No description available.

Ögonkontakt-sensorer skapar möjligheten att tolka användarens uppmärksamhet, vilketkan användas av system på en mängd olika vis. Dessa inkluderar att skapa nya möjligheterför människa-dator-interaktion och mäta mönster i uppmärksamhet hos individer.I den här uppsatsen gör vi ett försök till att konstruera en ögonkontakt-sensor med hjälpav en Raspberry Pi, med målet att göra den praktisk i verkliga scenarion. För att fastställaatt den är praktisk satte vi upp ett antal kriterier baserat på tidigare användning avögonkontakt-sensorer. För att möta dessa kriterier valde vi att använda en maskininlärningsmetodför att träna en klassificerare med bilder för att lära systemet att upptäcka omen användare har ögonkontakt eller ej. Vårt mål var att undersöka hur god prestanda vikunde uppnå gällande precision, hastighet och avstånd. Efter att ha testat kombinationerav fyra olika metoder för feature extraction kunde vi fastslå att den bästa övergripandeprecisionen uppnåddes genom att använda LDA-komprimering på pixeldatan från varjebild, medan PCA-komprimering var bäst när input-bilderna liknande de från träningen.När vi undersökte systemets hastighet fann vi att nedskalning av bilder hade en stor effektpå hastigheten, men detta sänkte också både precision och maximalt avstånd. Vi lyckadesminska den negativa effekten som en minskad skala hos en bild hade på precisionen, mendet maximala avståndet som sensorn fungerade på var fortfarande relativ till skalan och iförlängningen hastigheten. / Eye contact detection sensors have the possibility of inferring user attention, which can beutilized by a system in a multitude of different ways, including supporting human-computerinteraction and measuring human attention patterns. In this thesis we attempt to builda versatile eye contact sensor using a Raspberry Pi that is suited for real world practicalusage. In order to ensure practicality, we constructed a set of criteria for the system basedon previous implementations. To meet these criteria, we opted to use an appearance-basedmachine learning method where we train a classifier with training images in order to inferif users look at the camera or not. Our aim was to investigate how well we could detecteye contacts on the Raspberry Pi in terms of accuracy, speed and range. After extensivetesting on combinations of four different feature extraction methods, we found that LinearDiscriminant Analysis compression of pixel data provided the best overall accuracy, butPrincipal Component Analysis compression performed the best when tested on imagesfrom the same dataset as the training data. When investigating the speed of the system,we found that down-scaling input images had a huge effect on the speed, but also loweredthe accuracy and range. While we managed to mitigate the effects the scale had on theaccuracy, the range of the system is still relative to the scale of input images and byextension speed.

In statistical methodology of analyzing data, two of the most commonly used techniques are classification and regression modeling. As scientific technology progresses rapidly, complex data often occurs and requires novel classification and regression modeling methodologies according to the data structure. In this dissertation, I mainly focus on developing a few approaches for analyzing the data with complex structures. Classification problems commonly occur in many areas such as biomedical, marketing, sociology and image recognition. Among various classification methods, linear classifiers have been widely used because of computational advantages, ease of implementation and interpretation compared with non-linear classifiers. Specifically, linear discriminant analysis (LDA) is one of the most important methods in the family of linear classifiers. For high dimensional data with number of variables p larger than the number of observations n occurs more frequently, it calls for advanced classification techniques. In Chapter 2, I proposed a novel sparse LDA method which generalizes LDA through a regularized approach for the two-class classification problem. The proposed method can obtain an accurate classification accuracy with attractive computation, which is suitable for high dimensional data with p>n. In Chapter 3, I deal with the classification when the data complexity lies in the non-random missing responses in the training data set. Appropriate classification method needs to be developed accordingly. Specifically, I considered the "reject inference problem'' for the application of fraud detection for online business. For online business, to prevent fraud transactions, suspicious transactions are rejected with unknown fraud status, yielding a training data with selective missing response. A two-stage modeling approach using logistic regression is proposed to enhance the efficiency and accuracy of fraud detection. Besides the classification problem, data from designed experiments in scientific areas often have complex structures. Many experiments are conducted with multiple variance sources. To increase the accuracy of the statistical modeling, the model need to be able to accommodate more than one error terms. In Chapter 4, I propose a variance component mixed model for a nano material experiment data to address the between group, within group and within subject variance components into a single model. To adjust possible systematic error introduced during the experiment, adjustment terms can be added. Specifically a group adaptive forward and backward selection (GFoBa) procedure is designed to select the significant adjustment terms. / Ph. D.

