fault diagnosis scenarios is used: A Scania truck engine and an electrical system called Adapt. Two linear and one non-linear manifold learning methods are used: Principal Component Analysis and Linear Discriminant Analysis (linear) and Laplacian Eigenmaps (non-linear).Some improvements are achieved given certain conditions on the diagnosis scenarios. The improvements for different methods correspond to the systems in which they are achieved in terms of linearity. The positive results for the relatively linear electrical system are achieved mainly by the linear methods Principal Component Analysis and Linear Discriminant Analysis and the positive results for the non-linear Scania system are achieved by the non-linear method Laplacian Eigenmaps.The results for scenarios without these special conditions are not improved however, and it is uncertain wether the improvements in special condition scenarios are due to gained information or to the nature of the cases themselves.</p>

Nonlinearity in the Navier-Stokes equations can originate from a variety of sources, such as contributions stemming from the advective term, constitutive closure models or external factors such as chemical reactions and capillarity. Needless to say, a combination of any of the above sources has the potential to exasperate the problem significantly. This dissertation explores cases that predominantly feature advective and/or capillary effects. In particular, we first consider the inertia-dominated problem of single-phase flow past a confined square cylinder, followed by a study focused on the low-Re dynamics of rigid particles straddling non-planar interfaces. The first part of the thesis investigates transient, three-dimensional, incompressible and isothermal flow of a Newtonian fluid past a symmetrically confined obstacle at zero incidence. Results from both Laser Doppler Velocimetry (LDV) experiments and direct simulations upto Re = 250 have been reported. Beyond the onset of instability (Recr ≈ 58), an inflexion point around Re ≈ 115 is detected for the Strouhal number with no evidence of hysteresis in any of the measurements. Furthermore, incommensurate frequencies observed in the range 127 ≤ Re ≤ 175 suggest a quasi-periodic transition to three-dimensionality. This is shown to be followed by an intermediate periodic window starting around Re ≈ 180. Fourier analysis and spanwise velocity correlations are then used to characterize the observed phenomena. Subsequent analysis of consolidated data suggest that only a parametric variation of transverse and spanwise blockage ratios can bring closure to the subject of bluff-body wake transitions. The second part of the thesis implements and validates a physically consistent continuum model for the Moving Contact Line (MCL) through direct simulations. After elaborately discussing the MCL conundrum, a fundamental framework for the simulations is outlined in a theoretical orientation which combines the Level set method with a Fictitious domain approach in a finite-element scheme. The thesis objectives are then realized through simulation of various case studies that show favorable comparisons with theoretical and/or published experimental data. In short, the current work successfully illustrates the potential of novel boundary conditions (such as the GNBC) to accurately describe MCL dynamics. / Chemical Engineering

This dissertation investigates the implications of using the Advanced Measurement Approaches (AMA) as a method to assess operational risk capital charges for banks and insurance companies within Basel II paradigms and with regard to U.S. regulations. Operational risk has become recognized as a major risk class because of huge operational losses experienced by many financial firms over the last past decade. Unlike market risk, credit risk, and insurance risk, for which firms and scholars have designed efficient methodologies, there are few tools to help analyze and quantify operational risk. The new Basel Revised Framework for International Convergence of Capital Measurement and Capital Standards (Basel II) gives substantial flexibility to internationally active banks to set up their own risk assessment models in the context of the Advanced Measurement Approaches. The AMA developed in this thesis uses actuarial loss models complemented by the extreme value theory to determine the empirical probability distribution function of the overall capital charge in terms of various classes of copulas. Publicly available operational risk loss data set is used for the empirical exercise.

<p>The <italic>Chlamydiaceae</italic> are Gram-negative, obligate intracellular bacteria that are significant pathogens of humans and animals. Intracellularly, the bacteria reside in a membrane-bound vacuole, called the inclusion, from which they manipulate host processes to create a niche optimal for survival and propagation. Acquisition of host-derived lipids is essential for chlamydial growth, yet the source of lipids and mechanisms of trafficking to the inclusion are not well-established. The inclusion avoids interaction with several classical membrane and lipid transport pathways. In a functional genomic screen to identify host modulating chlamydial proteins, our lab identified cytosolic lipid droplets (LDs) as potential target organelles of <italic>Chlamydia</italic>. LDs are postulated to function in many cellular processes, such as lipid metabolism and transport, membrane trafficking, and cell signaling; therefore, we hypothesized that LDs may be important for <italic>Chlamydia</italic> pathogenesis as a source of lipids or as a platform for regulating other cellular functions. Here, we characterize the interaction between eukaryotic LDs and the chlamydial inclusion.</p><p> We find that LDs are recruited to the <italic>Chlamydia</italic> inclusion, chlamydial infection disrupts neutral lipid homeostasis, and pharmacological prevention of LD formation inhibits chlamydial replication. <italic>Chlamydia</italic> produces proteins (Ldas) that localize with LDs in yeast and mammalian cells when transiently expressed and are exported out of the inclusion to peripheral lipid-rich structures during infection. By electron microscopy and live cell imaging, we observe the translocation of intact LDs into the <italic>Chlamydia</italic> inclusion lumen. Biochemical and microscopic analysis of LDs from infected cells reveals that LD translocation may occur at specialized subregions of the inclusion membrane. The <italic>Chlamydia</italic> Lda3 protein is implicated in LD tethering to the inclusion membrane, and displacement of the protective coat protein, ADRP, from LD surfaces. This phenomenon could provide access for lipases to the LD core for utilization by the replicating bacteria. Additionally, the functional domains of Lda3 involved in binding to LD and inclusion membranes are identified. </p><p> In these studies, we identify eukaryotic lipid droplets (LDs) as a novel target organelle important for <italic>Chlamydia</italic> pathogenesis and describe a unique mechanism of whole organelle sequestration not previously observed for bacterial pathogens. These results represent a fundamental shift in our understanding of host interactions with the chlamydial inclusion, and may represent a new area for research in the field of cellular microbiology.</p> / Dissertation

