To see the other types of publications on this topic, follow the link: Kernal svm.
Dissertations / Theses on the topic 'Kernal svm'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 44 dissertations / theses for your research on the topic 'Kernal svm.'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse dissertations / theses on a wide variety of disciplines and organise your bibliography correctly.
1
Sýkora, Michal. "Automatické označování obrázků." Master's thesis, Vysoké učení technické v Brně. Fakulta informačních technologií, 2012. http://www.nusl.cz/ntk/nusl-236453.
Full textAbstract:
This work focuses on automatic classification of images into semantic classes based on their contentc, especially in using SVM classifiers. The main objective of this work is to improve classification accuracy on large datasets. Both linear and nonlinear SVM classifiers are considered. In addition, the possibility of transforming features by Restricted Boltzmann Machines and using linear SVM is explored as well. All these approaches are compared in terms of accuracy, computational demands, resource utilization, and possibilities for future research.
2
Bezerra, Pedro Correia Santos. "SVR-GARCH com misturas de kernels gaussianos." reponame:Repositório Institucional da UnB, 2016. http://repositorio.unb.br/handle/10482/20864.
Full textAbstract:
Dissertação (mestrado)—Universidade de Brasília, Departamento de Administração, Programa de Pós-graduação em Administração, 2016.<br>Submitted by Fernanda Percia França (fernandafranca@bce.unb.br) on 2016-06-24T13:54:25Z
No. of bitstreams: 1
2016_PedroCorreiaSantosBezerra.pdf: 1873991 bytes, checksum: 4cf775ac8f467cc83417f0bfde464f97 (MD5)<br>Approved for entry into archive by Raquel Viana(raquelviana@bce.unb.br) on 2016-07-04T20:32:00Z (GMT) No. of bitstreams: 1
2016_PedroCorreiaSantosBezerra.pdf: 1873991 bytes, checksum: 4cf775ac8f467cc83417f0bfde464f97 (MD5)<br>Made available in DSpace on 2016-07-04T20:32:00Z (GMT). No. of bitstreams: 1
2016_PedroCorreiaSantosBezerra.pdf: 1873991 bytes, checksum: 4cf775ac8f467cc83417f0bfde464f97 (MD5)<br>Durante o desenvolvimento deste trabalho o autor recebeu auxílio financeiro da CAPES.<br>A previsão da volatilidade dos retornos financeiros é fundamental em finanças empíricas. Nos últimos 15 anos, a máquina de suporte vetorial para regressão (Support Vector Regression (SVR)) foi proposta na literatura para estimação e previsão da volatilidade devido à sua capacidade de modelar as caudas pesadas, agrupamento de volatilidade e efeito de alavancagem dos retornos financeiros (Santamaria-Bonfil et al., 2015, Cavalcante et al., 2016). Evidências empíricas sugerem que o mercado de capitais oscila entre vários estados (ou regimes) (BenSaida, 2015), em que a distribuição global dos retornos é uma mistura de distribuições normais (Levy e Klaplansky, 2015). Neste contexto, o objetivo deste trabalho foi implementar misturas de kernels gaussianos no modelo SVR com variáveis de entrada do GARCH (1,1) (denominado SVR-GARCH) para capturar os regimes de mercado e aprimorar as previsões da volatilidade. O SVR-GARCH com combinação convexa de um, dois três e quatro kernels gaussianos foi comparado com o random walk, SVR-GARCH com kernel de ondaleta de Morlet, SVR-GARCH com kernel de ondaleta de Chapéu Mexicano, GARCH(1,1), EGARCH(1,1) e GJR(1,1) com distribuição normal, t-Student, t-Student assimétrica e distribuição de erro generalizada (GED) para a série de log-retornos diários do Ibovespa de 22 de dezembro de 2007 a 04 de janeiro de 2016. Para selecionar os parâmetros ótimos do SVR e do kernel, utilizou-se a técnica de validação combinada com o procedimento de grid-search e análise de sensibilidade. Para comparar o desempenho preditivo dos modelos, utilizou-se o Erro Quadrático Médio (MSE), Erro Quadrático Normalizado (NMSE), Raiz Quadrada do Erro Quadrático Médio (RMSE) e o teste de Diebold-Mariano. Os resultados empíricos indicam que o modelo SVR-GARCH com kernel de ondaleta de Chapéu Mexicano e random walk têm desempenho preditivo superior em relação aos demais modelos. Ademais, o SVR-GARCH com mistura de dois, três e quatro kernels gaussianos é superior ao SVR-GARCH com kernel de ondaleta de Morlet e um kernel gaussiano, o que também é uma novidade e contribuição deste trabalho. Por fim, esta dissertação confirma os achados da literatura em relação à superioridade do SVR na modelagem dos fatos estilizados da volatilidade das séries financeiras em relação aos modelos GARCH linear e não-linear com caudas pesadas. ________________________________________________________________________________________________ ABSTRACT<br>Volatility forecasting plays an important role in empirical finance. In the last 15 years, a number of studies has used the Support Vector Regression to estimate and predict volatility due to its ability to model leptokurtosis, volatility clustering, and leverage effect of financial returns (Santamaria-Bonfil et al., 2015, Cavalcante et al., 2016). Empirical evidence suggests that the capital market oscillates between several states (or regimes) (BenSaida, 2015), in which the overall distribution of returns is a mixture of normal distributions (Levy and Klaplansky, 2015). In this context, the objective of this dissertation is to use a mixture of Gaussian kernels in the SVR based on GARCH (1,1) (heretofore SVR-GARCH) in order to capture the regime behavior and to improve the one-period-ahead volatility forecasts. In order to choose the SVR parameters, I used the validation technique (holdout method) based on grid-search and sensitivity analysis. The SVR-GARCH with a linear combination of one, two, three and four Gaussian kernels is compared with \textit{random walk}, SVR-GARCH with Morlet wavelet kernel, SVR-GARCH with Mexican Hat wavelet kernel, GARCH, GJR and EGARCH models with normal, student-t, skewstudent- t and Generalized Error Distribution (GED) innovations by using the Mean Squared Error (MSE), Normalized Mean Squared Error (NMSE), Root Mean Squared Error (RMSE) and Diebold Mariano test. The out-sample results for the Ibovespa daily closing price from August 20, 2013 to January 04, 2016 shows that the random walk model and SVR-GARCH with Mexican Hat wavelet kernel provide the most accurate forecasts. The outcomes also highlight the fact that the SVR GARCH with a mixture of two, three and four Gaussian kernels has superior results than the SVR GARCH with Morlet wavelet kernel and a single Gaussian kernel. Moreover, consistent with the findings of the literature, I confirm that the SVR has superior empirical results in modeling financial time series stylized facts than the linear and non-linear GARCH models with fat-tailed distributions.
3
Long, Andrew Edmund. "Cokriging, kernels, and the SVD: Toward better geostatistical analysis." Diss., The University of Arizona, 1994. http://hdl.handle.net/10150/186892.
Full textAbstract:
Three forms of multivariate analysis, one very classical and the other two relatively new and little-known, are showcased and enhanced: the first is the Singular Value Decomposition (SVD), which is at the heart of many statistical, and now geostatistical, techniques; the second is the method of Variogram Analysis, which is one way of investigating spatial correlation in one or several variables; and the third is the process of interpolation known as cokriging, a method for optimizing the estimation of multivariate data based on the information provided through variogram analysis. The SVD is described in detail, and it is shown that the SVD can be generalized from its familiar matrix (two-dimensional) case to three, and possibly n, dimensions. This generalization we call the "Tensor SVD" (or TSVD), and we demonstrate useful applications in the field of geostatistics (and indicate ways in which it will be useful in other areas). Applications of the SVD to the tools of geostatistics are described: in particular, applications dependent on the TSVD, including variogram modelling in coregionalization. Variogram analysis in general is explored, and we propose broader use of an old tool (which we call the "corhogram ", based on the variogram) which proves useful in helping one choose variables for multivariate interpolation. The reasoning behind kriging and cokriging is discussed, and a better algorithm for solving the cokriging equations is developed, which results in simultaneous kriging estimates for comparison with those obtained from cokriging. Links from kriging systems to kernel systems are made; discovering kerneIs equivalent to kriging systems will be useful in the case where data are plentiful. Finally, some results of the application of geostatistical techniques to a data set concerning nitrate pollution in the West Salt River Valley of Arizona are described.
4
JÃnior, OtÃvio AlcÃntara de Lima. "SKMotes: um Kernel semipreemptivo para nÃs de redes de sensores sem fio." Universidade Federal do CearÃ, 2011. http://www.teses.ufc.br/tde_busca/arquivo.php?codArquivo=6767.
Full textAbstract:
Redes de sensores sem Fio (RSSFs) sÃo fruto dos recentes avanÃos nas tecnologias de sistemas micro-eletro-mecÃnicos, circuitos integrados de baixa potÃncia e comunicaÃÃo sem fio de baixa potÃncia. Estes avanÃos permitiram a criaÃÃo de minÃsculos dispositivos computacionais de baixo custo e baixa potÃncia, capazes de monitorar grandezas fÃsicas do ambiente e estabelecer comunicaÃÃo uns com os outros. Estes dispositivos, denominados nÃs sensores, sÃo dotados de um microcontrolador simples, elementos sensores, rÃdio transceptor e fonte de alimentaÃÃo.
