Dissertations / Theses on the topic 'Linear classifier'

Currently technologies such as EEG, EMG and EOG recordings are the established methods used in the analysis of sleep. But if these methods are to be employed to study sleep in rodents, extensive surgery and recovery is involved which can be both time consuming and costly. This thesis presents and analyzes a cost effective, non-invasive, high throughput system for detecting the sleep and wake patterns in mice using a piezoelectric sensor. This sensor was placed at the bottom of the mice cages to monitor the movements of the mice. The thesis work included the development of the instrumentation and signal acquisition system for recording the signals critical to sleep and wake classification. Classification of the mouse sleep and wake states were studied for a linear classifier and a Neural Network classifier based on 23 features extracted from the Power Spectrum (PS), Generalized Spectrum (GS), and Autocorrelation (AC) functions of short data intervals. The testing of the classifiers was done on two data sets collected from two mice, with each data set having around 5 hours of data. A scoring of the sleep and wake states was also done via human observation to aid in the training of the classifiers. The performances of these two classifiers were analyzed by looking at the misclassification error of a set of test features when run through a classifier trained by a set of training features. The best performing features were selected by first testing each of the 23 features individually in a linear classifier and ranking them according to their misclassification rate. A test was then done on the 10 best individually performing features where they were grouped in all possible combinations of 5 features to determine the feature combinations leading to the lowest error rates in a multi feature classifier. From this test 5 features were eventually chosen to do the classification. It was found that the features related to the signal energy and the spectral peaks in the 3Hz range gave the lowest errors. Error rates as low as 4% and 9% were achieved from a 5-feature linear classifier for the two data sets. The error rates from a 5-feature Neural Network classifier were found to be 6% and 12% respectively for these two data sets.

Intensive Care Units (ICUs) host patients in critical condition who are being monitored by sensors which measure their vital signs. These vital signs carry information about a patient’s physiology and can have a very rich structure at fine resolution levels. The task of analysing these biosignals for the purposes of monitoring a patient’s physiology is referred to as physiological condition monitoring. Physiological condition monitoring of patients in ICUs is of critical importance as their health is subject to a number of events of interest. For the purposes of this thesis, the overall task of physiological condition monitoring is decomposed into the sub-tasks of modelling a patient’s physiology a) under the effect of physiological or artifactual events and b) under the effect of drug administration. The first sub-task is concerned with modelling artifact (such as the taking of blood samples, suction events etc.), and physiological episodes (such as bradycardia), while the second sub-task is focussed on modelling the effect of drug administration on a patient’s physiology. The first contribution of this thesis is the formulation, development and validation of the Discriminative Switching Linear Dynamical System (DSLDS) for the first sub-task. The DSLDS is a discriminative model which identifies the state-of-health of a patient given their observed vital signs using a discriminative probabilistic classifier, and then infers their underlying physiological values conditioned on this status. It is demonstrated on two real-world datasets that the DSLDS is able to outperform an alternative, generative approach in most cases of interest, and that an a-mixture of the two models achieves higher performance than either of the two models separately. The second contribution of this thesis is the formulation, development and validation of the Input-Output Non-Linear Dynamical System (IO-NLDS) for the second sub-task. The IO-NLDS is a non-linear dynamical system for modelling the effect of drug infusions on the vital signs of patients. More specifically, in this thesis the focus is on modelling the effect of the widely used anaesthetic drug Propofol on a patient’s monitored depth of anaesthesia and haemodynamics. A comparison of the IO-NLDS with a model derived from the Pharmacokinetics/Pharmacodynamics (PK/PD) literature on a real-world dataset shows that significant improvements in predictive performance can be provided without requiring the incorporation of expert physiological knowledge.

In some classification problems where human judgments, qualitative and imprecise data exist, uncertainty comes from fuzziness rather than randomness. Limited number of fuzzy classification approaches is available for use for these classification problems to capture the effect of fuzzy uncertainty imbedded in data. The scope of this study mainly comprises two parts: new fuzzy classification approaches based on Tanaka&rsquo
s Fuzzy Linear Regression (FLR) approach, and an improvement of an existing one, Improved Fuzzy Classifier Functions (IFCF). Tanaka&rsquo
s FLR approach is a well known fuzzy regression technique used for the prediction problems including fuzzy type of uncertainty. In the first part of the study, three alternative approaches are presented, which utilize the FLR approach for a particular customer satisfaction classification problem. A comparison of their performances and their applicability in other cases are discussed. In the second part of the study, the improved IFCF method, Nonparametric Improved Fuzzy Classifier Functions (NIFCF), is presented, which proposes to use a nonparametric method, Multivariate Adaptive Regression Splines (MARS), in clustering phase of the IFCF method. NIFCF method is applied on three data sets, and compared with Fuzzy Classifier Function (FCF) and Logistic Regression (LR) methods.

Neste trabalho, foram utilizadas as vantagens da ferramenta matemática de análise temporal e espectral, a transformada wavelet discreta (DWT), além dos coeficientes de predição linear (LPC) e do algoritmo de inteligência artificial, Least Squares Support Vector Machines (LS-SVM), para aplicações em análise de sinais de voz e classificação de vozes patológicas. Inúmeros trabalhos na literatura têm demonstrado o grande interesse existente por ferramentas auxiliares ao diagnóstico de patologias da laringe. Os componentes da DWT forneceram parâmetros de medida para a análise e classificação das vozes patológicas, principalmente aquelas provenientes de pacientes com edema de Reinke e nódulo nas pregas vocais. O banco de dados com as vozes patológicas foi obtido do Departamento de Otorrinolaringologia e Cirurgia de Cabeça e Pescoço do Hospital das Clínicas da Faculdade de Medicina de Ribeirão Preto (FMRP-USP). Utilizando-se o algoritmo de reconhecimento de padrões, LS-SVM, mostrou-se que a combinação dos componentes da DWT de Daubechies com o filtro LP inverso levou a um classificador de bom desempenho alcançando mais de 90% de acerto na classificação das vozes patológicas.
The main objective of this work was to use the advantages of the time-frequency analysis mathematical tool, discrete wavelet transform (DWT), besides the linear prediction coefficients (LPC) and the artificial intelligence algorithm, Least Squares Support Vector Machines (LS-SVM), for applications in voice signal analysis and classification of pathological voices. A large number of works in the literature has been shown that there is a great interest for auxiliary tools to the diagnosis of laryngeal pathologies. DWT components gave measure parameters for the analysis and classification of pathological voices, mainly that ones from patients with Reinke\'s edema and nodule in the vocal folds. It was used a data bank with pathological voices from the Otolaryngology and the Head and Neck Surgery sector of the Clinical Hospital of the Faculty of Medicine at Ribeirão Preto, University of Sao Paulo (FMRP-USP), Brazil. Using the automatic learning algorithm applied in pattern recognition problems, LS-SVM, results have showed that the combination of Daubechies\' DWT components and inverse LP filter leads to a classifier with good performance reaching more than 90% of accuracy in the classification of the pathological voices.

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.

Face recognition is a popular research area where there are different approaches studied in the literature. In this thesis, a holistic Principal Component Analysis (PCA) based method, namely Eigenface method is studied in detail and three of the methods based on the Eigenface method are compared. These are the Bayesian PCA where Bayesian classifier is applied after dimension reduction with PCA, the Subspace Linear Discriminant Analysis (LDA) where LDA is applied after PCA and Eigenface where Nearest Mean Classifier applied after PCA. All the three methods are implemented on the Olivetti Research Laboratory (ORL) face database, the Face Recognition Technology (FERET) database and the CNN-TURK Speakers face database. The results are compared with respect to the effects of changes in illumination, pose and aging. Simulation results show that Subspace LDA and Bayesian PCA perform slightly well with respect to PCA under changes in pose
however, even Subspace LDA and Bayesian PCA do not perform well under changes in illumination and aging although they perform better than PCA.