The small-sample size issue is a prevalent problem in Genomics and Proteomics today. Bootstrap, a resampling method which aims at increasing the efficiency of data usage, is considered to be an effort to overcome the problem of limited sample size. This dissertation studies the application of bootstrap to two problems of supervised learning with small sample data: estimation of the misclassification error of Gaussian discriminant analysis, and the bagging ensemble classification method. Estimating the misclassification error of discriminant analysis is a classical problem in pattern recognition and has many important applications in biomedical research. Bootstrap error estimation has been shown empirically to be one of the best estimation methods in terms of root mean squared error. In the first part of this work, we conduct a detailed analytical study of bootstrap error estimation for the Linear Discriminant Analysis (LDA) classification rule under Gaussian populations. We derive the exact formulas of the first and the second moment of the zero bootstrap and the convex bootstrap estimators, as well as their cross moments with the resubstitution estimator and the true error. Based on these results, we obtain the exact formulas of the bias, the variance, and the root mean squared error of the deviation from the true error of these bootstrap estimators. This includes the moments of the popular .632 bootstrap estimator. Moreover, we obtain the optimal weight for unbiased and minimum-RMS convex bootstrap estimators. In the univariate case, all the expressions involve Gaussian distributions, whereas in the multivariate case, the results are written in terms of bivariate doubly non-central F distributions. In the second part of this work, we conduct an extensive empirical investigation of bagging, which is an application of bootstrap to ensemble classification. We investigate the performance of bagging in the classification of small-sample gene-expression data and protein-abundance mass spectrometry data, as well as the accuracy of small-sample error estimation with this ensemble classification rule. We observed that, under t-test and RELIEF filter-based feature selection, bagging generally does a good job of improving the performance of unstable, overtting classifiers, such as CART decision trees and neural networks, but that improvement was not sufficient to beat the performance of single stable, non-overtting classifiers, such as diagonal and plain linear discriminant analysis, or 3-nearest neighbors. Furthermore, the ensemble method did not improve the performance of these stable classifiers significantly. We give an explicit definition of the out-of-bag estimator that is intended to remove estimator bias, by formulating carefully how the error count is normalized, and investigate the performance of error estimation for bagging of common classification rules, including LDA, 3NN, and CART, applied on both synthetic and real patient data, corresponding to the use of common error estimators such as resubstitution, leave-one-out, cross-validation, basic bootstrap, bootstrap 632, bootstrap 632 plus, bolstering, semi-bolstering, in addition to the out-of-bag estimator. The results from the numerical experiments indicated that the performance of the out-of-bag estimator is very similar to that of leave-one-out; in particular, the out-of-bag estimator is slightly pessimistically biased. The performance of the other estimators is consistent with their performance with the corresponding single classifiers, as reported in other studies. The results of this work are expected to provide helpful guidance to practitioners who are interested in applying the bootstrap in supervised learning applications.

In this study, a 3D face representation and recognition method based on spherical harmonics expansion is proposed. The input data to the method is range image of the face. This data is called 2.5 dimensional. Input faces are manually marked on the two eyes, nose and chin points. In two dimensions, using the marker points, the human face is modeled as two concentric half ellipses for the selection of region of interest. These marker points are also used in three dimensions to register the faces so that the nose point tip is at the origin and the line across the two eyes lies parallel to the horizontal plane. A PCA based component analysis is done to further align the faces vertically. The aligned face is stitched and mapped to an ellipsoid and transformed using real spherical harmonics expansion. The real harmonics expansion coefficients are labeled and stored into a gallery. Using these coefficients as input, several classification algorithms are applied and the results are reported.