O sistema operacional (SO) Ã um componente essencial de um projeto de uma
aplicaÃÃo para RSSFs. Em relaÃÃo ao modelo de concorrÃncia, podem-se dividir os SOs
em duas categorias: baseados em eventos e baseados em threads. O modelo baseado em
eventos cria diculdades ao programador para controlar os
uxos de execuÃÃo e nÃo se
ajusta a problemas com longos perÃodos de computaÃÃao. Por outro lado, o modelo baseado
em threads tem alto consumo de memÃria, mas fornece um modelo de programaÃÃo mais
simples e com bons tempos de resposta. Dentro desse contexto, esta dissertaÃÃo propÃe um novo SO para RSSFs, chamado SKMotes, que explora as facilidades de programaÃÃo do modelo threads aliadas à baixa ocupaÃÃo de memÃria. Este SO utiliza um modelo de concorrÃncia baseado em threads, mas nÃo completamente preemptivo, pois em dado
momento apenas um subconjunto das threads do sistema està executando no modo preemptivo baseado em prioridades. O restante das threads permanece em espera, ocupando
apenas um contexto mÃnimo de execuÃÃo, que nÃo contempla a pilha de dados. O principal objetivo desse modelo à prover tempos de resposta baixos para threads de alta prioridade,
ao mesmo tempo que garante baixo consumo de energia e ocupaÃÃo de memÃria mais baixa do que kernels preemptivos. Estas caracterÃsticas permitem que o SKMotes seja
empregado em aplicaÃÃees de RSSFs que utilizem um conjunto de tarefas orientadas à E/S e a longos perÃodos de computaÃÃo. Por exemplo, aplicaÃÃes de RSSFs que realizem funÃÃes de compressÃo de dados, criptografia, dentre outras. A avaliaÃÃo de desempenho do SO proposto foi realizada em um ambiente de testes, baseado em uma FPGA, projetado para esta dissertaÃÃo, que permite realizar mediÃÃes da utilizaÃÃo da CPU e do tempo de
resposta das threads, ao mesmo tempo em que interage com a plataforma do nà sensor atravÃs da interface de comunicaÃÃo serial. Este ambiente de testes pode ser reutilizado em diferentes cenÃrios de avaliaÃÃo de desempenho de sistemas computacionais baseados em microcontroladores. Os testes de avaliaÃÃo de desempenho mostram que, para os casos
de teste realizados, o SKMotes apresenta ocupaÃÃo do processador equivalente Ãs soluÃÃes baseadas em multithreading preemptivo, mas com consumo de memÃria de dados, em
mÃdia, 20% menor. AlÃm disso, o SKMotes à capaz de garantir tempos de respostas, em mÃdia, 34% inferiores Ãs soluÃÃes baseadas em kernels de eventos. Quando se avalia
apenas os casos de teste que possuem threads orientadas à E/S e a longos perÃodos de computaÃÃo, o tempo de resposta chega a ser, em mÃdia, 63% inferior ao apresentado por kernels baseados em eventos.
5
Lima, Júnior Otávio Alcântara de. "SKMotes : um kernel semipreemptivo para nós de redes de sensores sem fio." reponame:Repositório Institucional da UFC, 2011. http://www.repositorio.ufc.br/handle/riufc/2043.
Full textAbstract:
LIMA JÚNIOR, O. A. de. SKMotes : um kernel semipreemptivo para nós de redes de sensores sem fio. 2011. 91 f. Dissertação (mestrado em Engenharia de Teleinformática) - Centro de Tecnologia, Universidade Federal do Ceará, Fortaleza, 2011.<br>Submitted by Marlene Sousa (mmarlene@ufc.br) on 2012-02-10T19:02:03Z
No. of bitstreams: 1
2011-dis_oalimajunior.pdf: 2767952 bytes, checksum: 003409fd0de13eaaac0200e8f9b5a633 (MD5)<br>Approved for entry into archive by Marlene Sousa(mmarlene@ufc.br) on 2012-02-10T19:02:31Z (GMT) No. of bitstreams: 1
2011-dis_oalimajunior.pdf: 2767952 bytes, checksum: 003409fd0de13eaaac0200e8f9b5a633 (MD5)<br>Made available in DSpace on 2012-02-10T19:02:31Z (GMT). No. of bitstreams: 1
2011-dis_oalimajunior.pdf: 2767952 bytes, checksum: 003409fd0de13eaaac0200e8f9b5a633 (MD5)
Previous issue date: 2011-10<br>The ever-increasing developments of low-power integrated circuits have made it possible the design of very small low-cost and low-power electronic sensors with wireless communication and computing capabilities. Those devices, in their turn, made it feasible the implementation of the so-called Wireless Sensors Networks (WSN). WSN is a network of such devices (known as nodes), each one having an embedded microcontroller and a communication module which makes it possible the nodes to be used as sensors which process and exchange information with the other nodes, in order to achieve a speci c purpose.
Usually, due to the nodes very limited processing power, a very simple operating
system (SO) is used to manage the node's processing and communicating capabilities by
executing tasks in a concurrent fashion. The SO is a very important part in the design of a WSN and, depending on the concurrence model used on its design, the SO can be divided into two types: event-based or thread-based SO's. Event-based models make it di cult for the programmer to control the execution ow and are not suitable for tasks with long computation time. Thread-based models, on the other hand, present heavy memory use, but have a much simpler programming model and good real-time responses. In this sense, this dissertation proposes a new semi-preemptive SO, called SKMotes has the relatively easy-programming model related to thread-based models and a low memory usage. Despite SKMotes be thread-based, it is not fully preemptive, since at any given time, only a subset of the system's threads is executing as preemptive priority-based tasks and the rest of them remains on hold, which makes for low context usage, since the threads do not need data stack. This approach provides low time response for high-priority threads
while at the same time guarantees lower memory usage than that of preemptive kernels.
These features make SKMotes very suitable for WSN applications where there is a
combination of I/O-oriented tasks and task with long computation times (for example, applications that perform data compression and/or cryptography). After being implemented, SKMotes' performance analysis was carried out by using a specially-designed
FPGA-based module, which made it possible to perform CPU-usage measurements as well as threads' time response, with the system on the y. The measurement's results showed that, for the considered test-scenario, SKMotes presents CPU-usage rates equal
to preemptive multi-threading approaches but having a lower memory usage (20).<br>Redes de Sensores sem Fio (RSSFs) são fruto dos recentes avan cos nas tecnologias
de sistemas micro-eletro-mecânicos, circuitos integrados de baixa potência e comunicação sem baixa potência. Estes avan ços permitiram a cria ção de min usculos dispositivos computacionais de baixo custo e baixa potência, capazes de monitorar grandezas fí sicas do ambiente e estabelecer comunica ção uns com os outros. Estes dispositivos, denominados n os sensores, são dotados de um microcontrolador simples, elementos sensores, r adio transceptor e fonte de alimenta ção. Desenvolver aplica çoes para RSSFs ée um grande desafio. O sistema operacional (SO) ée um componente essencial de um projeto de uma aplica ção para RSSFs. Em rela ção ao modelo de concorrência, podem-se dividir os SOs em duas categorias: baseados em eventos e baseados em threads. O modelo baseado em eventos cria dificuldades ao programador para controlar os fluxos de execu ção e não se ajusta a problemas com longos perí odos de computação. Por outro lado, o modelo baseado em threads tem alto consumo de mem oria, mas fornece um modelo de programa ção mais simples e com bons tempos de resposta. Dentro desse contexto, esta disserta ção propõe um novo SO para RSSFs, chamado SKMotes, que explora as facilidades de programa ção do modelo threads aliadas a baixa ocupa ção de mem oria. Este SO utiliza um modelo de concorrência baseado em threads, mas não completamente preemptivo, pois em dado momento apenas um subconjunto das threads do sistema est a executando no modo preemptivo baseado em prioridades. O restante das threads permanece em espera, ocupando apenas um contexto m nimo de execu ção, que não contempla a pilha de dados. O principal objetivo desse modelo é prover tempos de resposta baixos para threads de alta prioridade, ao mesmo tempo que garante baixo consumo de energia e ocupa ção de mem oria mais baixa do que kernels preemptivos. Estas caracter sticas permitem que o SKMotes seja empregado em aplica ções de RSSFs que utilizem um conjunto de tarefas orientadas a E/S e a longos per íodos de computação. Por exemplo, aplica ções de RSSFs que realizem fun ções de compressão de dados, criptogra a, dentre outras. A avalia ção de desempenho
do SO proposto foi realizada em um ambiente de testes, baseado em uma FPGA, projetado
para esta disserta ção, que permite realizar medi ções da utiliza ção da CPU e do tempo de
resposta das threads, ao mesmo tempo em que interage com a plataforma do n o sensor
atrav es da interface de comunica ção serial. Este ambiente de testes pode ser reutilizado
em diferentes cen arios de avaliação de desempenho de sistemas computacionais baseados em microcontroladores. Os testes de avalia ção de desempenho mostram que, para os casos de teste realizados, o SKMotes apresenta ocupa ção do processador equivalente as solu ções baseadas em multithreading preemptivo, mas com consumo de mem oria de dados, em m edia, 20% menor. Al em disso, o SKMotes e capaz de garantir tempos de respostas, em m edia, 34% inferiores as solu ções baseadas em kernels de eventos. Quando se avalia apenas os casos de teste que possuem threads orientadas a E/S e a longos per odos de computação, o tempo de resposta chega a ser, em m edia, 63% inferior ao apresentado por kernels baseados em eventos.