Made available in DSpace on 2016-06-02T19:05:36Z (GMT). No. of bitstreams: 1 2598.pdf: 3349204 bytes, checksum: 01649491fd1f03aa5a11b9191727f88b (MD5) Previous issue date: 2009-06-29
Financiadora de Estudos e Projetos
This master thesis describes an intelligent classifier system applied to multiclass non-linearly separable problems called Slicer. The system adopts a low computacional cost supervised learning strategy (evaluated as ) based on linear separability. During the learning period the system determines a set of hyperplanes associated to oneclass regions (sub-spaces). In classification tasks the classifier system uses the hyperplanes as a set of if-then-else rules to infer the class of the input attribute vector (non classified object). Among other characteristics, the intelligent classifier system is able to: deal with missing attribute values examples; reject noise examples during learning; adjust hyperplane parameters to improve the definition of the one-class regions; and eliminate redundant rules. The fuzzy theory is considered to design a hybrid version with features such as approximate reasoning and parallel inference computation. Different classification methods and benchmarks are considered for evaluation. The classifier system Slicer reaches acceptable results in terms of accuracy, justifying future investigation effort.
Este trabalho de mestrado descreve um sistema classificador inteligente aplicado a problemas multiclasse não-linearmente separáveis chamado Slicer. O sistema adota uma estratégia de aprendizado supervisionado de baixo custo computacional (avaliado em ) baseado em separabilidade linear. Durante o período de aprendizagem o sistema determina um conjunto de hiperplanos associados a regiões de classe única (subespaços). Nas tarefas de classificação o sistema classificador usa os hiperplanos como um conjunto de regras se-entao-senao para inferir a classe do vetor de atributos dado como entrada (objeto a ser classificado). Entre outras caracteristicas, o sistema classificador é capaz de: tratar atributos faltantes; eliminar ruídos durante o aprendizado; ajustar os parâmetros dos hiperplanos para obter melhores regiões de classe única; e eliminar regras redundantes. A teoria nebulosa é considerada para desenvolver uma versão híbrida com características como raciocínio aproximado e simultaneidade no mecanismo de inferência. Diferentes métodos de classificação e domínios são considerados para avaliação. O sistema classificador Slicer alcança resultados aceitáveis em termos de acurácia, justificando investir em futuras investigações.

This diploma thesis deals with a problem of spores venturia inaequlis recognition. These spores are captured on a special tape which is then analyzed using a microscope. The tape can be analyzed by a laboratorian or by the program Sporedetect v3. This program provides functions for complete picture processing and object recognition. In this diploma thesis, there are also described ways to automatically control a sliding stage of a microscope utilizing motorized translation stages and linear actuators. The information about automatic control of a microscope stage was obtained from catalogues of the companies Standa and Edmundoptics.

Includes bibliography.
This thesis describes methods of constructing models for cross-classified categorical data. In particular we discuss the construction of a class of approximating models and the selection of the most suitable model in the class. Examples of application are used to illustrate the methodology. The main purpose of the thesis is to demonstrate that it is both possible and advantageous to construct models which are specifically designed for the particular application under investigation. We believe that the methods described here allow the statistician to make good use of any expert knowledge which the client (typically a non-statistician) might possess on the subject to which the data relate.

This thesis describes the design and development of a system for detection and recognition of license plates. The work is divided into three basic parts: licence plates detection, finding of character positions and optical character recognition. To fullfill the goal of this work, a new dataset was taken. It contains 2814 license plates used for training classifiers and 2620 plates to evaluate the success rate of the system. Cascade Classifier was used to train detector of licence plates, which has success rate up to 97.8 %. After that, pozitions of individual characters were searched in detected pozitions of licence plates. If there was no character found, detected pozition was not the licence plate. Success rate of licence plates detection with all the characters found is up to 88.5 %. Character recognition is performed by SVM classifier. The system detects successfully with no errors up to 97.7 % of all licence plates.

Thesis (Electrical Engineer and M.S. in Electrical Engineering)--Naval Postgraduate School, March 2010.
Thesis Advisor(s): Fargues, Monique P. ; Cristi, Roberto ; Robertson, Ralph C. "March 2010." Description based on title screen as viewed on .April 27, 2010. Author(s) subject terms: Blind Modulation Classification, Cumulants, Principal Component Analysis, Linear Discriminant Analysis, Kernel-based functions. Includes bibliographical references (p. 211-212). Also available in print.

This thesis applies a hierarchical latent trait model system to a large quantity of data. The motivation for it was lack of viable approaches to analyse High Throughput Screening datasets which maybe include thousands of data points with high dimensions. We believe that a latent variable model (LTM) with a non-linear mapping from the latent space to the data space is a preferred choice for visualising a complex high-dimensional data set. As a type of latent variable model, the latent trait model can deal with either continuous data or discrete data, which makes it particularly useful in this domain. In addition, with the aid of differential geometry, we can imagine that distribution of data from magnification factor and curvature plots. Rather than obtaining the useful information just from a single plot, a hierarchical LTM arranges a set of LTMs and their corresponding plots in a tree structure. We model the whole data set with a LTM at the top level, which is broken down into clusters at deeper levels of the hierarchy. In this manner, the refined visualisation plots can be displayed in deeper levels and sub-clusters may be found. Hierarchy of LTMs is trained using expectation-maximisation (EM) algorithm to maximise its likelihood with respect to the data sample. Training proceeds interactively in a recursive fashion (top-down). The user subjectively identifies interesting regions on the visualisation plot that they would like to model in a greater detail. At each stage of hierarchical LTM construction, the EM algorithm alternates between the E - and M - step. Another problem that can occur when visualising a large data set is that there may be significant overlaps of data clusters. It is very difficult for the user to judge where centres of regions of interest should be put. We address this problem by employing the minimum message length technique, which can help the user to decide the optimal structure of the model.

La classification des images est aujourd'hui un défi d'une grande ampleur puisque ça concerne d'un côté les millions voir des milliards d'images qui se trouvent partout sur le web et d'autre part des images pour des applications temps réel critiques. Cette classification fait appel en général à des méthodes d'apprentissage et à des classifieurs qui doivent répondre à la fois à la précision ainsi qu'à la rapidité. Ces problèmes d'apprentissage touchent aujourd'hui un grand nombre de domaines d'applications: à savoir, le web (profiling, ciblage, réseaux sociaux, moteurs de recherche), les "Big Data" et bien évidemment la vision par ordinateur tel que la reconnaissance d'objets et la classification des images. La présente thèse se situe dans cette dernière catégorie et présente des algorithmes d'apprentissage supervisé basés sur la minimisation de fonctions de perte (erreur) dites "calibrées" pour deux types de classifieurs: k-Plus Proches voisins (kNN) et classifieurs linéaires. Ces méthodes d'apprentissage ont été testées sur de grandes bases d'images et appliquées par la suite à des images biomédicales. Ainsi, cette thèse reformule dans une première étape un algorithme de Boosting des kNN et présente ensuite une deuxième méthode d'apprentissage de ces classifieurs NN mais avec une approche de descente de Newton pour une convergence plus rapide. Dans une seconde partie, cette thèse introduit un nouvel algorithme d'apprentissage par descente stochastique de Newton pour les classifieurs linéaires connus pour leur simplicité et leur rapidité de calcul. Enfin, ces trois méthodes ont été utilisées dans une application médicale qui concerne la classification de cellules en biologie et en pathologie.