Le décodage des expressions faciales émotionnelles (EFE) est une fonction clé du système visuel humain puisqu’il est à la base de la communication non-verbale sur laquelle reposent les interactions sociales. De nombreuses études suggèrent un traitement différentiel des attributs diagnostiques du visage au sein des basses et des hautes fréquences spatiales (FS), respectivement sous-tendu par les voies magno- et parvocellulaires. En outre, des conditions telles que l’anxiété sociale sont susceptibles d’affecter ce traitement et d’entrainer une modulation des potentiels reliés aux évènements (PRE). Cette étude explore la possibilité de prédire le niveau d’anxiété social des individus à partir des corrélats électrophysiologiques du décodage d’EFE dans différentes bandes de FS. À cette fin, les PRE de 26 participants (âge moyen = 23.7 ± 4.7) ont été enregistrés lors de la présentation visuelle d’expressions neutres, de joie ou de colère filtrées pour ne retenir que les basses, moyennes ou hautes FS. L’anxiété sociale a été évaluée par l’administration préalable du questionnaire LSAS. Les latences et pics d’amplitude de la P100, N170, du complexe N2b/P3a et de la P3b ont été analysés statistiquement et utilisés pour entrainer différents algorithmes de classification. L’amplitude de la P100 était reliée au contenu en FS. La N170 a montré un effet des EFE. Le complexe N2b/P3a était plus ample pour les EFE et plus précoce pour les hautes FS. La P3b était moins ample pour les visages neutres, qui étaient aussi plus souvent omis. L’analyse discriminante linéaire a montré une précision de décodage d’en moyenne 56.11% au sein des attributs significatifs. La nature de ces attributs et leur sensibilité à l’anxiété sociale sera discutée. / The decoding of emotional facial expressions (EFE) is a key function of the human visual system since it lays at the basis of non-verbal communication that allows social interactions. Numerous studies suggests that the processing of faces diagnostic features may take place differently for low and high spatial frequencies (SF), respectively in the magno- and parvocellular pathways. Moreover, conditions such as social anxiety are supposed to influence this processing and the associated event-related potentials (ERP). This study explores the feasibility of predicting social anxiety levels using electrophysiological correlates of EFE processing across various SF bands. To this end, ERP from 26 participants (mean age = 23.7 ± 4.7) years old were recorded during visual presentation of neutral, angry and happy facial expressions, filtered to retain only low, medium or high SF. Social anxiety was previously assessed using the LSAS questionnary. Peak latencies and amplitudes of the P100, N170, N2b/P3a complex and P3b components were statistically analyzed and used to feed supervised machine learning algorithms. P100 amplitude was linked to SF content. N170 was effected by EFE. N2b/P3a complex was larger for EFE and earlier for high SF. P3b was lower for neutral faces, which were also more often omitted. The linear discriminant analysis showed a decoding accuracy across significant features with a mean of 56.11%. The nature of these features and their sensitivity to social anxiety will be discussed.