6
Bonesso, Diego. "Estimação dos parâmetros do kernel em um classificador SVM na classificação de imagens hiperespectrais em uma abordagem multiclasse." reponame:Biblioteca Digital de Teses e Dissertações da UFRGS, 2013. http://hdl.handle.net/10183/86168.
Full textAbstract:
Nessa dissertação é investigada e testada uma metodologia para otimizar os parâmetros do kernel do classificador Support Vector Machines (SVM). Experimentos são realizados utilizando dados de imagens em alta dimensão. Imagens em alta dimensão abrem novas possibilidades para a classificação de imagens de sensoriamento remoto que capturam cenas naturais. É sabido que classes que são espectralmente muito similares, i.e, classes que possuem vetores de média muito próximos podem não obstante serem separadas com alto grau de acurácia em espaço de alta dimensão, desde que a matriz de covariância apresente diferenças significativas. O uso de dados de imagens em alta dimensão pode apresentar, no entanto, alguns desafios metodológicos quando aplicado um classificador paramétrico como o classificador de Máxima Verossimilhança Gaussiana. Conforme aumenta a dimensionalidade dos dados, o número de parâmetros a serem estimados a partir de um número geralmente limitado de amostras de treinamento também aumenta. Esse fato pode ocasionar estimativas pouco confiáveis, que por sua vez resultam em baixa acurácia na imagem classificada. Existem diversos abordagens propostas na literatura para minimizar esse problema. Os classificadores não paramétricos podem ser uma boa alternativa para mitigar esse problema. O SVM atualmente tem sido investigado na classificação de dados de imagens em alta-dimensão com número limitado de amostras de treinamento. Para que o classificador SVM seja utilizado com sucesso é necessário escolher uma função de kernel adequada, bem como os parâmetros dessa função. O kernel RBF tem sido frequentemente mencionado na literatura por obter bons resultados na classificação de imagens de sensoriamento remoto. Neste caso, dois parâmetro devem ser escolhidos para o classificador SVM: (1) O parâmetro de margem (C) que determina um ponto de equilíbrio razoável entre a maximização da margem e a minimização do erro de classificação, e (2) o parâmetro que controla o raio do kernel RBF. Estes dois parâmetros podem ser vistos como definindo um espaço de busca. O problema nesse caso consiste em procurar o ponto ótimo que maximize a acurácia do classificador SVM. O método de Busca em Grade é baseado na exploração exaustiva deste espaço de busca. Esse método é proibitivo do ponto de vista do tempo de processamento, sendo utilizado apenas com propósitos comparativos. Na prática os métodos heurísticos são a abordagem mais utilizada, proporcionado níveis aceitáveis de acurácia e tempo de processamento. Na literatura diversos métodos heurísticos são aplicados ao problema de classificação de forma global, i.e, os valores selecionados são aplicados durante todo processo de classificação. Esse processo, no entanto, não considera a diversidade das classes presentes nos dados. Nessa dissertação investigamos a aplicação da heurística Simulated Annealing (Recozimento Simulado) para um problema de múltiplas classes usando o classificador SVM estruturado como uma arvore binária. Seguindo essa abordagem, os parâmetros são estimados em cada nó da arvore binária, resultado em uma melhora na acurácia e tempo razoável de processamento. Experimentos são realizados utilizando dados de uma imagem hiperespectral disponível, cobrindo uma área de teste com controle terrestre bastante confiável.<br>In this dissertation we investigate and test a methodology to optimize the kernel parameters in a Support Vector Machines classifier. Experiments were carried out using remote sensing high-dimensional image data. High dimensional image data opens new possibilities in the classification of remote sensing image data covering natural scenes. It is well known that classes that are spectrally very similar, i.e., classes that show very similar mean vectors can notwithstanding be separated with an high degree of accuracy in high dimensional spaces, provided that their covariance matrices differ significantly. The use of high-dimensional image data may present, however, some drawbacks when applied in parametric classifiers such as the Gaussian Maximum Likelihood classifier. As the data dimensionality increases, so does the number of parameters to be estimated from a generally limited number of training samples. This fact results in unreliable estimates for the parameters, which in turn results in low accuracy in the classified image. There are several approaches proposed in the literature to minimize this problem. Non-parametric classifiers may provide a sensible way to overcome this problem. Support Vector Machines (SVM) have been more recently investigated in the classification of high-dimensional image data with a limited number of training samples. To achieve this end, a proper kernel function has to be implemented in the SVM classifier and the respective parameters selected properly. The RBF kernel has been frequently mentioned in the literature as providing good results in the classification of remotely sensed data. In this case, two parameters must be chosen in the SVM classification: (1) the margin parameter (C) that determines the trade-off between the maximization of the margin in the SVM and minimization of the classification error, and (2) the parameter that controls the radius in the RBF kernel. These two parameters can be seen as defining a search space, The problem here consists in finding an optimal point that maximizes the accuracy in the SVM classifier. The Grid Search approach is based on an exhaustive exploration in the search space. This approach results prohibitively time consuming and is used only for comparative purposes. In practice heuristic methods are the most commonly used approaches, providing acceptable levels of accuracy and computing time. In the literature several heuristic methods are applied to the classification problem in a global fashion, i.e., the selected values are applied to the entire classification process. This procedure, however, does not take into consideration the diversity of the classes present in the data. In this dissertation we investigate the application of Simulated Annealing to a multiclass problem using the SVM classifier structured as a binary tree. Following this proposed approach, the parameters are estimated at every level of the binary tree, resulting in better accuracy and a reasonable computing time. Experiments are done using a set of hyperspectral image data, covering a test area with very reliable ground control available.
7
Kohram, Mojtaba. "Experiments with Support Vector Machines and Kernels." University of Cincinnati / OhioLINK, 2013. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1378112059.
Full text
8
MANSOORI, MOHAMMADAMIR. "FPGA Acceleration of Domain-specific Kernels via High-Level Synthesis." Doctoral thesis, Politecnico di Torino, 2022. http://hdl.handle.net/11583/2962967.
Full text
9
Zwald, Laurent. "PERFORMANCES STATISTIQUES D'ALGORITHMES D'APPRENTISSAGE : ``KERNEL PROJECTION MACHINE'' ET ANALYSE EN COMPOSANTES PRINCIPALES A NOYAU." Phd thesis, Université Paris Sud - Paris XI, 2005. http://tel.archives-ouvertes.fr/tel-00012011.
Full textAbstract:
La thèse se place dans le cadre de l'apprentissage statistique. Elle apporte<br />des contributions à la communauté du machine learning en utilisant des<br />techniques de statistiques modernes basées sur des avancées dans l'étude<br />des processus empiriques. Dans une première partie, les propriétés statistiques de<br />l'analyse en composantes principales à noyau (KPCA) sont explorées. Le<br />comportement de l'erreur de reconstruction est étudié avec un point de vue<br />non-asymptotique et des inégalités de concentration des valeurs propres de la matrice de<br />Gram sont données. Tous ces résultats impliquent des vitesses de<br />convergence rapides. Des propriétés <br />non-asymptotiques concernant les espaces propres de la KPCA eux-mêmes sont également<br />proposées. Dans une deuxième partie, un nouvel <br />algorithme de classification a été<br />conçu : la Kernel Projection Machine (KPM). <br />Tout en s'inspirant des Support Vector Machines (SVM), il met en lumière que la sélection d'un espace vectoriel par une méthode de<br />réduction de la dimension telle que la KPCA régularise <br />convenablement. Le choix de l'espace vectoriel utilisé par la KPM est guidé par des études statistiques de sélection de modéle par minimisation pénalisée de la perte empirique. Ce<br />principe de régularisation est étroitement relié à la projection fini-dimensionnelle étudiée dans les travaux statistiques de <br />Birgé et Massart. Les performances de la KPM et de la SVM sont ensuite comparées sur différents jeux de données. Chaque thème abordé dans cette thèse soulève de nouvelles questions d'ordre théorique et pratique.
10
Grasso, Antonio. "Sviluppo di un modello debolmente supervisionato a bassa complessità per la ricerca di anomalie in immagini." Master's thesis, Alma Mater Studiorum - Università di Bologna, 2020. http://amslaurea.unibo.it/20024/.
Full textAbstract:
Il lavoro di questa tesi nasce dall’esigenza di rilevare difetti di produzione in ambito industriale tramite sensori di visione intelligenti, che si traduce in un problema di anomaly detection nelle immagini. Le tecniche allo stato dell’arte, come gli autoencoder convoluzionali, garantiscono ottimi risultati a fronte tuttavia di un’alta complessità. In questa tesi viene dunque proposto un modello di apprendimento supervisionato compatibile con i requisiti del contesto industriale. In particolare, ci si è posti l’obiettivo di sviluppare un modello in grado di essere addestrato con un numero limitato di esempi in tempi rapidi e con un basso consumo di risorse computazionali, capace di funzionare senza la presenza di parametri di configurazione.