Object recognition in video has seen giant strides in accuracy improvements in the last few years, a testament to the computational capacity of deep convolutional neural networks. However, this computational capacity of software-based neural networks coincides with high power consumption compared to that of some spiking neural networks (SNNs), up to 300,000 times more energy per synaptic event in IBM's TrueNorth chip, for example. SNNs are also well-suited to exploit the precise timing of event-driven image sensors, which transmit asynchronous "events" only when the luminance of a pixel changes above or below a threshold value. The combination of event-based imagers and SNNs becomes a straightforward way to achieve low power consumption in object recognition tasks. This thesis compares different linear classifiers for two low-power, hardware-friendly, spiking, unsupervised neural network architectures, SSLCA and HFirst, in response to asynchronous event-based data, and explores their ability to learn and recognize patterns from two event-based image datasets, N-MNIST and CIFAR10-DVS. By performing a grid search of important SNN and classifier hyperparameters, we also explore how to improve classification performance of these architectures. Results show that a softmax regression classifier exhibits modest accuracy gains (0.73%) over the next-best performing linear support vector machine (SVM), and considerably outperforms a single layer perceptron (by 5.28%) when classification performance is averaged over all datasets and spiking neural network architectures with varied hyperparameters. Min-max normalization of the inputs to the linear classifiers aides in classification accuracy, except in the case of the single layer perceptron classifier. We also see the highest reported classification accuracy for spiking convolutional networks on N-MNIST and CIFAR10-DVS, increasing this accuracy from 97.77% to 97.82%, and 29.67% to 31.76%, respectively. These findings are relevant for any system employing unsupervised SNNs to extract redundant features from event-driven data for recognition.

The topic of this thesis is an audio recording classification with 15 different acoustic scene classes that represent common scenes and places where people are situated on a regular basis. The thesis describes 2 approaches based on GMM and i-vectors and a fusion of the both approaches. The score of the best GMM system which was evaluated on the evaluation dataset of the DCASE Challenge is 60.4%. The best i-vector system's score is 68.4%. The fusion of the GMM system and the best i-vector system achieves score of 69.3%, which would lead to the 20th place in the all systems ranking of the DCASE 2017 Challenge (among 98 submitted systems from all over the world).

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2017.
Cataloged from PDF version of thesis.
Includes bibliographical references (pages 203-221).
Scoring systems are linear classification models that let users make quick predictions by adding, subtracting, and multiplying a few small numbers. These models are widely used in applications where humans have traditionally made decisions because they are easy to understand and validate. In spite of extensive deployment, many scoring systems are still built using ad hoc approaches that combine statistical techniques, heuristics, and expert judgement. Such approaches impose steep trade-offs with performance, making it difficult for practitioners to build scoring systems that will be used and accepted. In this dissertation, we present two new machine learning methods to learn scoring systems from data: Supersparse Linear Integer Models (SLIM) for decision-making applications; and Risk-calibrated Supersparse Linear Integer Models (RiskSLIM) for risk assessment applications. Both SLIM and RiskSLIM solve discrete optimization problems to learn scoring systems that are fully optimized for feature selection, small integer coefficients, and operational constraints. We formulate these problems as integer programming problems and develop specialized algorithms to recover certifiably optimal solutions with an integer programming solver. We illustrate the benefits of this approach by building scoring systems for realworld problems such as recidivism prediction, sleep apnea screening, ICU seizure prediction, and adult ADHD diagnosis. Our results show that a discrete optimization approach can learn simple models that perform well in comparison to the state-of-the-art, but that are far easier to customize, understand, and validate.
by Berk Ustun.
Ph. D.

Submitted by Thiago Oliveira (thiago.oliveira@ifpb.edu.br) on 2017-03-23T17:51:30Z No. of bitstreams: 1 31- Maria de F?tima de Kallynna Bezerra Couras - CLASSIFICA??O DE DESVIOS VOCAIS UTILIZANDO CARACTER?STICAS BASEADAS NO MODELO LINEAR DE PRODU??O DA FALA.pdf: 4019202 bytes, checksum: 06d0b1b292ae0bd918d5b8358ca11b0f (MD5)
Approved for entry into archive by Thiago Oliveira (thiago.oliveira@ifpb.edu.br) on 2017-03-23T17:53:11Z (GMT) No. of bitstreams: 1 31- Maria de F?tima de Kallynna Bezerra Couras - CLASSIFICA??O DE DESVIOS VOCAIS UTILIZANDO CARACTER?STICAS BASEADAS NO MODELO LINEAR DE PRODU??O DA FALA.pdf: 4019202 bytes, checksum: 06d0b1b292ae0bd918d5b8358ca11b0f (MD5)
Made available in DSpace on 2017-03-23T17:53:11Z (GMT). No. of bitstreams: 1 31- Maria de F?tima de Kallynna Bezerra Couras - CLASSIFICA??O DE DESVIOS VOCAIS UTILIZANDO CARACTER?STICAS BASEADAS NO MODELO LINEAR DE PRODU??O DA FALA.pdf: 4019202 bytes, checksum: 06d0b1b292ae0bd918d5b8358ca11b0f (MD5) Previous issue date: 2017-02
A avalia??o perceptivo-auditiva tem papel fundamental na avalia??o da qualidade vocal. No entanto, por ser uma avalia??o subjetiva, est? sujeita a imprecis?es e varia??es, sendo necess?ria a utiliza??o de t?cnicas que tragam maior confiabilidade aos resultados. A an?lise ac?stica surge como uma ferramenta que proporciona a avalia??o da qualidade vocal de forma objetiva. Neste trabalho, s?o empregadas t?cnicas de processamento digital de sinais, baseadas no modelo linear de produ??o da fala, para analisar a qualidade vocal. ? avaliado o desempenho de medidas tradicionalmente empregadas na an?lise ac?stica, tais como frequ?ncia fundamental, medidas de perturba??o (jitter e shimmer), GNE (Glottal to Noise Excitation Ratio) e frequ?ncias form?nticas. Tambem ? avaliado o potencial discriminativo dos coeficientes da an?lise de predi??o linear (Linear Predictive Coding- LPC), coeficientes cepstrais e mel-cepstrais na classifica??o de desvios vocais (rugosidade, soprosidade e tens?o). Com o aux?lio de um classificador, baseado em redes neurais artificiais MLP (Multilayer Perceptron), ? realizada a classifica??o dos sinais utilizando as medidas extra?das individualmente e de forma combinada. Foram obtidas taxas de classifica??o de 86% na discrimina??o entre vozes soprosas e vozes saud?veis.

Conventional investigations into the performance of piston-rings in internal combustion engines are performed at relatively low speeds and consider only steady state operation conditions. Loss of power in internal combustion (IC) engines is becoming an increasing issue when they are operated at high engine speeds. This project is directed at developing technology to establish whether this phenomenon is influenced by a lubricant related effect. In a normal operating environment automotive engines typically operate under transient operating conditions. These rapid changes in operation conditions may influence the thickness of the hydrodynamic film which lubricates the interfaces between the piston-ring and liner. During this project two capacitance methods were employed in a fired compression ignition engine, an amplitude modulated (AM) system originally developed by Grice and a new "high speed" capacitance technique based on a frequency modulated principle. The first part of this thesis is concerned with the development and implementation of a new apparatus suitable for measuring the thickness and extent of the hydrodynamic oil film which lubricates the piston-rings and liner. The nature of the working principle of the high speed capacitance measurement system required the design, manufacture, assembly and commissioning of a novel dynamic calibration apparatus. The new system can also be used for static calibration (AM system) of capacitance based distance measuring systems. It uses a manufacturer calibrated closed loop controlled piezo-actuator to present a target relative to the sensor face. Some previous investigations concluded a stable oil film thickness. However, this work shows that there are cyclic variations of the oil film thickness OFT on a stroke to stroke and cycle to cycle basis. A series of measurements was conducted under various fixed speed load points. The effects of using lubricants of different viscosity on the minimum (OFT) between liner and piston ring have been little studied and this work shows that it was possible to speciate measurements of different lubricants. This thesis also describes a measurement of the oil film thickness during abrupt changes in engine operating conditions.