Feature extraction is an extremely important pre-processing step to pattern recognition, and machine learning problems. This thesis highlights how one can best extract features from the data in an exhaustively online and purely adaptive manner. The solution to this problem is given for both labeled and unlabeled datasets, by presenting a number of novel on-line learning approaches. Specifically, the differential equation method for solving the generalized eigenvalue problem is used to derive a number of novel machine learning and feature extraction algorithms. The incremental eigen-solution method is used to derive a novel incremental extension of linear discriminant analysis (LDA). Further the proposed incremental version is combined with extreme learning machine (ELM) in which the ELM is used as a preprocessor before learning. In this first key contribution, the dynamic random expansion characteristic of ELM is combined with the proposed incremental LDA technique, and shown to offer a significant improvement in maximizing the discrimination between points in two different classes, while minimizing the distance within each class, in comparison with other standard state-of-the-art incremental and batch techniques. In the second contribution, the differential equation method for solving the generalized eigenvalue problem is used to derive a novel state-of-the-art purely incremental version of slow feature analysis (SLA) algorithm, termed the generalized eigenvalue based slow feature analysis (GENEIGSFA) technique. Further the time series expansion of echo state network (ESN) and radial basis functions (EBF) are used as a pre-processor before learning. In addition, the higher order derivatives are used as a smoothing constraint in the output signal. Finally, an online extension of the generalized eigenvalue problem, derived from James Stone’s criterion, is tested, evaluated and compared with the standard batch version of the slow feature analysis technique, to demonstrate its comparative effectiveness. In the third contribution, light-weight extensions of the statistical technique known as canonical correlation analysis (CCA) for both twinned and multiple data streams, are derived by using the same existing method of solving the generalized eigenvalue problem. Further the proposed method is enhanced by maximizing the covariance between data streams while simultaneously maximizing the rate of change of variances within each data stream. A recurrent set of connections used by ESN are used as a pre-processor between the inputs and the canonical projections in order to capture shared temporal information in two or more data streams. A solution to the problem of identifying a low dimensional manifold on a high dimensional dataspace is then presented in an incremental and adaptive manner. Finally, an online locally optimized extension of Laplacian Eigenmaps is derived termed the generalized incremental laplacian eigenmaps technique (GENILE). Apart from exploiting the benefit of the incremental nature of the proposed manifold based dimensionality reduction technique, most of the time the projections produced by this method are shown to produce a better classification accuracy in comparison with standard batch versions of these techniques - on both artificial and real datasets.

O sensoriamento remoto é uma estratégia que pode ajudar no monitoramento da qualidade das pastagens. Objetivou-se com esse estudo analisar a resposta espectral das folhas de Brachiaria brizantha cv. Marandú, adubada com doses crescentes de ureia, para diferenciar e predizer teores foliares de nitrogênio (TFN). Os tratamentos foram distribuídos em blocos ao acaso (DBC), composto por quatro blocos e quatro tratamentos, totalizando 16 parcelas. Foram utilizadas doses crescentes de adubação com ureia: 0, 25, 50, 75 kg de N/ha/corte. Ao longo do experimento foram realizadas 7 coletas, sendo coletadas 8 folhas por parcela. Essas folhas foram submetidas à análise hiperespectral e posterior análise química do teor de nitrogênio. Ao analisar a resposta espectral das folhas, observou-se diferenças estatísticas entre os tratamentos na região do visível em todas as coletas, com ênfase na região de 550 nm (verde). Por meio de análise discriminante linear (LDA) realizada para cada coleta, os centróides gerados por todos os tratamentos apresentaram diferenças significativas, com exceção do LD1 nas coletas 6 e 7 que não apresentou distinção entre os tratamentos de 50 e 75 kg de N/ha/corte, e LD2 na coleta 5 que não apresentou distinção entre os tratamentos de 0 e 50 kg de N/ha/corte. As equações de regressão multivariada obtidas pelo método de quadrados mínimos parciais (PLSR), geraram valores razoáveis a bons de R2 (0,53 a 0,83) na predição dos TFN, onde os comprimentos de onda com maior peso nessas regressões estão na região do red edge (715 a 720 nm). Por fim, ao testar a performance de alguns Índices de Vegetação da literatura, as coletas 4, 6 e 7 apresentaram bons coeficientes de determinação (R2) que variaram de 0,65 a 0,73; uma característica em comum nos índices que melhor estimaram os TFN é a presença de comprimentos de ondas que fazem parte da região do red edge. / Remote sensing is a set of techniques that can help to monitor pasture quality. The object of this study is to analyze the spectral response from Brachiaria brizantha cv. Marandú leaves, under contrasting nitrogen levels, to differentiate and predict leaf nitrogen content. The treatments were set in a Randomized Block Design, composed of four blocks and four treatments, totaling 16 plots. Increasing doses of urea fertilization were used: 0, 25, 50, 75 kg N/ha/mowing. During the experiment, 7 data collections were performed, and 8 leaves per plot were extracted for each data collection. These leaves were submitted to hyperspectral data extraction and subsequent chemical analysis to quantify the nitrogen content. When analyzing the spectral pattern of the leaves, statistical differences among samples with different nitrogen levels were noticeable in the visible range of the spectrum in all the collections, with emphasis on the 550 nm region (green). Through linear discriminant analysis (LDA), performed for each collection, the generated centroids by the samples of each nitrogen level presented significant differences, except for LD1 in collections 6 and 7, which did not present a distinction between treatments of 50 and 75 kg of N/ha/mowing, and LD2 in collection 5 that did not distinguish between treatments of 0 and 50 kg of N/ha/mowing. The partial least square regression (PLSR) method generated reasonable to good values of R2 (0.53 to 0.83) for the prediction of leaf nitrogen content, where the wavelengths with the highest coefficient in these models are in the red edge region of the spectrum (715 to 720 nm). Finally, when testing the performance of some Vegetation Indexes from literature, collections 4, 6 and 7 presented good determination coefficients (R2) ranging from 0.65 to 0.73; a common feature in the indexes that best estimate the nitrogen content is the presence of wavelengths from the red edge region of the spectrum.