Tale modello si compone di un descrittore efficiente ed efficace per l’estrazione di feature dalle immagini e di un classificatore a valle. La scelta del classificatore è ricaduta sulla one-class SVM per via delle sue molteplici caratteristiche vantaggiose, tra cui la possibilità di essere addestrata in tempi rapidi. Per migliorare le performance del modello nella ricerca di anomalie, è stato prima messo a punto un metodo di data augmentation in grado di riprodurre le feature estratte dal descrittore tramite tecniche statistiche di sampling da distribuzioni discrete. In secondo luogo, sono state proposte nuove funzioni kernel per la one-class SVM in grado di sfruttare la struttura delle feature del descrittore al fine di avvicinare gli esempi simili e allontanare gli esempi diversi nello spazio delle feature. I risultati ottenuti dimostrano che il modello e le modifiche del modello proposte rappresentano una buona soluzione al problema di ricerca di anomalie nelle immagini capace di rispettare i requisiti di bassa complessità del contesto industriale.
11
Wang, Xiaoguang. "Design and Analysis of Techniques for Multiple-Instance Learning in the Presence of Balanced and Skewed Class Distributions." Thesis, Université d'Ottawa / University of Ottawa, 2015. http://hdl.handle.net/10393/32184.
Full textAbstract:
With the continuous expansion of data availability in many large-scale, complex, and
networked systems, such as surveillance, security, the Internet, and finance, it becomes critical to advance the fundamental understanding of knowledge discovery and analysis from raw data to support decision-making processes. Existing knowledge discovery and data analyzing techniques have shown great success in many real-world applications such as applying Automatic Target Recognition (ATR) methods to detect targets of interest in imagery, drug activity prediction, computer vision recognition, and so on. Among these techniques, Multiple-Instance (MI) learning is different from standard classification since it uses a set of bags containing many instances as input. The instances in each bag are not labeled | instead the bags themselves are labeled. In this area many researchers have accomplished a lot of work and made a lot of progress. However, there still exist some areas which are not covered. In this thesis, we focus on two topics of MI learning: (1) Investigating the relationship between MI learning and other multiple pattern learning
methods, which include multi-view learning, data fusion method and multi-kernel SVM. (2) Dealing with the class imbalance problem of MI learning.
In the first topic, three different learning frameworks will be presented for general MI
learning. The first uses multiple view approaches to deal with MI problem, the second is a data fusion framework, and the third framework, which is an extension of the first framework, uses multiple-kernel SVM. Experimental results show that the approaches presented work well on solving MI problem.
The second topic is concerned with the imbalanced MI problem. Here we investigate
the performance of learning algorithms in the presence of underrepresented data and severe class distribution skews. For this problem, we propose three solution frameworks: a data re-sampling framework, a cost-sensitive boosting framework and an adaptive instance-weighted boosting SVM (with the name IB_SVM) for MI learning. Experimental results - on both benchmark datasets and application datasets - show that the proposed frameworks are proved to be effective solutions for the imbalanced problem of MI learning.
12
Wehmann, Adam. "A Spatial-Temporal Contextual Kernel Method for Generating High-Quality Land-Cover Time Series." The Ohio State University, 2014. http://rave.ohiolink.edu/etdc/view?acc_num=osu1398866264.
Full text
13
Marek, Tomáš. "Klasifikace dokumentů podle tématu." Master's thesis, Vysoké učení technické v Brně. Fakulta informačních technologií, 2013. http://www.nusl.cz/ntk/nusl-236158.
Full textAbstract:
This thesis deals with a document classification, especially with a text classification method. Main goal of this thesis is to analyze two arbitrary document classification algorithms to describe them and to create an implementation of those algorithms. Chosen algorithms are Bayes classifier and classifier based on support vector machines (SVM) which were analyzed and implemented in the practical part of this thesis. One of the main goals of this thesis is to create and choose optimal text features, which are describing the input text best and thus lead to the best classification results. At the end of this thesis there is a bunch of tests showing comparison of efficiency of the chosen classifiers under various conditions.
14
Henchiri, Yousri. "L'approche Support Vector Machines (SVM) pour le traitement des données fonctionnelles." Thesis, Montpellier 2, 2013. http://www.theses.fr/2013MON20187/document.
Full textAbstract:
L'Analyse des Données Fonctionnelles est un domaine important et dynamique en statistique. Elle offre des outils efficaces et propose de nouveaux développements méthodologiques et théoriques en présence de données de type fonctionnel (fonctions, courbes, surfaces, ...). Le travail exposé dans cette thèse apporte une nouvelle contribution aux thèmes de l'apprentissage statistique et des quantiles conditionnels lorsque les données sont assimilables à des fonctions. Une attention particulière a été réservée à l'utilisation de la technique Support Vector Machines (SVM). Cette technique fait intervenir la notion d'Espace de Hilbert à Noyau Reproduisant. Dans ce cadre, l'objectif principal est d'étendre cette technique non-paramétrique d'estimation aux modèles conditionnels où les données sont fonctionnelles. Nous avons étudié les aspects théoriques et le comportement pratique de la technique présentée et adaptée sur les modèles de régression suivants. Le premier modèle est le modèle fonctionnel de quantiles de régression quand la variable réponse est réelle, les variables explicatives sont à valeurs dans un espace fonctionnel de dimension infinie et les observations sont i.i.d.. Le deuxième modèle est le modèle additif fonctionnel de quantiles de régression où la variable d'intérêt réelle dépend d'un vecteur de variables explicatives fonctionnelles. Le dernier modèle est le modèle fonctionnel de quantiles de régression quand les observations sont dépendantes. Nous avons obtenu des résultats sur la consistance et les vitesses de convergence des estimateurs dans ces modèles. Des simulations ont été effectuées afin d'évaluer la performance des procédures d'inférence. Des applications sur des jeux de données réelles ont été considérées. Le bon comportement de l'estimateur SVM est ainsi mis en évidence<br>Functional Data Analysis is an important and dynamic area of statistics. It offers effective new tools and proposes new methodological and theoretical developments in the presence of functional type data (functions, curves, surfaces, ...). The work outlined in this dissertation provides a new contribution to the themes of statistical learning and quantile regression when data can be considered as functions. Special attention is devoted to use the Support Vector Machines (SVM) technique, which involves the notion of a Reproducing Kernel Hilbert Space. In this context, the main goal is to extend this nonparametric estimation technique to conditional models that take into account functional data. We investigated the theoretical aspects and practical attitude of the proposed and adapted technique to the following regression models.The first model is the conditional quantile functional model when the covariate takes its values in a bounded subspace of the functional space of infinite dimension, the response variable takes its values in a compact of the real line, and the observations are i.i.d.. The second model is the functional additive quantile regression model where the response variable depends on a vector of functional covariates. The last model is the conditional quantile functional model in the dependent functional data case. We obtained the weak consistency and a convergence rate of these estimators. Simulation studies are performed to evaluate the performance of the inference procedures. Applications to chemometrics, environmental and climatic data analysis are considered. The good behavior of the SVM estimator is thus highlighted
15
Zaremba, Wojciech. "Modeling the variability of EEG/MEG data through statistical machine learning." Habilitation à diriger des recherches, Ecole Polytechnique X, 2012. http://tel.archives-ouvertes.fr/tel-00803958.
Full textAbstract:
Brain neural activity generates electrical discharges, which manifest as electrical and magnetic potentials around the scalp. Those potentials can be registered with magnetoencephalography (MEG) and electroencephalography (EEG) devices. Data acquired by M/EEG is extremely difficult to work with due to the inherent complexity of underlying brain processes and low signal-to-noise ratio (SNR). Machine learning techniques have to be employed in order to reveal the underlying structure of the signal and to understand the brain state. This thesis explores a diverse range of machine learning techniques which model the structure of M/EEG data in order to decode the mental state. It focuses on measuring a subject's variability and on modeling intrasubject variability. We propose to measure subject variability with a spectral clustering setup. Further, we extend this approach to a unified classification framework based on Laplacian regularized support vector machine (SVM). We solve the issue of intrasubject variability by employing a model with latent variables (based on a latent SVM). Latent variables describe transformations that map samples into a comparable state. We focus mainly on intrasubject experiments to model temporal misalignment.
16
Ouyang, Hua. "Optimal stochastic and distributed algorithms for machine learning." Diss., Georgia Institute of Technology, 2013. http://hdl.handle.net/1853/49091.