This thesis examines the application of a Support Vector Machine (SVM) classifier to automatically detect sleep and quiet wake (rest) behavior in mice from pressure signals on their cage floor. Previous work employed Neural Networks (NN) and Linear Discriminant Analysis (LDA) to successfully detect sleep and wake behaviors in mice. Although the LDA was successful in distinguishing between the sleep and wake behaviors, it has several limitations, which include the need to select a threshold and difficulty separating additional behaviors with subtle differences, such as sleep and rest. The SVM has advantages in that it offers greater degrees of freedom than the LDA for working with complex data sets. In addition, the SVM has direct methods to limit overfitting for the training sets (unlike the NN method). This thesis develops an SVM classifier to characterize the linearly non separable sleep and rest behaviors using a variety of features extracted from the power spectrum, autocorrelation function, and generalized spectrum (autocorrelation of complex spectrum). A genetic algorithm (GA) optimizes the SVM parameters and determines a combination of 5 best features. Experimental results from over 9 hours of data scored by human observation indicate 75% classification accuracy for SVM compared to 68% accuracy for LDA.

Submitted by Programa de P?s-Gradua??o Engenharia El?trica (ppgee@ifpb.edu.br) on 2018-03-27T13:16:45Z No. of bitstreams: 1 36- Giulliana Karla Lacerda Pereira de Queiroz - AN?LISE DIN?MICA N?O LINEAR E AN?LISE DE QUANTIFICA??O DE RECORR?NCIA APLICADAS NA CLASSIFICA??O DE DESVIOS VOCAIS.pdf: 2447855 bytes, checksum: 4d6f1915d45e04e18c97b0d760f5a8ae (MD5)
Approved for entry into archive by Alex Sandro R?go (alex@ifpb.edu.br) on 2018-04-05T19:22:26Z (GMT) No. of bitstreams: 1 36- Giulliana Karla Lacerda Pereira de Queiroz - AN?LISE DIN?MICA N?O LINEAR E AN?LISE DE QUANTIFICA??O DE RECORR?NCIA APLICADAS NA CLASSIFICA??O DE DESVIOS VOCAIS.pdf: 2447855 bytes, checksum: 4d6f1915d45e04e18c97b0d760f5a8ae (MD5)
Made available in DSpace on 2018-04-05T19:22:26Z (GMT). No. of bitstreams: 1 36- Giulliana Karla Lacerda Pereira de Queiroz - AN?LISE DIN?MICA N?O LINEAR E AN?LISE DE QUANTIFICA??O DE RECORR?NCIA APLICADAS NA CLASSIFICA??O DE DESVIOS VOCAIS.pdf: 2447855 bytes, checksum: 4d6f1915d45e04e18c97b0d760f5a8ae (MD5) Previous issue date: 2018-03-27
PRPIPG - IFPB
Disfonia representa qualquer dificuldade na emiss?o vocal que prejudique a produ??o natural da voz. T?cnicas de processamento digital de sinais v?m sendo empregadas como ferramenta auxiliar na avalia??o de desvios vocais, trazendo maior conforto ao paciente. Algumas medidas n?o lineares, baseadas na teoria do caos, foram empregadas,neste trabalho, em conjunto com medidas de quantifica??o de recorr?ncia para a an?lise discriminativa destes desvios. Dois estudos de caso foram realizados nesta pesquisa. No caso 1 foi feita a discrimina??o de vozes adultas saud?veis e desviadas (rugosidade, soprosidade e tens?o) e no caso 2 foi avaliada a discrimina??o da intensidade dos graus dos desvios vocais de vozes adultas (Grau 1-voz normal, Grau 2 - voz considerada com desvio leve e Grau 3 - voz considerada com desvio moderado). As caracter?sticas da an?lise din?mica n?o linear empregada no processo de classifica??o foram a Dimens?o de Correla??o e o Primeiro M?nimo da Fun??o de Informa??o M?tua. As medidas de quantifica??o empregadas foram o Determinismo, a Entropia de Shannon, o Comprimento M?dio das Linhas Diagonais, o Comprimento M?ximo das Linhas Verticais e a Transitividade. O Passo de Reconstru??o tamb?m foi utilizado no processo de classifica??o. Por meio dos testes estat?sticos, foi avaliado o potencial de cada caracter?stica em discriminar os tipos de sinais de voz considerados. Foi utilizada a rede neural MLP (Multilayer Perceptron), com o algoritmo de aprendizado supervisionado Gradiente Conjugado Escalonado (SCG), no processo de classifica??o. Avaliando o desempenho do classificador utilizando as medidas, de forma individual e combinada, foram obtidos, como melhores resultados, uma acur?cia m?dia de 91,17% na distin??o entre as vozes saud?veis e soprosas com as medidas Transitividade e Passo de reconstru??o. Com rela??o ? discrimina??o entre a intensidade dos graus dos desvios, obteve-se uma acur?cia m?dia de 94,5% entre os Graus 1 e 3, com a combina??o das medidas Determinismo, Entropia, Transitividade, Primeiro M?nimo da Fun??o de Informa??o M?tua e o Comprimento m?ximo das linhas verticais. Os resultados encontrados, nesta pesquisa, indicam que as medidas n?o lineares, baseadas na teoria do caos, com as medidas de quantifica??o de recorr?ncia foram eficientes para detectar a presen?a e o grau dos desvios vocais, podendo ser empregada em m?todos de avalia??o, triagem e monitoramento vocal.

Esse trabalho está baseado na investigação dos detectores de falhas aplicando classificadores de classe única. As falhas a serem detectadas são relativas ao estado de funcionamento de cada componente do circuito, especificamente de suas tolerâncias (falha paramétrica). Usando a função de transferência de cada um dos circuitos são gerados e analisados os sinais de saída com os componentes dentro e fora da tolerância. Uma função degrau é aplicada à entrada do circuito, o sinal de saída desse circuito passa por uma função diferenciadora e um filtro. O sinal de saída do filtro passa por um processo de redução de atributos e finalmente, o sinal segue simultaneamente para os classificadores multiclasse e classe única. Na análise são empregados ferramentas de reconhecimento de padrões e de classificação de classe única. Os classficadores multiclasse são capazes de classificar o sinal de saída do circuito em uma das classes de falha para o qual foram treinados. Eles apresentam um bom desempenho quando as classes de falha não possuem superposição e quando eles não são apresentados a classes de falhas para os quais não foram treinados. Comitê de classificadores de classe única podem classificar o sinal de saída em uma ou mais classes de falha e também podem classificá-lo em nenhuma classe. Eles apresentam desempenho comparável ao classificador multiclasse, mas também são capazes detectar casos de sobreposição de classes de falhas e indicar situações de falhas para os quais não foram treinados (falhas desconhecidas). Os resultados obtidos nesse trabalho mostraram que os classificadores de classe única, além de ser compatível com o desempenho do classificador multiclasse quando não há sobreposição, também detectou todas as sobreposições existentes sugerindo as possíveis falhas.
This work deals with the application of one class classifiers in fault detection. The faults to be detected are related parametric faults. The transfer function of each circuit was generated and the outputs signals with the components in and out of tolerance were analyzed. Pattern recognition and one class classifications tools are employed to perform the analysis. The multiclass classifiers are able to classify the circuit output signal in one of the trained classes. They present a good performance when the fault classes do not overlap or when they are not presented to fault classes that were not presented in the training. The one class classifier committee may classify the output signal in one or more fault classes and may also classify them in none of the trained class faults. They present comparable performance to multiclass classifiers, but also are able to detect overlapping fault classes and show fault situations that were no present in the training (unknown faults).