A qualidade microbiológica de medicamentos é fundamental para garantir sua eficácia e segurança. Os métodos convencionais para identificação microbiana em produtos não estéreis são amplamente utilizados, entretanto são demorados e trabalhosos. O objetivo deste trabalho é desenvolver método microbiológico rápido (MMR) para a identificação de contaminantes em produtos farmacêuticos utilizando a espectrofotometria de infravermelho com transformada de Fourier com reflectância total atenuada (FTIR-ATR). Análise de componentes principais (PCA) e análise de discriminantes (LDA) foram utilizadas para obter um modelo de predição com a capacidade de diferenciar o crescimento de oriundo de contaminação por Bacillus subtilis (ATCC 6633), Candida albicans (ATCC 10231), Enterococcus faecium (ATCC 8459), Escherichia coli (ATCC 8739), Micrococcus luteus (ATCC 10240), Pseudomonas aeruginosa (ATCC 9027), Salmonella Typhimurium (ATCC 14028), Staphylococcus aureus (ATCC 6538) e Staphylococcus epidermidis (ATCC 12228). Os espectros de FTIR-ATR forneceram informações quanto à composição de proteínas, DNA/RNA, lipídeos e carboidratos provenientes do crescimento microbiano. As identificações microbianas fornecidas pelo modelo PCA/LDA baseado no método FTIR-ATR foram compatíveis com aquelas obtidas pelos métodos microbiológicos convencionais. O método de identificação microbiana rápida por FTIR-ATR foi validado quanto à sensibilidade (93,5%), especificidade (83,3%) e limite de detecção (17-23 UFC/mL de amostra). Portanto, o MMR proposto neste trabalho pode ser usado para fornecer uma identificação rápida de contaminantes microbianos em produtos farmacêuticos. / Microbiological quality of pharmaceuticals is fundamental in ensuring efficacy and safety of medicines. Conventional methods for microbial identification in non-sterile drugs are widely used, however are time-consuming and laborious. The aim of this paper was to develop a rapid microbiological method (RMM) for identification of contaminants in pharmaceutical products using Fourier transform infrared with attenuated total reflectance spectrometry (FTIR-ATR). Principal components analysis (PCA) and linear discriminant analysis (LDA) were used to obtain a predictive model with capable to distinguish Bacillus subtilis (ATCC 6633), Candida albicans (ATCC 10231), Enterococcus faecium (ATCC 8459), Escherichia coli (ATCC 8739), Micrococcus luteus (ATCC 10240), Pseudomonas aeruginosa (ATCC 9027), Salmonella Typhimurium (ATCC 14028), Staphylococcus aureus (ATCC 6538), and Staphylococcus epidermidis (ATCC 12228) microbial growth. FTIR-ATR spectra provide information of protein, DNA/RNA, lipids, and carbohydrates constitution of microbial growth. Microbial identification provided by PCA/LDA based on FTIR-ATR method were compatible to those obtained using conventional microbiological methods. FTIR-ATR method for rapid identification of microbial contaminants in pharmaceutical products was validated by assessing the sensitivity (93.5%), specificity (83.3%), and limit of detection (17-23 CFU/mL of sample). Therefore, the RMM proposed in this work may be used to provide a rapid identification of microbial contaminants in pharmaceutical products.