Full textAbstract:
Stochastic and data-distributed optimization algorithms have received lots of attention from the machine learning community due to the tremendous demand from the large-scale learning and the big-data related optimization. A lot of stochastic and deterministic learning algorithms are proposed recently under various application scenarios. Nevertheless, many of these algorithms are based on heuristics and their optimality in terms of the generalization error is not sufficiently justified. In this talk, I will explain the concept of an optimal learning algorithm, and show that given a time budget and proper hypothesis space, only those achieving the lower bounds of the estimation error and the optimization error are optimal. Guided by this concept, we investigated the stochastic minimization of nonsmooth convex loss functions, a central problem in machine learning. We proposed a novel algorithm named Accelerated Nonsmooth Stochastic Gradient Descent, which exploits the structure of common nonsmooth loss functions to achieve optimal convergence rates for a class of problems including SVMs. It is the first stochastic algorithm that can achieve the optimal O(1/t) rate for minimizing nonsmooth loss functions. The fast rates are confirmed by empirical comparisons with state-of-the-art algorithms including the averaged SGD. The Alternating Direction Method of Multipliers (ADMM) is another flexible method to explore function structures. In the second part we proposed stochastic ADMM that can be applied to a general class of convex and nonsmooth functions, beyond the smooth and separable least squares loss used in lasso. We also demonstrate the rates of convergence for our algorithm under various structural assumptions of the stochastic function: O(1/sqrt{t}) for convex functions and O(log t/t) for strongly convex functions. A novel application named Graph-Guided SVM is proposed to demonstrate the usefulness of our algorithm. We also extend the scalability of stochastic algorithms to nonlinear kernel machines, where the problem is formulated as a constrained dual quadratic optimization. The simplex constraint can be handled by the classic Frank-Wolfe method. The proposed stochastic Frank-Wolfe methods achieve comparable or even better accuracies than state-of-the-art batch and online kernel SVM solvers, and are significantly faster. The last part investigates the problem of data-distributed learning. We formulate it as a consensus-constrained optimization problem and solve it with ADMM. It turns out that the underlying communication topology is a key factor in achieving a balance between a fast learning rate and computation resource consumption. We analyze the linear convergence behavior of consensus ADMM so as to characterize the interplay between the communication topology and the penalty parameters used in ADMM. We observe that given optimal parameters, the complete bipartite and the master-slave graphs exhibit the fastest convergence, followed by bi-regular graphs.
17
Sangnier, Maxime. "Outils d'apprentissage automatique pour la reconnaissance de signaux temporels." Rouen, 2015. http://www.theses.fr/2015ROUES064.
Full textAbstract:
Les travaux présentés ici couvrent deux thématiques de la reconnaissance de signaux temporels par des systèmes numériques dont certains paramètres sont inférés à partir d’un ensemble limité d’observations. La première est celle de la détermination automatique de caractéristiques discriminantes. Pour ce faire, nous proposons un algorithme de génération de colonnes capable d’apprendre une transformée Temps-Fréquence (TF), mise sous la forme d’un banc de filtres, de concert à une machine à vecteurs supports. Cet algorithme est une extension des techniques existantes d’apprentissage de noyaux multiples, combinant de manière non-linéaire une infinité de noyaux. La seconde thématique dans laquelle s’inscrivent nos travaux est l’appréhension de la temporalité des signaux. Si cette notion a été abordée au cours de notre première contribution, qui a pointé la nécessité de déterminer au mieux la résolution temporelle d’une représentation TF, elle prend tout son sens dans une prise de décision au plus tôt. Dans ce contexte, notre seconde contribution fournit un cadre méthodologique permettant de détecter précocement un événement particulier au sein d’une séquence, c’est à dire avant que ledit événement ne se termine. Celui-ci est construit comme une extension de l’apprentissage d’instances multiples et des espaces de similarité aux séries temporelles. De plus, nous accompagnons cet outil d’un algorithme d’apprentissage efficace et de garanties théoriques de généralisation. L’ensemble de nos travaux a été évalué sur des signaux issus d’interfaces cerveau-machine, des paysages sonores et des vidéos représentant des actions humaines<br>The work presented here tackles two different subjects in the wide thematic of how to build a numerical system to recognize temporal signals, mainly from limited observations. The first one is automatic feature extraction. For this purpose, we present a column generation algorithm, which is able to jointly learn a discriminative Time-Frequency (TF) transform, cast as a filter bank, with a support vector machine. This algorithm extends the state of the art on multiple kernel learning by non-linearly combining an infinite amount of kernels. The second direction of research is the way to handle the temporal nature of the signals. While our first contribution pointed out the importance of correctly choosing the time resolution to get a discriminative TF representation, the role of the time is clearly enlightened in early recognition of signals. Our second contribution lies in this field and introduces a methodological framework for early detection of a special event in a time-series, that is detecting an event before it ends. This framework builds upon multiple instance learning and similarity spaces by fitting them to the particular case of temporal sequences. Furthermore, our early detector comes with an efficient learning algorithm and theoretical guarantees on its generalization ability. Our two contributions have been empirically evaluated with brain-computer interface signals, soundscapes and human actions movies
18
Ahlenius, Camilla. "Automatic Pronoun Resolution for Swedish." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-289192.
Full textAbstract:
This report describes a quantitative analysis performed to compare two different methods on the task of pronoun resolution for Swedish. The first method, an implementation of Mitkov’s algorithm, is a heuristic-based method — meaning that the resolution is determined by a number of manually engineered rules regarding both syntactic and semantic information. The second method is data-driven — a Support Vector Machine (SVM) using dependency trees and word embeddings as features. Both methods are evaluated on an annotated corpus of Swedish news articles which was created as a part of this thesis. SVM-based methods significantly outperformed the implementation of Mitkov’s algorithm. The best performing SVM model relies on tree kernels applied to dependency trees. The model achieved an F1-score of 0.76 for the positive class and 0.9 for the negative class, where positives are pairs of pronoun and noun phrase that corefer, and negatives are pairs that do not corefer.<br>Rapporten beskriver en kvantitativ analys som genomförts för att jämföra två olika metoder för automatisk pronomenbestämning på svenska. Den första metoden, en implementation av Mitkovs algoritm, är en heuristisk metod vilket innebär att pronomenbestämningen görs med ett antal manuellt utformade regler som avser att fånga både syntaktisk och semantisk information. Den andra metoden är datadriven, en stödvektormaskin (SVM) som använder dependensträd och ordvektorer som särdrag. Båda metoderna utvärderades med hjälp av en annoterad datamängd bestående av svenska nyhetsartiklar som skapats som en del av denna avhandling. Den datadrivna metoden överträffade Mitkovs algoritm. Den SVM-modell som ger bäst resultat bygger på trädkärnor som tillämpas på dependensträd. Modellen uppnådde ett F1-värde på 0.76 för den positiva klassen och 0.9 för den negativa klassen, där de positiva datapunkterna utgörs av ett par av pronomen och nominalfras som korefererar, och de negativa datapunkterna utgörs av par som inte korefererar.
19
Holmes, Michael P. "Multi-tree Monte Carlo methods for fast, scalable machine learning." Diss., Georgia Institute of Technology, 2009. http://hdl.handle.net/1853/33865.
Full textAbstract:
As modern applications of machine learning and data mining are forced to deal with ever more massive quantities of data, practitioners quickly run into difficulty with the scalability of even the most basic and fundamental methods. We propose to provide scalability through a marriage between classical, empirical-style Monte Carlo approximation and deterministic multi-tree techniques. This union entails a critical compromise: losing determinism in order to gain speed. In the face of large-scale data, such a compromise is arguably often not only the right but the only choice. We refer to this new approximation methodology as Multi-Tree Monte Carlo. In particular, we have developed the following fast approximation methods:
1. Fast training for kernel conditional density estimation, showing speedups as high as 10⁵ on up to 1 million points.
2. Fast training for general kernel estimators (kernel density estimation, kernel regression, etc.), showing speedups as high as 10⁶ on tens of millions of points.
3. Fast singular value decomposition, showing speedups as high as 10⁵ on matrices containing billions of entries.
The level of acceleration we have shown represents improvement over the prior state of the art by several orders of magnitude. Such improvement entails a qualitative shift, a commoditization, that opens doors to new applications and methods that were previously invisible, outside the realm of practicality. Further, we show how these particular approximation methods can be unified in a Multi-Tree Monte Carlo meta-algorithm which lends itself as scaffolding to the further development of new fast approximation methods. Thus, our contribution includes not just the particular algorithms we have derived but also the Multi-Tree Monte Carlo methodological framework, which we hope will lead to many more fast algorithms that can provide the kind of scalability we have shown here to other important methods from machine learning and related fields.
20
Ježek, Michal. "Podpora snapshotu a rollbacku pro konfigurační soubory v distribuci Fedora." Master's thesis, Vysoké učení technické v Brně. Fakulta informačních technologií, 2008. http://www.nusl.cz/ntk/nusl-235433.
Full textAbstract:
The purpose of this thesis is to design and implement tools for support of a snapshot and a rollback for configuration files on the GNU/Linux distribution. The set of the tools enables an automatic/periodical saving of the configuration files into the selected placement. The creation of backups reacts to file events by watching the changes with kernel subsystem inotify. Tools are enabling to return to the selected backup. The way of the backup actualization is configurable. This tool permits the data comparison from selected backups, to show the differences in configurations and eventually to manage a merge among actual and selected backup. Tools also allows a comparison of a configurations of one client or configurations among clients, and to display the mutual differences, eventually to manage their merge.