The classical search engine result page is used for many interactions with search results. While these are effective at communicating relevance, they do not present the context well. By giving the user an overview in the form of a spatialized display, in a domain that has a physical analog that the user is familiar with, context should become pre-attentive and obvious to the user. A prototype has been built that takes public medical information articles and assigns these to parts of the human body. The articles are indexed and made searchable. A visualization presents the coverage of a query on the human body and allows the user to interact with it to explore the results. Through usage cases the function and utility of the approach is shown.

Supervised learning of piecewise linear models is a well studied problem in machine learning community. The key idea in piecewise linear modeling is to properly partition the input space and learn a linear model for every partition. Decision trees and regression trees are classic examples of piecewise linear models for classification and regression problems. The existing approaches for learning decision/regression trees can be broadly classified in to two classes, namely, fixed structure approaches and greedy approaches. In the fixed structure approaches, tree structure is fixed before hand by fixing the number of non leaf nodes, height of the tree and paths from root node to every leaf node of the tree. Mixture of experts and hierarchical mixture of experts are examples of fixed structure approaches for learning piecewise linear models. Parameters of the models are found using, e.g., maximum likelihood estimation, for which expectation maximization(EM) algorithm can be used. Fixed structure piecewise linear models can also be learnt using risk minimization under an appropriate loss function. Learning an optimal decision tree using fixed structure approach is a hard problem. Constructing an optimal binary decision tree is known to be NP Complete. On the other hand, greedy approaches do not assume any parametric form or any fixed structure for the decision tree classifier. Most of the greedy approaches learn tree structured piecewise linear models in a top down fashion. These are built by binary or multi-way recursive partitioning of the input space. The main issues in top down decision tree induction is to choose an appropriate objective function to rate the split rules. The objective function should be easy to optimize. Top-down decision trees are easy to implement and understand, but there are no optimality guarantees due to their greedy nature. Regression trees are built in the similar way as decision trees. In regression trees, every leaf node is associated with a linear regression function. All piece wise linear modeling techniques deal with two main tasks, namely, partitioning of the input space and learning a linear model for every partition. However, Partitioning of the input space and learning linear models for different partitions are not independent problems. Simultaneous optimal estimation of partitions and learning linear models for every partition, is a combinatorial problem and hence computationally hard. However, piecewise linear models provide better insights in to the classification or regression problem by giving explicit representation of the structure in the data. The information captured by piecewise linear models can be summarized in terms of simple rules, so that, they can be used to analyze the properties of the domain from which the data originates. These properties make piecewise linear models, like decision trees and regression trees, extremely useful in many data mining applications and place them among top data mining algorithms. In this thesis, we address the problem of supervised learning of piecewise linear models for classification and regression. We propose novel algorithms for learning piecewise linear classifiers and regression functions. We also address the problem of noise tolerant learning of classifiers in presence of label noise. We propose a novel algorithm for learning polyhedral classifiers which are the simplest form of piecewise linear classifiers. Polyhedral classifiers are useful when points of positive class fall inside a convex region and all the negative class points are distributed outside the convex region. Then the region of positive class can be well approximated by a simple polyhedral set. The key challenge in optimally learning a fixed structure polyhedral classifier is to identify sub problems, where each sub problem is a linear classification problem. This is a hard problem and identifying polyhedral separability is known to be NP complete. The goal of any polyhedral learning algorithm is to efficiently handle underlying combinatorial problem while achieving good classification accuracy. Existing methods for learning a fixed structure polyhedral classifier are based on solving non convex constrained optimization problems. These approaches do not efficiently handle the combinatorial aspect of the problem and are computationally expensive. We propose a method of model based estimation of posterior class probability to learn polyhedral classifiers. We solve an unconstrained optimization problem using a simple two step algorithm (similar to EM algorithm) to find the model parameters. To the best of our knowledge, this is the first attempt to form an unconstrained optimization problem for learning polyhedral classifiers. We then modify our algorithm to find the number of required hyperplanes also automatically. We experimentally show that our approach is better than the existing polyhedral learning algorithms in terms of training time, performance and the complexity. Most often, class conditional densities are multimodal. In such cases, each class region may be represented as a union of polyhedral regions and hence a single polyhedral classifier is not sufficient. To handle such situation, a generic decision tree is required. Learning optimal fixed structure decision tree is a computationally hard problem. On the other hand, top-down decision trees have no optimality guarantees due to the greedy nature. However, top-down decision tree approaches are widely used as they are versatile and easy to implement. Most of the existing top-down decision tree algorithms (CART,OC1,C4.5, etc.) use impurity measures to assess the goodness of hyper planes at each node of the tree. These measures do not properly capture the geometric structures in the data. We propose a novel decision tree algorithm that ,at each node, selects hyperplanes based on an objective function which takes into consideration geometric structure of the class regions. The resulting optimization problem turns out to be a generalized eigen value problem and hence is efficiently solved. We show through empirical studies that our approach leads to smaller size trees and better performance compared to other top-down decision tree approaches. We also provide some theoretical justification for the proposed method of learning decision trees. Piecewise linear regression is similar to the corresponding classification problem. For example, in regression trees, each leaf node is associated with a linear regression model. Thus the problem is once again that of (simultaneous) estimation of optimal partitions and learning a linear model for each partition. Regression trees, hinge hyperplane method, mixture of experts are some of the approaches to learn continuous piecewise linear regression models. Many of these algorithms are computationally intensive. We present a method of learning piecewise linear regression model which is computationally simple and is capable of learning discontinuous functions as well. The method is based on the idea of K plane regression that can identify a set of linear models given the training data. K plane regression is a simple algorithm motivated by the philosophy of k means clustering. However this simple algorithm has several problems. It does not give a model function so that we can predict the target value for any given input. Also, it is very sensitive to noise. We propose a modified K plane regression algorithm which can learn continuous as well as discontinuous functions. The proposed algorithm still retains the spirit of k means algorithm and after every iteration it improves the objective function. The proposed method learns a proper Piece wise linear model that can be used for prediction. The algorithm is also more robust to additive noise than K plane regression. While learning classifiers, one normally assumes that the class labels in the training data set are noise free. However, in many applications like Spam filtering, text classification etc., the training data can be mislabeled due to subjective errors. In such cases, the standard learning algorithms (SVM, Adaboost, decision trees etc.) start over fitting on the noisy points and lead to poor test accuracy. Thus analyzing the vulnerabilities of classifiers to label noise has recently attracted growing interest from the machine learning community. The existing noise tolerant learning approaches first try to identify the noisy points and then learn classifier on remaining points. In this thesis, we address the issue of developing learning algorithms which are inherently noise tolerant. An algorithm is inherently noise tolerant if, the classifier it learns with noisy samples would have the same performance on test data as that learnt from noise free samples. Algorithms having such robustness (under suitable assumption on the noise) are attractive for learning with noisy samples. Here, we consider non uniform label noise which is a generic noise model. In non uniform label noise, the probability of the class label for an example being incorrect, is a function of the feature vector of the example.(We assume that this probability is less than 0.5 for all feature vectors.) This can account for most cases of noisy data sets. There is no provably optimal algorithm for learning noise tolerant classifiers in presence of non uniform label noise. We propose a novel characterization of noise tolerance of an algorithm. We analyze noise tolerance properties of risk minimization frame work as risk minimization is a common strategy for classifier learning. We show that risk minimization under 01 loss has the best noise tolerance properties. None of the other convex loss functions have such noise tolerance properties. Empirical risk minimization under 01 loss is a hard problem as 01 loss function is not differentiable. We propose a gradient free stochastic optimization technique to minimize risk under 01 loss function for noise tolerant learning of linear classifiers. We show (under some conditions) that the algorithm converges asymptotically to the global minima of the risk under 01 loss function. We illustrate the noise tolerance of our algorithm through simulations experiments. We demonstrate the noise tolerance of the algorithm through simulations.