De nos jours, avec l'abondance croissante de données de très grande taille, les problèmes de classification de grande dimension ont été mis en évidence comme un challenge dans la communauté d'apprentissage automatique et ont beaucoup attiré l'attention des chercheurs dans le domaine. Au cours des dernières années, les techniques d'apprentissage avec la parcimonie et l'optimisation stochastique se sont prouvées être efficaces pour ce type de problèmes. Dans cette thèse, nous nous concentrons sur le développement des méthodes d'optimisation pour résoudre certaines classes de problèmes concernant ces deux sujets. Nos méthodes sont basées sur la programmation DC (Difference of Convex functions) et DCA (DC Algorithm) étant reconnues comme des outils puissants d'optimisation non convexe. La thèse est composée de trois parties. La première partie aborde le problème de la sélection des variables. La deuxième partie étudie le problème de la sélection de groupes de variables. La dernière partie de la thèse liée à l'apprentissage stochastique. Dans la première partie, nous commençons par la sélection des variables dans le problème discriminant de Fisher (Chapitre 2) et le problème de scoring optimal (Chapitre 3), qui sont les deux approches différentes pour la classification supervisée dans l'espace de grande dimension, dans lequel le nombre de variables est beaucoup plus grand que le nombre d'observations. Poursuivant cette étude, nous étudions la structure du problème d'estimation de matrice de covariance parcimonieuse et fournissons les quatre algorithmes appropriés basés sur la programmation DC et DCA (Chapitre 4). Deux applications en finance et en classification sont étudiées pour illustrer l'efficacité de nos méthodes. La deuxième partie étudie la L_p,0régularisation pour la sélection de groupes de variables (Chapitre 5). En utilisant une approximation DC de la L_p,0norme, nous prouvons que le problème approché, avec des paramètres appropriés, est équivalent au problème original. Considérant deux reformulations équivalentes du problème approché, nous développons différents algorithmes basés sur la programmation DC et DCA pour les résoudre. Comme applications, nous mettons en pratique nos méthodes pour la sélection de groupes de variables dans les problèmes de scoring optimal et d'estimation de multiples matrices de covariance. Dans la troisième partie de la thèse, nous introduisons un DCA stochastique pour des problèmes d'estimation des paramètres à grande échelle (Chapitre 6) dans lesquelles la fonction objectif est la somme d'une grande famille des fonctions non convexes. Comme une étude de cas, nous proposons un schéma DCA stochastique spécial pour le modèle loglinéaire incorporant des variables latentes / These days with the increasing abundance of data with high dimensionality, high dimensional classification problems have been highlighted as a challenge in machine learning community and have attracted a great deal of attention from researchers in the field. In recent years, sparse and stochastic learning techniques have been proven to be useful for this kind of problem. In this thesis, we focus on developing optimization approaches for solving some classes of optimization problems in these two topics. Our methods are based on DC (Difference of Convex functions) programming and DCA (DC Algorithms) which are wellknown as one of the most powerful tools in optimization. The thesis is composed of three parts. The first part tackles the issue of variable selection. The second part studies the problem of group variable selection. The final part of the thesis concerns the stochastic learning. In the first part, we start with the variable selection in the Fisher's discriminant problem (Chapter 2) and the optimal scoring problem (Chapter 3), which are two different approaches for the supervised classification in the high dimensional setting, in which the number of features is much larger than the number of observations. Continuing this study, we study the structure of the sparse covariance matrix estimation problem and propose four appropriate DCA based algorithms (Chapter 4). Two applications in finance and classification are conducted to illustrate the efficiency of our methods. The second part studies the L_p,0regularization for the group variable selection (Chapter 5). Using a DC approximation of the L_p,0norm, we indicate that the approximate problem is equivalent to the original problem with suitable parameters. Considering two equivalent reformulations of the approximate problem we develop DCA based algorithms to solve them. Regarding applications, we implement the proposed algorithms for group feature selection in optimal scoring problem and estimation problem of multiple covariance matrices. In the third part of the thesis, we introduce a stochastic DCA for large scale parameter estimation problems (Chapter 6) in which the objective function is a large sum of nonconvex components. As an application, we propose a special stochastic DCA for the loglinear model incorporating latent variables