In der vorliegenden Arbeit wurden 3D-Strömungszustände und die Bewegung an der Badoberfläche in Abhängigkeit von den Betriebsparametern mit der LDA-Methode im 1:2 Modell einer Stranggießkokille mit Fr-Zahl als Ähnlichkeitskriterium untersucht. Auf Basis der Messdaten wurde der obere Kokillenbereich in 7 Teilräumen stromabwärts unterteilt. Der Verlauf der Freistrahlausbreitung wurde durch eine Exponentialfunktion beschrieben. Nahe am Tauchrohraustritt wurde das Medium sowohl in den austretenden Strahl als auch in das Tauchrohr hinein eingesaugt. Die Frequenz und die Amplitude der Oberflächenschwankungen wurden nach der Leitfähigkeitsmethode gemessen. Dort dominierten drei Frequenzbereiche. In der Strömungsgeschwindigkeit beim Austritt des turbulenten Freistrahles wurden auch entsprechende nieder- und hochfrequente Anteile gemessen. Die Badoberflächenschwankungen wurden an vier Positionen gleichzeitig bestimmt. Mit steigender Fr-Zahl nahm der Mittelwert der Badoberflächenschwankung zu. Zwischen den Fr-Zahlen und den normierten Amplituden der Badoberflächenschwankungen (Mittelwert der Amplitude / hydraulischer Durchmesser der Tauchrohraustrittsfläche), konnte ein linearer Zusammenhang festgestellt werden. Die numerischen Ergebnisse (Fluent), die mit unterschiedlich definierten Randbedingungen des Strahlaustrittes berechnet wurden, ergaben eine gute Übereinstimmung mit den experimentellen Ergebnissen, wenn die Randbedingungen am Tauchrohraustritt auf experimentellen Messdaten basierten. Liegen diese Messdaten nicht vor, dann kann das Ergebnis einer numerischen Untersuchung verbessert werden, indem die stromaufwärts liegenden Systemteile in die Rechensimulation einbezogen werden.

Topic models are a new research field within the computer sciences information retrieval and text mining. They are generative probabilistic models of text corpora inferred by machine learning and they can be used for retrieval and text mining tasks. The most prominent topic model is latent Dirichlet allocation (LDA), which was introduced in 2003 by Blei et al. and has since then sparked off the development of other topic models for domain-specific purposes. This thesis focuses on LDA's practical application. Its main goal is the replication of the data analyses from the 2004 LDA paper ``Finding scientific topics'' by Thomas Griffiths and Mark Steyvers within the framework of the R statistical programming language and the R~package topicmodels by Bettina Grün and Kurt Hornik. The complete process, including extraction of a text corpus from the PNAS journal's website, data preprocessing, transformation into a document-term matrix, model selection, model estimation, as well as presentation of the results, is fully documented and commented. The outcome closely matches the analyses of the original paper, therefore the research by Griffiths/Steyvers can be reproduced. Furthermore, this thesis proves the suitability of the R environment for text mining with LDA. (author's abstract) / Series: Theses / Institute for Statistics and Mathematics

No description available.

O uso de dados biométricos da face para verificação automática de identidade é um dos maiores desafios em sistemas de controle de acesso seguro. O processo é extremamente complexo e influenciado por muitos fatores relacionados à forma, posição, iluminação, rotação, translação, disfarce e oclusão de características faciais. Hoje existem muitas técnicas para se reconhecer uma face. Esse trabalho apresenta uma investigação buscando identificar uma face no banco de dados ORL com diferentes grupos de treinamento. É proposto um algoritmo para o reconhecimento de faces baseado na técnica de subespaço LDA (PCA + LDA) utilizando uma rede neural SOM para representar cada classe (face) na etapa de classificação/identificação. Aplicando o método do subespaço LDA busca-se extrair as características mais importantes na identificação das faces previamente conhecidas e presentes no banco de dados, criando um espaço dimensional menor e discriminante com relação ao espaço original. As redes SOM são responsáveis pela memorização das características de cada classe. O algoritmo oferece maior desempenho (taxas de reconhecimento entre 97% e 98%) com relação às adversidades e fontes de erros que prejudicam os métodos de reconhecimento de faces tradicionais. / The use of biometric technique for automatic personal identification is one of the biggest challenges in the security field. The process is complex because it is influenced by many factors related to the form, position, illumination, rotation, translation, disguise and occlusion of face characteristics. Now a days, there are many face recognition techniques. This work presents a methodology for searching a face in the ORL database with some different training sets. The algorithm for face recognition was based on sub-space LDA (PCA + LDA) technique using a SOM neural net to represent each class (face) in the stage of classification/identification. By applying the sub-space LDA method, we extract the most important characteristics in the identification of previously known faces that belong to the database, creating a reduced and more discriminated dimensional space than the original space. The SOM nets are responsible for the memorization of each class characteristic. The algorithm offers great performance (recognition rates between 97% and 98%) considering the adversities and sources of errors inherent to the traditional methods of face recognition.