21
Costa, Victor Hugo Teles. "Análise de desempenho de sistemas de comunicação OFDM-TDMA utilizando cadeias de Markov e curva de serviço." Universidade Federal de Goiás, 2013. http://repositorio.bc.ufg.br/tede/handle/tede/3795.
Full textAbstract:
Submitted by Jaqueline Silva (jtas29@gmail.com) on 2014-12-12T17:31:05Z
No. of bitstreams: 1
Dissertação-Victor Hugo Teles Costa-2013.pdf: 20678399 bytes, checksum: a39c778934ebe127bd74f506467fe0a3 (MD5)<br>Rejected by Jaqueline Silva (jtas29@gmail.com), reason: on 2014-12-12T17:31:57Z (GMT)<br>Submitted by Jaqueline Silva (jtas29@gmail.com) on 2014-12-12T19:42:30Z
No. of bitstreams: 2
license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5)
Dissertação-Victor Hugo Teles Costa-2013.pdf: 20678399 bytes, checksum: a39c778934ebe127bd74f506467fe0a3 (MD5)<br>Approved for entry into archive by Jaqueline Silva (jtas29@gmail.com) on 2014-12-16T09:25:22Z (GMT) No. of bitstreams: 2
license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5)
Dissertação-Victor Hugo Teles Costa-2013.pdf: 20678399 bytes, checksum: a39c778934ebe127bd74f506467fe0a3 (MD5)<br>Made available in DSpace on 2014-12-16T09:25:22Z (GMT). No. of bitstreams: 2
license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5)
Dissertação-Victor Hugo Teles Costa-2013.pdf: 20678399 bytes, checksum: a39c778934ebe127bd74f506467fe0a3 (MD5)
Previous issue date: 2013-12-06<br>This paper presents a model based on Markov Chains and enhanced with
the use of Kernel Density Estimation and of MMFM (Markov Modulated Fluid
Model) in order to evaluate the performance of the transmission link in OFDMTDMA
systems. For that purpose, traffic models based on the Kernel method and
the MMFM with adjusted autocorrelation function are proposed. From the model
implemented for the OFDM-TDMA system, it was derived equations for estimation
of QoS parameters such as delay and average queue size in the buffer. The
obtained results confirm that the proposed model is efficient in describing the link
performance indicators. The use of MMFM to model the arrival process improves
the QoS parameter estimates of the queueing model making their values very close
to those of the simulation results. It was also developed an equation to the OFDMTDMA
system’s service curve. Through this equation and the concept of Envelope
Process, it was proposed an equation to estimate the probability of buffer overflow
in OFDM-TDMA systems. The results show that the estimates of the overflow probability
based on the system’s service curve are very close to the ones obtained by
simulations and that the computational complexity to obtain them is significantly
reduced compared to the model based on Markov Chains due to the absence of
matrix computation.<br>Este trabalho apresenta um modelo baseado em Cadeias de Markov e aprimorado
com o uso do método de Kernel de estimação não-paramétrica e de MMFM
(Markov Modulated Fluid Model) com o objetivo de avaliar e descrever o desempenho
do enlace de transmissão em sistemas OFDM-TDMA. Para tal, modelos de tráfego
baseados no Método de Kernel e em MMFM com ajuste da função de autocorrelação
são propostos. A partir do modelo implementado para o sistema OFDM-TDMA,
foram obtidas equações para estimação de parâmetros de QoS como retardo e tamanho
médio da fila no buffer. Os resultados obtidos confirmam que o modelo proposto
é bastante eficiente ao descrever os indicadores de desempenho do sistema. O uso de
MMFM para modelar o processo de chegada de pacotes aprimora os estimadores de
parâmetros de QoS tornando as estimativas bem próximas dos valores obtidos com
as simulações. Também deduziu-se uma equação para a curva de serviço de Sistemas
OFDM-TDMA. Em seguida, utilizando-se desta curva de serviço e do conceito de
Processo Envelope foi proposta uma equação para estimação de probabilidade de
transbordo do buffer em sistemas OFDM-TDMA. Os resultados obtidos mostram
que as estimativas de probabilidade de transbordo baseadas na curva de serviço do
sistema se aproximam bem dos resultados da simulação e a complexidade computacional
do cálculo necessário para obtê-los é significativamente reduzida em relação
ao modelo definido utilizando Cadeias de Markov.
22
Abo, Al Ahad George, and Abbas Salami. "Machine Learning for Market Prediction : Soft Margin Classifiers for Predicting the Sign of Return on Financial Assets." Thesis, Linköpings universitet, Produktionsekonomi, 2018. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-151459.
Full textAbstract:
Forecasting procedures have found applications in a wide variety of areas within finance and have further shown to be one of the most challenging areas of finance. Having an immense variety of economic data, stakeholders aim to understand the current and future state of the market. Since it is hard for a human to make sense out of large amounts of data, different modeling techniques have been applied to extract useful information from financial databases, where machine learning techniques are among the most recent modeling techniques. Binary classifiers such as Support Vector Machines (SVMs) have to some extent been used for this purpose where extensions of the algorithm have been developed with increased prediction performance as the main goal. The objective of this study has been to develop a process for improving the performance when predicting the sign of return of financial time series with soft margin classifiers. An analysis regarding the algorithms is presented in this study followed by a description of the methodology that has been utilized. The developed process containing some of the presented soft margin classifiers, and other aspects of kernel methods such as Multiple Kernel Learning have shown pleasant results over the long term, in which the capability of capturing different market conditions have been shown to improve with the incorporation of different models and kernels, instead of only a single one. However, the results are mostly congruent with earlier studies in this field. Furthermore, two research questions have been answered where the complexity regarding the kernel functions that are used by the SVM have been studied and the robustness of the process as a whole. Complexity refers to achieving more complex feature maps through combining kernels by either adding, multiplying or functionally transforming them. It is not concluded that an increased complexity leads to a consistent improvement, however, the combined kernel function is superior during some of the periods of the time series used in this thesis for the individual models. The robustness has been investigated for different signal-to-noise ratio where it has been observed that windows with previously poor performance are more exposed to noise impact.
23
Haglund, Robin. "Automated analysis of battery articles." Thesis, Uppsala universitet, Strukturkemi, 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-403738.
Full textAbstract:
Journal articles are the formal medium for the communication of results among scientists, and often contain valuable data. However, manually collecting article data from a large field like lithium-ion battery chemistry is tedious and time consuming, which is an obstacle when searching for statistical trends and correlations to inform research decisions. To address this a platform for the automatic retrieval and analysis of large numbers of articles is created and applied to the field of lithium-ion battery chemistry. Example data produced by the platform is presented and evaluated and sources of error limiting this type of platform are identified, with problems related to text extraction and pattern matching being especially significant. Some solutions to these problems are presented and potential future improvements are proposed.
24
Zhang, Hang. "Distributed Support Vector Machine With Graphics Processing Units." ScholarWorks@UNO, 2009. http://scholarworks.uno.edu/td/991.
Full textAbstract:
Training a Support Vector Machine (SVM) requires the solution of a very large quadratic programming (QP) optimization problem. Sequential Minimal Optimization (SMO) is a decomposition-based algorithm which breaks this large QP problem into a series of smallest possible QP problems. However, it still costs O(n2) computation time. In our SVM implementation, we can do training with huge data sets in a distributed manner (by breaking the dataset into chunks, then using Message Passing Interface (MPI) to distribute each chunk to a different machine and processing SVM training within each chunk). In addition, we moved the kernel calculation part in SVM classification to a graphics processing unit (GPU) which has zero scheduling overhead to create concurrent threads. In this thesis, we will take advantage of this GPU architecture to improve the classification performance of SVM.
25
Tsuda, Larissa Sayuri. "Análise dos atropelamentos de mamíferos em uma rodovia no estado de São Paulo utilizando Self-Organizing Maps." Universidade de São Paulo, 2018. http://www.teses.usp.br/teses/disponiveis/3/3138/tde-21092018-134154/.
