Developments in measurement technology, communication and data storage have resulted in measurement systems that produce large amount of data. Together with the long existing need for characterizing the performance of the power system this has resulted in demand for automatic and efficient information-extraction methods. The objective of the research work presented in this thesis was therefore to develop new robust methods that extract additional information from voltage and current measurements in power systems. This work has contributed to two specific areas of interest.The first part of the work has been the development of a measurement method that gives information how voltage flicker propagates (with respect to a monitoring point) and how to trace a flicker source. As part of this work the quantity of flicker power has been defined and integrated in a perceptionally relevant measurement method. The method has been validated by theoretical analysis, by simulations, and by two field tests (at low-voltage and at 130-kV level) with results that matched the theory. The conclusion of this part of the work is that flicker power can be used for efficient tracing of a flicker source and to determine how flicker propagates.The second part of the work has been the development of a voltage disturbance classification system based on the statistical learning theory-based Support Vector Machine method. The classification system shows always high classification accuracy when training data and test data originate from the same source. High classification accuracy is also obtained when training data originate from one power network and test data from another. The classification system shows, however, lower performance when training data is synthetic and test data originate from real power networks. It was concluded that it is possible to develop a classification system based on the Support Vector Machine method with “global settings” that can be used at any location without the need to retrain. The conclusion is that the proposed classification system works well and shows sufficiently high classification accuracy when trained on data that originate from real disturbances. However, more research activities are needed in order to generate synthetic data that have statistical characteristics close enough to real disturbances to replace actual recordings as training data.

Tanzania’s Ruaha landscape is an international priority area for large carnivore conservation, harbouring roughly 10% of the world’s lions, and important populations of leopards and spotted hyaenas. However, these large carnivore populations are threatened by intense retaliatory killing due to human-carnivore conflict on village land around Ruaha National Park (RNP), mostly as a result of livestock predation by lions, leopards and spotted hyaenas. Moreover, a current lack of ecological data on the distribution of these carnivores hinders the development of effective strategies for conservation and targeted conflict mitigation in this landscape. This study aimed to identify the most significant ecogeographical variables (EGVs) influencing the distribution of lions, leopards and spotted hyaenas across the Ruaha landscape, and to map areas of conservation importance for these species. In addition, the study assessed the influence of EGVs on livestock predation risk by these carnivores in the village land around RNP, and generated a predictive map of predation risk. The relative importance of livestock husbandry practices and EGVs in terms of influencing predation risk within enclosures was also investigated. Proximity to rivers was the most important variable influencing the distribution of large carnivores in Ruaha, and contributed to predation risk of grazing livestock. The traditional livestock husbandry adopted in bomas appeared insufficient to alleviate the inherent risk of predation by large carnivores. The study produced the first detailed maps of lion, leopard and spotted hyaena distribution in the critically important Ruaha landscape, and identified likely livestock depredation hotspots. These results will target conflict mitigation approaches around Ruaha, by identifying particularly high-risk areas for livestock enclosures and grazing stock. Improving husbandry in these areas could help reduce livestock depredation and retaliatory carnivore killing, therefore reducing one of the most significant conservation threats in this critically important landscape.

This dissertation presents a sampling-based method that can be used for uncertainty quantification and deterministic or probabilistic optimization. The objective is to simultaneously address several difficulties faced by classical techniques based on response values and their gradients. In particular, this research addresses issues with discontinuous and binary (pass or fail) responses, and multiple failure modes. All methods in this research are developed with the aim of addressing problems that have limited data due to high cost of computation or experiment, e.g. vehicle crashworthiness, fluid-structure interaction etc.The core idea of this research is to construct an explicit boundary separating allowable and unallowable behaviors, based on classification information of responses instead of their actual values. As a result, the proposed method is naturally suited to handle discontinuities and binary states. A machine learning technique referred to as support vector machines (SVMs) is used to construct the explicit boundaries. SVM boundaries can be highly nonlinear, which allows one to use a single SVM for representing multiple failure modes.One of the major concerns in the design and uncertainty quantification communities is to reduce computational costs. To address this issue, several adaptive sampling methods have been developed as part of this dissertation. Specific sampling methods have been developed for reliability assessment, deterministic optimization, and reliability-based design optimization. Adaptive sampling allows the construction of accurate SVMs with limited samples. However, like any approximation method, construction of SVM is subject to errors. A new method to quantify the prediction error of SVMs, based on probabilistic support vector machines (PSVMs) is also developed. It is used to provide a relatively conservative probability of failure to mitigate some of the adverse effects of an inaccurate SVM. In the context of reliability assessment, the proposed method is presented for uncertainties represented by random variables as well as spatially varying random fields.In order to validate the developed methods, analytical problems with known solutions are used. In addition, the approach is applied to some application problems, such as structural impact and tolerance optimization, to demonstrate its strengths in the context of discontinuous responses and multiple failure modes.