Dissertation submitted in partial fulfillment of the requirements for the Degree of Master of Science in Geospatial Technologies.
Land Cover and Land Use (LCLU) maps are very important tools for understanding the relationships between human activities and the natural environment. Defining accurately all the features over the Earth's surface is essential to assure their management properly. The basic data which are being used to derive those maps are remote sensing imagery (RSI), and concretely, satellite images. Hence, new techniques and methods able to deal with those data and at the same time, do it accurately, have been demanded. In this work, our goal was to have a brief review over some of the currently approaches in the scientific community to face this challenge, to get higher accuracy in LCLU maps. Although, we will be focus on the study of the classifiers ensembles and the different strategies that those ensembles present in the literature. We have proposed different ensembles strategies based in our data and previous work, in order to increase the accuracy of previous LCLU maps made by using the same data and single classifiers. Finally, only one of the ensembles proposed have got significantly higher accuracy, in the classification of LCLU map, than the better single classifier performance with the same data. Also, it was proved that diversity did not play an important role in the success of this ensemble.

碩士
臺灣大學
資訊工程學研究所
98
The classification problem appears in many applications such as document classification and web page search. Support vector machine(SVM) is one of the most popular tools used in classification task. One of the component in SVM is the kernel trick. We use kernels to map data into a higher dimentional space. And this technique is applied in non-linear SVMs. For large-scale sparce data, we use the linear kernel to deal with it. We call such SVM as the linear SVM. There are many kinds of SVMs in which different loss functions are applied. We call these SVMs as L1-SVM and L2-SVM in which L1-loss and L2-loss functions are used respectively. We can also apply SVMs to deal with multi-class classification with one-against-one or one-against-all approaches. In this thesis several models such as logistic regression, L1-SVM, L2-SVM, Crammer and Singer, and maximum entropy will be compared in the multi-class classification task.

The quantitative analysis of decomposed electromyographic (EMG) signals reveals information for diagnosing and characterizing neuromuscular disorders. Neuromuscular jitter is an important measure that reflects the stability of the operation of a neuromuscular junction. It is conventionally measured using single fiber electromyographic (SFEMG) techniques. SFEMG techniques require substantial physician dexterity and subject cooperation. Furthermore, SFEMG needles are expensive, and their re-use increases the risk of possible transmission of infectious agents. Using disposable concentric needle (CN) electrodes and automating the measurment of neuromuscular jitter would greatly facilitate the study of neuromuscular disorders. An improved automated jitter measurment system based on the decomposition of CN detected EMG signals is developed and evaluated in this thesis. Neuromuscular jitter is defined as the variability of time intervals between two muscle fiber potentials (MFPs). Given the candidate motor unit potentials (MUPs) of a decomposed EMG signal, which is represented by a motor unit potential train (MUPT), the automated jitter measurement system designed in this thesis can be summarized as a three-step procedure: 1) identify isolated motor unit potentials in a MUPT, 2) detect the significant MFPs of each isolated MUP, 3) track significant MFPs generated by the same muscle fiber across all isolated MUPs, select typical MFP pairs, and calculate jitter. In Step one, a minimal spanning tree-based 2-phase clustering algorithm was developed for identifying isolated MUPs in a train. For the second step, a pattern recognition system was designed to classify detected MFP peaks. At last, the neuromuscular jitter is calculated based on the tracked and selected MFP pairs in the third step. These three steps were simulated and evaluated using synthetic EMG signals independently, and the whole system is preliminary implemented and evaluated using a small simulated data base. Compared to previous work in this area, the algorithms in this thesis showed better performance and great robustness across a variety of EMG signals, so that they can be applied widely to similar scenarios. The whole system developed in this thesis can be implemented in a large EMG signal decomposition system and validated using real data.

碩士
中信金融管理學院
金融管理研究所
107
Predicting the movement of the stock price has been always a popular research direction for securities market. Several methods has been developed during the past, such as the Fundamental analysis, using the information of financial statement, or Technical analysis, which is using the Stock price of Trading market. However setting parameters is important for Technical analysis, Patel, Shan, Thakkar, Kotecha (2015) used data of two stock price indices (CNX Nifty, S&P BSE Sensex) and two stocks (Reliance Industries, Infosys Ltd.) from 2003 to 2012, through Random forest, Naive-Bayes classifier and other algorithm to test. Result confirms converting technical indicators from Continuous-representation(Normalization) to Discrete-representation (Trend Classification) is better Continuous-representation. Therefore, we used Artificial Neural Network and Support Vector Machine as the model, Relative Strength Index, Stochastic Oscillator, Moving Average Convergence/ Divergence and On Balance Volume as the Feature(Variable), consider financial status, it’s different between listed companies(stock exchange market and over-the-counter market), we adopt data from TAIEX and TPEx, total of 30 years and 20 years to run the test(each batch as five years). We evaluated models through Indicators (Accuracy, Recall, Precision, F1 Score, AUC), and our result shows winning ratio of Trend classification is 56.67% higher than Normalization during Artificial Neural Network- TAIEX, winning ratio of Trend classification is 50.0% equal to Normalization during Support Vector Machine - TAIEX, winning ratio of Trend classification is 85.00% higher than Normalization during Artificial Neural Network- TPEx, winning ratio of Trend classification is 67.71% higher than Normalization during Support Vector Machine – TPEx.

碩士
國立臺灣科技大學
資訊工程系
92
Biometric recognition systems perform personal identification with physiological characteristics. These physiological characteristics usually include the following: faces, irises, retinas, hand textures, and fingerprints. Irises are not easy to be copied and do not change forever. Moreover, everyone has different irises. According to such cues, irises have high quality of uniqueness and stability, and they are great for biometric recognition. In this thesis, we present a human iris recognition system with a high recognition rate. The iris recognition system consists of three major processing phases. First, the system captures images of human’s eyes from a web camera, and obtains iris images from them. We further manipulate the iris images using digital image processing techniques, so that the resulting iris images are suited to recognition. Second, the system makes feature vectors from the iris images. Before extraction of feature vectors, we must unwrap the iris images. In this phase, the problem of rotation invariant is solved. We then adopt direct linear discriminant analysis to extract feature vectors such that the distance between the feature vectors of different classes is the largest but the distance between those in the same class is the smallest. Finally, the system employs the nearest feature classifiers to discriminate the feature vectors. To verify the effectiveness of the proposed methods, we realize a human iris recognition system. The experimental results reveal that the recognition rate achieves 96.47% in the case of fewer sampling feature vectors, whereas it can attain 98.50% if more sampling feature vectors are added to each class.