Full textAbstract:
A construção e ampliação de rodovias gera impactos significativos ao meio ambiente. Os principais impactos ao meio biótico são a supressão de vegetação, redução da riqueza e abundância de espécies de fauna como decorrência da fragmentação de habitats e aumento dos riscos de atropelamento de animais silvestres e domésticos. O objetivo geral do trabalho foi identificar padrões espaciais nos atropelamentos de fauna silvestre por espécie (nome popular) utilizando ferramentas de análise espacial e machine learning. Especificamente, buscou-se compreender a relação entre atropelamentos de animais silvestres e variáveis que representam características de uso e cobertura do solo e caracterização da rodovia, tais como formação florestal, corpos d\'água, silvicultura, áreas edificadas, velocidade máxima permitida, volume de tráfego, entre outras. Os atropelamentos de fauna silvestre foram analisados por espécie atropelada, a fim de identificar os padrões espaciais dos atropelamentos específicos para cada espécie. As ferramentas de análise espacial empregadas foram a Função K - para determinar o padrão de distribuição dos registros de atropelamento de fauna, o Estimador de Densidade de Kernel - para gerar estimativas de densidade de pontos sobre a rodovia, a Análise de Hotspots - para identificar os trechos mais críticos de atropelamento de fauna e, por fim, o Self-Organizing Maps (SOM), um tipo de rede neural artificial, que reorganiza amostras de dados n-dimensionais de acordo com a similaridade entre elas. Os resultados das análises de padrões pontuais foram importantes para entender que os pontos de atropelamento possuem padrões de distribuição espacial que variam por espécie. Os eventos ocorrem espacialmente agrupados e não estão homogeneamente distribuídos ao longo da rodovia. De maneira geral, os animais apresentam trechos de maior intensidade de atropelamento em locais distintos. O SOM permitiu analisar as relações entre múltiplas variáveis, lineares e não-lineares, tais como são os dados ecológicos, e encontrar padrões espaciais distintos por espécie. A maior parte dos animais foi atropelada próxima de fragmentos florestais e de corpos d\'água, e distante de cultivo de cana-de-açúcar, silvicultura e área edificada. Porém, uma parte considerável das mortes de animais dos tipos com maior número de atropelamentos ocorreu em áreas com paisagem diversificada, incluindo alta densidade de drenagem, fragmentos florestais, silvicultura e áreas edificadas.<br>The construction and expansion of roads cause significant impacts on the environment. The main potential impacts to biotic environment are vegetation suppression, reduction of the abundance and richness of species due to forest fragmentation and increase of animal (domestic and wildlife) vehicle collisions. The general objective of this work was to identify spatial patterns in wildlife-vehicle collisions individually per species by using spatial analysis and machine learning. Specifically, the relationship between wildlife-vehicle collisions and variables that represent land use and road characterization features - such as forests, water bodies, silviculture, sugarcane fields, built environment, speed limit and traffic volume - was investigated. The wildlife-vehicle collisions were analyzed per species, in order to identify the spatial patterns for each species separately. The spatial analysis tools used in this study were K-Function - to determine the distribution pattern of roadkill, Kernel Density Estimator (KDE) - to identify the location and intensity of hotspots and hotzones. Self-Organizing Maps (SOM), an artificial neural network (ANN), was selected to reorganize the multi-dimensional data according to the similarity between them. The results of the spatial pattern analysis were important to perceive that the point data pattern varies between species. The events occur spatially clustered and are not uniformly distributed along the highway. In general, wildlife-vehicle collsions have their hotzones in different locations. SOM was able to analyze the relationship between multiple variables, linear and non-linear, such as ecological data, and established distinct spatial patterns per each species. Most of the wildlife was run over close to forest area and water bodies, and distant from sugarcane, silviculture and built environments. But a considerable part of the wildlife-vehicle collisions occurred in areas with diverse landscape, including high density of water bodies, silviculture and built environments.
26
Abdallah, Fahed. "Noyaux reproduisants et critères de contraste pour l'élaboration de détecteurs à structure imposée." Troyes, 2004. http://www.theses.fr/2004TROY0002.
Full textAbstract:
Les travaux réalisés pendant cette thèse sont relatifs à la synthèse de détecteurs à partir d'une base d'exemples étiquetés. La théorie développée fait appel aux espaces de Hilbert à noyaux reproduisants pour l'élaboration de détecteurs linéaires généralisés dans des espaces transformés de dimension importante, voire infinie, sans qu'aucun calcul n'y soit effectué explicitement. Elle repose sur l'optimisation du meilleur critère de contraste pour le problème traité, après s'être assuré que de telles mesures de performance permettant l'obtention sous des conditions restrictives assez faibles, à une statistique équivalente au rapport de vraisemblance. Pour une meilleure prise en compte de phénomènes tels que la malédiction de la dimensionnalité, l'approche proposée s'appuie sur la théorie de l'apprentissage. Celle-ci lui permet d'offrir des garanties de performances en généralisation. On propose ainsi des méthodes qui permettent le contrôle de complexité des détecteurs obtenus. Les résultats obtenus sur des données synthétiques et réelles montrent que notre approche est en mesure de rivaliser avec les structures de décision les plus étudiées actuellement que sont les Support Vector Machines<br>In this thesis, we consider statistical learning machines with try to infer rules from a given set or observations in order to make correct predictions on unseen examples. Building upon the theory of reproducing kernels, we develop a generalized linear detector in transformed spaces of high dimension, without explicitly doing any calculus in these spaces. The method is based on the optimization of the best second-order criterion with respect to the problem to solve. In fact, theoretical results show that second-order criteria are able, under some mild conditions, to guarantee the best solution in the sense of classical detection theories. Achieving a good generalisation performance with a receiver requires matching its complexity to the amount of available training data. This problem, known as the curse of dimensionality, has been studied theoretically by Vapnik and Chervonenkis. In this dissertation, we propose complexity control procedures in order to improve the performance of these receivers when few training data are available. Simulation results on real and synthetic data show clearly the competitiveness of our approach compared with other state of the art existing kernel methods like Support Vector Machines
27
Popel, Aleksej. "The effect of radiation damage by fission fragments on the structural stability and dissolution of the UO2 fuel matrix." Thesis, University of Cambridge, 2017. https://www.repository.cam.ac.uk/handle/1810/265103.
Full textAbstract:
The aim of this work was to study the separate effect of fission fragment damage on the structural integrity and matrix dissolution of uranium dioxide in water. Radiation damage similar to fission damage was created by irradiating bulk undoped and doped ‘SIMFUEL’ disks of UO2, undoped bulk CeO2 and thin films of UO2 and CeO2 with high energy Xe and U ions. The UO2 thin films, with thicknesses in the range of 90 – 150 nm, were deposited onto (001), (110) and (111) orientations of single crystal LSAT (Al10La3O51Sr14Ta7) and YSZ (Yttria-Stabilised Zirconia) substrates. The CeO2 thin films were deposited onto single crystal silicon (001) substrates. Part of the bulk UO2 and CeO2 samples, the thin films of UO2 on the LSAT substrates and the thin films of CeO2 were irradiated with 92 MeV 129Xe23+ ions to a fluence of 4.8 × 1015 ions/cm2 to simulate the damage produced by fission fragments in uranium dioxide nuclear fuel. Part of the bulk UO2 and CeO2 samples and the thin films of UO2 on the YSZ substrates were irradiated with 110 MeV 238U31+ ions to a fluence of 5 × 1010, 5 × 1011 and 5 × 1012 ions/cm2 to study the accumulation of the damage induced. The irradiated and unirradiated samples were studied using scanning electron microscopy (SEM), focused ion beam (FIB), atomic force microscopy (AFM), energy dispersive X-ray (EDX) spectroscopy, electron probe microanalysis (EPMA), X-ray diffraction (XRD), electron backscatter diffraction (EBSD), secondary ion mass spectrometry (SIMS) and X-ray photoelectron spectroscopy (XPS) techniques to characterise the as-produced samples and assess the effects of the ion irradiations. Dissolution experiments were conducted to assess the effect of the Xe ion irradiation on the dissolution of the thin film UO2 samples on the LSAT substrates and the bulk and thin film CeO2 samples. The solutions obtained from the leaching of the irradiated and unirradiated samples were analysed using inductively coupled plasma mass spectrometry (ICP-MS). XRD studies of the bulk UO2 samples showed that the ion irradiations resulted in an increased lattice parameter, microstrain and decreased crystallite size, as expected. The irradiated UO2 thin films on the LSAT substrates underwent significant microstructural and crystallographic rearrangements. It was shown that by irradiating thin films of UO2 with high energy, high fluence ions, it is possible to produce a structure that is similar to a thin slice through the high burn-up structure. It is expected that the ion irradiation induced chemical mixing of the UO2 films with the substrate elements (La, Sr, Al, Ta). As a result, a material similar to a doped SIMFUEL with induced radiation damage was produced.
28
Nguyen, Van Toi. "Visual interpretation of hand postures for human-machine interaction." Thesis, La Rochelle, 2015. http://www.theses.fr/2015LAROS035/document.