This thesis aims to contribute to the area of kernel methods, which are a class of machine learning methods known for their wide applicability and state-of-the-art performance, but which suffer from high training and evaluation complexity. The work in this thesis utilizes the notion of reduced-set models to alleviate the training and testing complexities of these methods in a unified manner. In the first part of the thesis, we use recent results in kernel smoothing and integral-operator learning to design a generic strategy to speed up various kernel methods. In Chapter 3, we present a method to speed up kernel PCA (KPCA), which is one of the fundamental kernel methods for manifold learning, by using reduced-set density estimates (RSDE) of the data. The proposed method induces an integral operator that is an approximation of the ideal integral operator associated to KPCA. It is shown that the error between the ideal and approximate integral operators is related to the error between the ideal and approximate kernel density estimates of the data. In Chapter 4, we derive similar approximation algorithms for Gaussian process regression, diffusion maps, and kernel embeddings of conditional distributions. In the second part of the thesis, we use reduced-set models for kernel methods to tackle online learning in model-reference adaptive control (MRAC). In Chapter 5, we relate the properties of the feature spaces induced by Mercer kernels to make a connection between persistency-of-excitation and the budgeted placement of kernels to minimize tracking and modeling error. In Chapter 6, we use a Gaussian process (GP) formulation of the modeling error to accommodate a larger class of errors, and design a reduced-set algorithm to learn a GP model of the modeling error. Proofs of stability for all the algorithms are presented, and simulation results on a challenging control problem validate the methods.

A non-invasive, on-line method for detection of mechanical (rotor, bearings eccentricity) and stator winding faults in a 3-phase induction motors from observation of motor line current supply input. The main aim is to avoid the consequence of unexpected failure of critical equipment which results in extended process shutdown, costly machinery repair, and health and safety problems. This thesis looks into the possibility of utilizing machine learning techniques in the field of condition monitoring of electromechanical systems. Induction motors are chosen as an example for such application. Electrical motors play a vital role in our everyday life. Induction motors are kept in operation through monitoring its condition in a continuous manner in order to minimise their off times. The author proposes a model free sensor-less monitoring system, where the only monitored signal is the input to the induction motor. The thesis considers different methods available in literature for condition monitoring of induction motors and adopts a simple solution that is based on monitoring of the motor current. The method proposed use the feature extraction and Support Vector Machines (SVM) to set the limits for healthy and faulty data based on the statistical methods. After an extensive overview of the related literature and studies, the motor which is the virtual sensor in the drive system is analysed by considering its construction and principle of operation. The mathematical model of the motor is used for analysing the system. This is followed by laboratory testing of healthy motors and comparing their output signals with those of the same motors after being intentionally failed, concluding with the development of a full monitoring system. Finally, a monitoring system is proposed that can detect the presence of a fault in the monitored machine and diagnose the fault type and severity

The dissertation introduces Sphere Support Vector Machines (SphereSVM) and Minimal Norm Support Vector Machines (MNSVM) as the new fast classification algorithms that use geometrical properties of the underlying classification problems to efficiently obtain models describing training data. SphereSVM is based on combining minimal enclosing ball approach, state of the art nearest point problem solvers and probabilistic techniques. The blending of the three speeds up the training phase of SVMs significantly and reaches similar (i.e., practically the same) accuracy as the other classification models over several big and large real data sets within the strict validation frame of a double (nested) cross-validation (CV). MNSVM is further simplification of SphereSVM algorithm. Here, relatively complex classification task was converted into one of the simplest geometrical problems -- minimal norm problem. This resulted in additional speedup compared to SphereSVM. The results shown are promoting both SphereSVM and MNSVM as outstanding alternatives for handling large and ultra-large datasets in a reasonable time without switching to various parallelization schemes for SVMs algorithms proposed recently. The variants of both algorithms, which work without explicit bias term, are also presented. In addition, other techniques aiming to improve the time efficiency are discussed (such as over-relaxation and improved support vector selection scheme). Finally, the accuracy and performance of all these modifications are carefully analyzed and results based on nested cross-validation procedure are shown.

The aim of this work is to implement a conversion from the scanned math formula to the editable form as a TEX file as an extension of the working OCR system. In this work we closely analyze this problem, its division into several smaller parts, such as math symbol recognition and a recognition of structure of math formulas, and their solutions together with a description of various solutions. We test our implementations using our database of symbols and math formulas. An important part of the work is also a creation of a set of complex applications with a sophisticated graphical user interface, which allow easy accommodation of conversion to the user's needs. During the conversion we work with images, which may contain insignificant noise caused by a scanner of lower quality.