L'application de classifieurs linéaires à l'analyse des données d'imagerie cérébrale (fMRI) a mené à plusieurs percées intéressantes au cours des dernières années. Ces classifieurs combinent linéairement les réponses des voxels pour détecter et catégoriser différents états du cerveau. Ils sont plus agnostics que les méthodes d'analyses conventionnelles qui traitent systématiquement les patterns faibles et distribués comme du bruit. Dans le présent projet, nous utilisons ces classifieurs pour valider une hypothèse portant sur l'encodage des sons dans le cerveau humain. Plus précisément, nous cherchons à localiser des neurones, dans le cortex auditif primaire, qui détecteraient les modulations spectrales et temporelles présentes dans les sons. Nous utilisons les enregistrements fMRI de sujets soumis à 49 modulations spectro-temporelles différentes. L'analyse fMRI au moyen de classifieurs linéaires n'est pas standard, jusqu'à maintenant, dans ce domaine. De plus, à long terme, nous avons aussi pour objectif le développement de nouveaux algorithmes d'apprentissage automatique spécialisés pour les données fMRI. Pour ces raisons, une bonne partie des expériences vise surtout à étudier le comportement des classifieurs. Nous nous intéressons principalement à 3 classifieurs linéaires standards, soient l'algorithme machine à vecteurs de support (linéaire), l'algorithme régression logistique (régularisée) et le modèle bayésien gaussien naïf (variances partagées).
The application of linear machine learning classifiers to the analysis of brain imaging data (fMRI) has led to several interesting breakthroughs in recent years. These classifiers combine the responses of the voxels to detect and categorize different brain states. They allow a more agnostic analysis than conventional fMRI analysis that systematically treats weak and distributed patterns as unwanted noise. In this project, we use such classifiers to validate an hypothesis concerning the encoding of sounds in the human brain. More precisely, we attempt to locate neurons tuned to spectral and temporal modulations in sound. We use fMRI recordings of brain responses of subjects listening to 49 different spectro-temporal modulations. The analysis of fMRI data through linear classifiers is not yet a standard procedure in this field. Thus, an important objective of this project, in the long term, is the development of new machine learning algorithms specialized for neuroimaging data. For these reasons, an important part of the experiments is dedicated to studying the behaviour of the classifiers. We are mainly interested in 3 standard linear classifiers, namely the support vectors machine algorithm (linear), the logistic regression algorithm (regularized) and the naïve bayesian gaussian model (shared variances).

碩士
國立交通大學
光電工程研究所
107
In order to satisfy the Internet data traffic and explosive growth of cloud computing and Internet of Things (IoT). The latest IEEE 802.3bs 400G Ethernet standard has been announced to support 2-km and 10-km transmission. However, it is obvious that the requirements for inter-data center applications or other beyond 10-km applications are insufficient. Therefore, the IEEE 802 LMSC Executive Committee has chartered a Study Group under the IEEE 802.3 Ethernet Working Group to develop the 200-Gb/s and 400-Gb/s Ethernet standard for beyond 10-km optical PHYs. Digital signal processing (DSP) is considered for expanding available bandwidth and capacity and cost-effective solution in the next generation network. Nowadays, artificial neural networks have already been used for optical transmission systems. Organizations in this field are developing and researching the artificial neural network based non-linear equalizers to compensated distorted signals caused by the non-linear effect. In addition, in order to apply in DSP ICs, it is necessary to reduce model complexity and computational complexity since the power consumption is critical. In this work, we establish an 80-Gb/s PAM-4 1293-nm EML-based optical link over 40-km transmission and successfully use artificial neural network based non-linear classifiers (ANN-based NLCs) to compensate distorted signals. Furthermore, we adopt the pruning method to reduce model complexity and use 8-bit quantization to decrease computational complexity.

碩士
國立屏東商業技術學院
不動產經營系(所)
99
Houses are not only influenced by their attributes, but also the location. A good location represents better quality of life, and enhances the added value of the houses. It is an important issue to indicate the quality of the districts by objective regional characteristic variables. This study divides the districts into administrative districts and school districts. Neighborhood affluent index is treated as the characteristic variable of administrative districts, and star schools and first-choice admission rate in junior high schools are treated as the characteristic variables of school districts for evaluating the quality of districts. According to different districts, hierarchical linear modeling and hierarchical cross-classified linear modeling are adopted to estimate housing prices. Empirical result demonstrates that housing prices must be estimated by hierarchical linear model or hierarchical cross-classified linear model. In these two models, the housing prices are not only affected by the housing attributes, but also the regional characteristic variables (Neighborhood affluent index and promotion rate). Housing attributes are also mediated by neighborhood affluent index and promotion rate.

When starch granules are gelatinized, many complex structural changes occur as a result of large quantity of water being absorbed. The enlargement of granule sizes and the leaching out water-soluble macromolecules contribute to the viscoelasticity. Starch pasting behavior greatly influences the texture of a variety of food products such as canned soup, sauces, baby foods, batter mixes etc. It is important to characterize the relationship between the structure, composition and architecture of the starch granules with its pasting behavior in order to arrive at a rational methodology to design modified starch of desirable digestion rate and texture. Five types of starch used in this study were waxy maize starch (WMS), normal maize starch (NMS), waxy rice starch (WRS), normal rice starch (NRS) and STMP cross linked normal maize starch. Evolution of volume fraction φ and pasting of 8% w/w starch suspension when heated at 60, 65, 70, 75, 80, 85 and 90 °C were characterized by particle size distribution and G’, G” in the frequency range of 0.01 to 10 Hz respectively. As expected, granule swelling was more pronounced at higher temperatures. At a fixed temperature, most of the swelling occurred within the first 5 min of heating. The pastes exhibited elastic behavior with G’ being much greater than G”. G’ increased with time for waxy maize and rice starch at all times. G’ and G’’ were found to correlated only to the temperature of pasting and not change much with the rate of heating. For WMS, WRS and STMP crosslinked NMS, G’ approached a limiting value for long heating times (30 min and above) especially at heating temperatures of 85°C and above. This behavior is believed to be due to the predominant effect of swelling at small times. For normal maize and rice starch, however, G’ reached a maximum and decreased at longer times for temperatures above 80 °C due to softening of granules as evidenced by peak force measurements. For each starch sample, the experimental data of G’ at different heating temperatures and times could be collapsed into a single curve. The limiting value of G’ at high volume fraction was related to granule size and granule interfacial energy using a foam rheology model. The interfacial free energy of granules were obtained from contact angle measurements and was employed to evaluate the limiting G’. The experimental data of G’ for all starches when subjected to different heating temperatures and times were normalized with respect to the limiting value at high volume fractions. The master curve for normalized G’ was employed to predict the evolution of G’ with time for different starches which was found to agree well with experimental data of storage modulus. A mechanistic model for starch swelling that is based on Flory Huggins polymer swelling theory was employed to predict the evolution of volume fraction of swollen granules. The model accounts for the structure and composition of different types of starches through starch-solvent interaction as quantified by static light scattering, gelatinization temperature and enthalpy of gelatinization, porosity and its variation with swelling and crosslinking of starch molecules within the granule from equilibrium swelling. Consequently, one could predict the evolution of texture of these starch suspension from the knowledge of their swelling behavior. Expressing the limiting storage modulus of complete swelling (volume fraction approaching unity) of starch suspension in terms of foam rheology, we were able to normalize the storage modulus of different types of starches with respect to its limiting value which is found to fall into a master curve. This master curve when employed along with the swelling model resulted in the successful prediction of development of texture for different types of starches. The above methodology can quantify the effects of structure and composition of starch on its pasting behavior and would therefore provide a rational guideline for modification and processing of starch-based material to obtain desirable texture and rheological properties.

This work explores how randomization can be exploited to deliver sophisticated

algorithms with provable bounds for: (i) The approximation of matrix functions, such

as the log-determinant and the Von-Neumann entropy; and (ii) The low-rank approximation

of matrices. Our algorithms are inspired by recent advances in Randomized

Numerical Linear Algebra (RandNLA), an interdisciplinary research area that exploits

randomization as a computational resource to develop improved algorithms for

large-scale linear algebra problems. The main goal of this work is to encourage the

practical use of RandNLA approaches to solve Big Data bottlenecks at industrial

level. Our extensive evaluation tests are complemented by a thorough theoretical

analysis that proves the accuracy of the proposed algorithms and highlights their

scalability as the volume of data increases. Finally, the low computational time and

memory consumption, combined with simple implementation schemes that can easily

be extended in parallel and distributed environments, render our algorithms suitable

for use in the development of highly efficient real-world software.