With the advent of computers and the information age, vast amounts of data generated in a great deal of science and industry fields require the statisticians to explore further. In particular, statistical and computational problems in biology and medicine have created a new field of bioinformatics, which is attracting more and more statisticians, computer scientists, and biologists. Several procedures have been developed for tracing the source of fecal pollution in water resources based on certain characteristics of certain microorganisms. Use of this collection of techniques has been termed microbial source tracking (MST). Most of the current methods for MST are based on patterns of either phenotypic or genotypic variation in indicator organisms. Studies also suggested that patterns of genotypic variation might be more reliable due to their less association with environmental factors than those of phenotypic variation. Among the genotypic methods for source tracking, fingerprinting via rep-PCR is most common. Thus, identifying the specific pollution sources in contaminated waters based on rep-PCR fingerprinting techniques, viewed as a classification problem, has become an increasingly popular research topic in bioinformatics. In the project, several statistical methods for classification were studied, including linear discriminant analysis, quadratic discriminant analysis, logistic regression, and $k$-nearest-neighbor rules, neural networks and support vector machine. This project report summaries each of these methods and relevant statistical theory. In addition, an application of these methods to a particular set of MST data is presented and comparisons are made.

Submitted by Rosivalda Pereira (mrs.pereira@ufma.br) on 2017-08-14T18:15:08Z No. of bitstreams: 1 DanielCosta.pdf: 1087754 bytes, checksum: ada5f863f42efd8298fff788c37bded3 (MD5) / Made available in DSpace on 2017-08-14T18:15:08Z (GMT). No. of bitstreams: 1 DanielCosta.pdf: 1087754 bytes, checksum: ada5f863f42efd8298fff788c37bded3 (MD5) Previous issue date: 2008-02-25 / Female breast cancer is the major cause of death in western countries. Efforts in Computer Vision have been made in order to add improve the diagnostic accuracy by radiologists. In this work, we present a methodology that uses independent component analysis (ICA) along with support vector machine (SVM) and linear discriminant analysis (LDA) to distinguish between mass or non-mass and benign or malign tissues from mammograms. As a result, it was found that: LDA reaches 90,11% of accuracy to discriminante between mass or non-mass and 95,38% to discriminate between benign or malignant tissues in DDSM database and in mini-MIAS database we obtained 85% to discriminate between mass or non-mass and 92% of accuracy to discriminate between benign or malignant tissues; SVM reaches 99,55% of accuracy to discriminate between mass or non-mass and the same percentage to discriminate between benign or malignat tissues in DDSM database whereas, and in MIAS database it was obtained 98% to discriminate between mass or non-mass and 100% to discriminate between benign or malignant tissues. / Câncer de mama feminino é o câncer que mais causa morte nos países ocidentais. Esforços em processamento de imagens foram feitos para melhorar a precisão dos diagnósticos por radiologistas. Neste trabalho, nós apresentamos uma metodologia que usa análise de componentes independentes (ICA) junto com análise discriminante linear (LDA) e máquina de vetor de suporte (SVM) para distinguir as imagens entre nódulos ou não-nódulos e os tecidos em benignos ou malignos. Como resultado, obteve-se com LDA 90,11% de acurácia na discriminação entre nódulo ou não-nódulo e 95,38% na discriminação de tecidos benignos ou malignos na base de dados DDSM. Na base de dados mini- MIAS, obteve-se 85% e 92% na discriminação entre nódulos ou não-nódulos e tecidos benignos ou malignos respectivamente. Com SVM, alcançou-se uma taxa de até 99,55% na discriminação de nódulos ou não-nódulos e a mesma porcentagem na discriminação entre tecidos benignos ou malignos na base de dados DDSM enquanto que na base de dados mini-MIAS, obteve-se 98% e até 100% na discriminação de nódulos ou não-nódulos e tecidos benignos ou malignos, respectivamente.