The learning using privileged information paradigm has allowed support vector machine models to incorporate privileged information, variables available in the training set but not in the test set, to improve predictive ability. The consequent introduction of the knowledge transfer method has enabled a practical application of support vector machine models utilizing privileged information. This thesis describes a modified knowledge transfer method inspired by cross-validation, which unlike the current standard knowledge transfer method does not create the knowledge transfer function and the approximated privileged features used in the support vector machines on the same observations. The modified method, the robust knowledge transfer, is described and evaluated versus the standard knowledge transfer method and is shown to be able to improve the predictive performance of the support vector machines for both binary classification and regression.

Treinamento semi-supervisionado é uma metodologia de aprendizado de máquina que conjuga características de treinamento supervisionado e não-supervisionado. Ela se baseia no uso de bases semi-rotuladas (bases contendo dados rotulados e não-rotulados) para o treinamento de classificadores. A adição de dados não-rotulados, mais baratos e geralmente disponíveis em maior quantidade do que os dados rotulados, pode aumentar o desempenho e/ou baratear o custo de treinamento desses classificadores (a partir da diminuição da quantidade de dados rotulados necessários). Esta dissertação analisa duas estratégias para se executar treinamento semi-supervisionado, especificamente em Support Vector Machines (SVMs): formas direta e indireta. A estratégia direta é atualmente mais conhecida e estudada, e permite o uso de dados rotulados e não-rotulados, ao mesmo tempo, em tarefas de aprendizagem de classificadores. Entretanto, a inclusão de muitos dados não-rotulados pode tornar o treinamento demasiadamente lento. Já a estratégia indireta é mais recente, sendo capaz de agregar os benefícios do treinamento semi-supervisionado direto com tempos menores para o aprendizado de classificadores. Esta opção utiliza os dados não-rotulados para pré-processar a base de dados previamente à tarefa de aprendizagem do classificador, permitindo, por exemplo, a filtragem de eventuais ruídos e a reescrita da base em espaços de variáveis mais convenientes. Dentro do escopo da forma indireta, está a principal contribuição dessa dissertação: idealização, implementação e análise do algoritmo split learning. Foram obtidos ótimos resultados com esse algoritmo, que se mostrou eficiente em treinar SVMs de melhor desempenho e em períodos menores a partir de bases semi-rotuladas. / Semi-supervised learning is a machine learning methodology that mixes features of supervised and unsupervised learning. It allows the use of partially labeled databases (databases with labeled and unlabeled data) to train classifiers. The addition of unlabeled data, which are cheaper and generally more available than labeled data, can enhance the performance and/or decrease the costs of learning such classifiers (by diminishing the quantity of required labeled data). This work analyzes two strategies to perform semi-supervised learning, specifically with Support Vector Machines (SVMs): direct and indirect concepts. The direct strategy is currently more popular and studied; it allows the use of labeled and unlabeled data, concomitantly, in learning classifiers tasks. However, the addition of many unlabeled data can lead to very long training times. The indirect strategy is more recent; it is able to attain the advantages of the direct semi-supervised learning with shorter training times. This alternative uses the unlabeled data to pre-process the database prior to the learning task; it allows denoising and rewriting the data in better feature espaces. The main contribution of this Master thesis lies within the indirect strategy: conceptualization, experimentation, and analysis of the split learning algorithm, that can be used to perform indirect semi-supervised learning using SVMs. We have obtained promising empirical results with this algorithm, which is efficient to train better performance SVMs in shorter times from partially labeled databases.

A detecção de patologias na laringe tem ocorrido basicamente por meio de diagnósticos médicos apoiados em videolaringoscopia, que é considerado um procedimento invasivo e causa certo deconforto ao paciente. Além disso, esse tipo de exame é realizado com solicitação médica e apenas quando as alterações na fala já são marcantes ou estão causando dor. Nesse ponto, muitas vezes a doença já está em grau avançado, dificultando o seu tratamento. Com o objetivo de realizar um pré-diagnóstico de tais patologias, este trabalho propõe uma técnica não invasiva baseada em um novo algoritmo que combina duas Máquinas de Vetores de Suporte, treinadas com o uso de um procedimento de aprendizado semi-supervisionado, alimentadas por um conjunto de parâmetros obtidos com o uso da Transformada Wavelet Discreta do sinal de voz do locutor. Os testes realizados com uma base de dados de vozes normais e afetadas por diversas patologias demonstram a eficácia da técnica proposta que pode, inclusive, ser implementada em tempo-real. / Larynx pathology detection is a process that depends basically on medical diagnosis and is based on videolaringoscopy, which is considered as being an invasive and uncomfortable procedure. Furthermore, this kind of examination depends on a physicists requirement and is carried out only when speech is considerably modified or causing pain. At that level, the problem is in an advanced stage which difficults its treatment. In order to get a pre-diagnosis of such pathologies, this work proposes a non-invasive technique which is based on a new algorithm that combines two support vector machines, trained with a semi-supervised approach, powered by a set of parameters derived from the discrete wavelet transform of the speakers voice signal. Tests carried out with the use of a database of normal and pathological voices show the efficacy of the proposed technique which can also be implemented for use in real-time.