Static terrestrial laser scanners have been increasingly used in three-dimensional data acquisition since it can rapidly provide accurate measurements with high resolution. Several scans from multiple viewpoints are necessary to achieve complete coverage of the surveyed objects due to occlusion and large object size. Therefore, in order to reconstruct three-dimensional models of the objects, the task of registration is required to transform several individual scans into a common reference frame. This thesis introduces three alternative approaches for the coarse registration of two adjacent scans, namely, feature-based approach, pseudo-conjugate point-based method, and closed-form solution. In the feature-based approach, linear and planar features in the overlapping area of adjacent scans are selected as registration primitives. The pseudo-conjugate point-based method utilizes non-corresponding points along common linear and planar features to estimate transformation parameters. The pseudo-conjugate point-based method is simpler than the feature-based approach since the partial derivatives are easier to compute. In the closed-form solution, a rotation matrix is first estimated by using a unit quaternion, which is a concise description of the rotation. Afterward, the translation parameters are estimated with non-corresponding points along the linear or planar features by using the pseudo-conjugate point-based method. Alternative approaches for fitting a line or plane to data with errors in three-dimensional space are investigated.

Experiments are conducted using simulated and real datasets to verify the effectiveness of the introduced registration procedures and feature fitting approaches. The proposed two approaches of line fitting are tested with simulated datasets. The results suggest that these two approaches can produce identical line parameters and variance-covariance matrix. The three registration approaches are tested with both simulated and real datasets. In the simulated datasets, all three registration approaches produced equivalent transformation parameters using linear or planar features. The comparison between the simulated linear and planar features shows that both features can produce equivalent registration results. In the real datasets, the three registration approaches using the linear or planar features also produced equivalent results. In addition, the results using real data indicates that the registration approaches using planar features produced better results than the approaches using linear features. The experiments show that the pseudo-conjugate point-based approach is easier to implement than the feature-based approach. The pseudo-conjugate point-based method and feature-based approach are nonlinear, so an initial guess of transformation parameters is required in these two approaches. Compared to the nonlinear approaches, the closed-form solution is linear and hence it can achieve the registration of two adjacent scans without the requirement of any initial guess for transformation parameters. Therefore, the pseudo-conjugate point-based method and closed-form solution are the preferred approaches for coarse registration using linear or planar features. In real practice, the planar features would have a better preference when compared to linear features since the linear features are derived indirectly by the intersection of neighboring planar features. To get enough lines with different orientations, planes that are far apart from each other have to be extrapolated to derive lines.

Emotional intelligence has a prominent role in education, health care, and day to day interaction. With the increasing use of computer technology, computers are interacting with more and more individuals. This interaction provides an opportunity to increase knowledge about human emotion for human consumption, well-being, and improved computer adaptation. This thesis explores the efficacy of using up to four different sensors in three domains for computational affect detection. We first consider computer-based education, where a collection of four sensors is used to detect student emotions relevant to learning, such as frustration, confidence, excitement and interest while students use a computer geometry tutor. The best classier of each emotion in terms of accuracy ranges from 78% to 87.5%. We then use voice data collected in a clinical setting to differentiate both gender and culture of the speaker. We produce classifiers with accuracies between 84% and 94% for gender, and between 58% and 70% for American vs. Asian culture, and we find that classifiers for distinguishing between four cultures do not perform better than chance. Finally, we use video and audio in a health care education scenario to detect students' emotions during a clinical simulation evaluation. The video data provides classifiers with accuracies between 63% and 88% for the emotions of confident, anxious, frustrated, excited, and interested. We find the audio data to be too complex to single out the voice source of the student by automatic means. In total, this work is a step forward in the automatic computational detection of affect in realistic settings.

Individuals with disabilities and persons operating in inaccessible environments can greatly benefit from the aid of robotic manipulators in performing activities of daily living (ADLs) and other remote tasks. Users relying on robotic manipulators to interact with their environment are restricted by the lack of sensory information available through traditional operator interfaces. These interfaces only allow visual task access and deprive users of somatosensory feedback that would be available through direct contact. Multimodal sensory feedback can bridge these perceptual gaps effectively. Given a set of object properties (e.g. temperature, weight) to be conveyed and sensory modalities (e.g. visual, haptic) available, it is necessary to determine which modality should be assigned to each property for an effective interface design. However, the effectiveness of assigning properties to modalities has varied with application and context. The goal of this study was to develop an effective multisensory interface for robot-assisted pouring tasks, which delivers nuanced sensory feedback while permitting high visual demand necessary for precise teleoperation. To that end, an optimization approach is employed to generate a combination of feedback properties to modality assignments that maximizes effective feedback perception and minimizes cognitive load. A set of screening experiments tested twelve possible individual assignments to form the combination. Resulting perceptual accuracy, load, and user preference measures were input into a cost function. Formulating and solving as a linear assignment problem, a minimum cost combination was generated. Results from experiments evaluating efficacy in practical use cases for pouring tasks indicate that the solution is significantly more effective than no feedback and has considerable advantage over an arbitrary design.

Crude oil is a mixture of hydrocarbons so complex that it is predicted to comprise as many compounds as there are genes in the human genome. Developing methods to not only recover crude oil from the ground but also to convert crude oil into desirable products is challenging due to its complex nature. Thus, the petroleum industry relies heavily on analytical techniques to characterize the oil in reservoirs prior to enhanced oil recovery efforts and to evaluate the chemical compositions of their crude oil based products. Mass spectrometry (MS) is the only analytical technique that has the potential to provide elemental composition as well as structural information for the individual compounds that comprise petroleum samples. The continuous development of ionization techniques and mass analyzers, and other instrumentation advances, have primed mass spectrometry as the go-to analytical technique for providing solutions to problems faced by the petroleum industry. The research discussed in this dissertation can be divided into three parts: developing novel mass spectrometry-based methods to characterize mixtures of saturated hydrocarbons in petroleum products (Chapters 3 and 5), exploring the cause of fragmentation of saturated hydrocarbons upon atmospheric pressure chemical ionization to improve the analysis of samples containing these compounds (Chapter 4), and developing a better understanding of the chemical composition of crude oil that tightly binds to reservoir surfaces to improve chemically enhanced oil recovery (Chapter 6).

This dissertation provides a novel methodology to assess the impact of additive manufacturing (AM) on the aerospace supply chain. The focus is serialized production of structural parts for the aeroengine. This methodology has three fundamental steps. First, a screening heuristic is used to identify which parts and assemblies would be candidates for AM displacement. Secondly, the production line is characterized and evaluated to understand how these changes in the bill of material might impact plant workflow, and ultimately, part and assembly cost. Finally, the third step employs an integer linear program (ILP) to predict the impact on the supply chain network. The network nodes represent the various companies – and depending upon their tier – each tier has a dedicated function. The output of the ILP is the quantity and connectivity of these nodes between the tiers.

It was determined that additive manufacturing can be used to displace certain conventional manufacturing parts and assemblies as additive manufacturing’s technology matures sufficiently. Additive manufacturing is particularly powerful if adopted by the artifact’s design authority (usually the original equipment manufacturer – OEM) since it can then print its own parts on demand. Given this sourcing flexibility, these entities can in turn apply pricing pressure on its suppliers. This phenomena increasing has been seen within the industry.