Full textAbstract:
Aujourd'hui, les utilisateurs souhaitent interagir plus naturellement avec les systèmes numériques. L'une des modalités de communication la plus naturelle pour l'homme est le geste de la main. Parmi les différentes approches que nous pouvons trouver dans la littérature, celle basée sur la vision est étudiée par de nombreux chercheurs car elle ne demande pas de porter de dispositif complémentaire. Pour que la machine puisse comprendre les gestes à partir des images RGB, la reconnaissance automatique de ces gestes est l'un des problèmes clés. Cependant, cette approche présente encore de multiples défis tels que le changement de point de vue, les différences d'éclairage, les problèmes de complexité ou de changement d'environnement. Cette thèse propose un système de reconnaissance de gestes statiques qui se compose de deux phases : la détection et la reconnaissance du geste lui-même. Dans l'étape de détection, nous utilisons un processus de détection d'objets de Viola Jones avec une caractérisation basée sur des caractéristiques internes d'Haar-like et un classifieur en cascade AdaBoost. Pour éviter l'influence du fond, nous avons introduit de nouvelles caractéristiques internes d'Haar-like. Ceci augmente de façon significative le taux de détection de la main par rapport à l'algorithme original. Pour la reconnaissance du geste, nous avons proposé une représentation de la main basée sur un noyau descripteur KDES (Kernel Descriptor) très efficace pour la classification d'objets. Cependant, ce descripteur n'est pas robuste au changement d'échelle et n'est pas invariant à l'orientation. Nous avons alors proposé trois améliorations pour surmonter ces problèmes : i) une normalisation de caractéristiques au niveau pixel pour qu'elles soient invariantes à la rotation ; ii) une génération adaptative de caractéristiques afin qu'elles soient robustes au changement d'échelle ; iii) une construction spatiale spécifique à la structure de la main au niveau image. Sur la base de ces améliorations, la méthode proposée obtient de meilleurs résultats par rapport au KDES initial et aux descripteurs existants. L'intégration de ces deux méthodes dans une application montre en situation réelle l'efficacité, l'utilité et la faisabilité de déployer un tel système pour l'interaction homme-robot utilisant les gestes de la main<br>Nowadays, people want to interact with machines more naturally. One of the powerful communication channels is hand gesture. Vision-based approach has involved many researchers because this approach does not require any extra device. One of the key problems we need to resolve is hand posture recognition on RGB images because it can be used directly or integrated into a multi-cues hand gesture recognition. The main challenges of this problem are illumination differences, cluttered background, background changes, high intra-class variation, and high inter-class similarity. This thesis proposes a hand posture recognition system consists two phases that are hand detection and hand posture recognition. In hand detection step, we employed Viola-Jones detector with proposed concept Internal Haar-like feature. The proposed hand detection works in real-time within frames captured from real complex environments and avoids unexpected effects of background. The proposed detector outperforms original Viola-Jones detector using traditional Haar-like feature. In hand posture recognition step, we proposed a new hand representation based on a good generic descriptor that is kernel descriptor (KDES). When applying KDES into hand posture recognition, we proposed three improvements to make it more robust that are adaptive patch, normalization of gradient orientation in patches, and hand pyramid structure. The improvements make KDES invariant to scale change, patch-level feature invariant to rotation, and final hand representation suitable to hand structure. Based on these improvements, the proposed method obtains better results than original KDES and a state of the art method
29
楊健炘. "Kernel-Based SVM: Theory and Application." Thesis, 2006. http://ndltd.ncl.edu.tw/handle/96957635162943930250.
Full text
30
Chen, Yen-Ting, and 陳衍廷. "Face Detection Using Kernel PCA and Imbalanced SVM." Thesis, 2006. http://ndltd.ncl.edu.tw/handle/04609871515301078957.
Full textAbstract:
碩士<br>中原大學<br>機械工程研究所<br>94<br>Abstract
Fully automatic face and expression recognition systems have received more and more attention in recent years. However, in such kind of systems, the first step is to detect faces from images and the recognition rate extremely relies on the accuracy of face detection process, therefore, face detection system plays a crucial role in a face recognition system. This thesis aims to develop a multi-view face detection system. Our system is composed of two sub-systems：(1) one is for detecting faces from still images, (2) and the other is for detecting faces from image sequences. In order to solve the problem of imbalanced dataset, we apply imbalanced SVM (ISVM) and KPCA to data sets in order to extract image features. The experimental results show that combing the KPCA feature extractor and the ISVM classifier can yield higher face-detecting rate compared to SVM. Before detection in still image mode, the method of skin-region segmentation is utilized to find the possible regions, while both moving objects tracking technique and skin-region segmentation method are applied to eliminate the redundant searching regions in dynamic sequences mode. The system also aggregates with CAMSHIFT algorithm for fast face tracking. The experimental results show that the proposed multi-view face detection system can detect and track faces varying in orientations, sizes, and expressions in images efficiently.
Keywords: Face recognition, face detection, support vector machines, kernel principal components analysis (KPCA), imbalanced dataset
31
Ruei-YaoHuang and 黃瑞堯. "Video and Image Applications Based on Kernel Support Vector Machine (SVM)." Thesis, 2010. http://ndltd.ncl.edu.tw/handle/67090107280042729427.
Full textAbstract:
碩士<br>國立成功大學<br>電機工程學系碩博士班<br>98<br>We use a classification method based on Kernel support vector machines (Kernel SVM), that can be applied to various types of data. We use Kernel SVM to extract the video highlights of sport and classify textile grade. Different form original classification method, we optimize the parameters and the features by Genetic Algorithm. The Kernel SVM is composed of the training mode and the analysis mode. In the training mode, we adopt the Kernel SVM to train classification function. In the analysis mode, we use the classification function to generate the classification result. We use the video and audio features without predefining any highlight rule of the events. The precision of highlight extraction by Kernel SVM can achieve about 81%, while that of textile grade classification is approximately 83% The experimental results show the proposed method can extract video highlights of sport, and it can also be applied to textile grade classification.
32
Lee, Jen-Hao, and 李仁豪. "Model Selection of the Bounded SVM Formulation Using the RBF Kernel." Thesis, 2001. http://ndltd.ncl.edu.tw/handle/98455433223164841493.
Full textAbstract:
碩士<br>國立臺灣大學<br>資訊工程學研究所<br>89<br>The support vector machine (SVM) has become one of the most promising and
popular methods in machine learning. Sound theory and careful implementation
make SVM efficient enough to solve moderate to large problems, and
the performance has been shown to be competitive with existing methods such as
neural networks and decision trees.
One remaining problem on the practical use of SVM is the model selection.
That is, there are several parameters to tune
so that better general accuracy can be achieved.
This thesis works on the case of
an SVM classification formulation with only bounded
constraints using the RBF kernel
according to the leave-one-out (loo) rate. A simple framework
is approached in the first place, in which loo rates are exactly computed
through a given model space. Next, some tricks are utilized
to avoid unnecessary computation. Some heuristics are also proposed for
locating good areas in more efficient ways according to observations on
loo rates and time distribution over the model space.
The experiments show that the software developed here performs
well both in terms of computational time and loo rates. And the heuristics
proposed here should be helpful for other SVM model selection software.
33
Singh, Shyam Kishor. "Brain Tumor Detection and Segmentaion in MR Images Using Kernel SVM." Thesis, 2017. http://ethesis.nitrkl.ac.in/8805/1/2017_MT_SK_Singh.pdf.
Full textAbstract:
Automatic detection of brain tumors and accurate classification of magnetic resonance image (MRI) of brain tumors has been a perplexed technique to obtain variability and complexity of the position, texture, size, and outlining of the lesions. In our method, we presented a novel approach to distinguish the normal and abnormal brain tumors in MR brain image. We implemented the technique over 200 brain MRIs in which 160 are abnormal and 40 are normal brain MRIs. Abnormal MRI is of two types cancerous or non- cancerous MRI. In this method first we applying some pre-processing technique and DWT technique after that we used the principal component analysis (PCA) followed by training and testing the brain MRIs using machine learning technique kernel support vector machine (KSVM). Kernel- SVM gives a good classification result in overall classification techniques of machine learning. We compares the linear and Non- linear and the RGB kernel support vector machines accuracy for classification of brain tumor.
34
Chow, Yi-Zheng, and 周詒徵. "The Kernel of Constant-sum Simple Game." Thesis, 1999. http://ndltd.ncl.edu.tw/handle/24658603945444865069.
Full text
35
Melnyk, Igor V. "Empirical investigation of models produced by kernel LARS-type and SVM-type algorithms." Thesis, 2009. http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqdiss&rft_dat=xri:pqdiss:1464520.
Full text
36
Asharaf, S. "Efficient Kernel Methods For Large Scale Classification." Thesis, 2007. https://etd.iisc.ac.in/handle/2005/1076.
Full textAbstract:
Classification algorithms have been widely used in many application domains. Most of these domains deal with massive collection of data and hence demand classification algorithms that scale well with the size of the data sets involved. A classification algorithm is said to be scalable if there is no significant increase in time and space requirements for the algorithm (without compromising the generalization performance) when dealing with an increase in the training set size. Support Vector Machine (SVM) is one of the most celebrated kernel based classification methods used in Machine Learning. An SVM capable of handling large scale classification problems will definitely be an ideal candidate in many real world applications. The training process involved in SVM classifier is usually formulated as a Quadratic Programing(QP) problem. The existing solution strategies for this problem have an associated time and space complexity that is (at least) quadratic in the number of training points. This makes the SVM training very expensive even on classification problems having a few thousands of training examples.
This thesis addresses the scalability of the training algorithms involved in both two class and multiclass Support Vector Machines. Efficient training schemes reducing the space and time requirements of the SVM training process are proposed as possible solutions. The classification schemes discussed in the thesis for handling large scale two class classification problems are a) Two selective sampling based training schemes for scaling Non-linear SVM and b) Clustering based approaches for handling unbalanced data sets with Core Vector Machine. To handle large scale multicalss classification problems, the thesis proposes Multiclass Core Vector Machine (MCVM), a scalable SVM based multiclass classifier. In MVCM, the multiclass SVM problem is shown to be equivalent to a Minimum Enclosing Ball (MEB) problem and is then solved using a fast approximate MEB finding algorithm. Experimental studies were done with several large real world data sets such as IJCNN1 and Acoustic data sets from LIBSVM page, Extended USPS data set from CVM page and network intrusion detection data sets of DARPA, US Defense used in KDD 99 contest. From the empirical results it is observed that the proposed classification schemes achieve good generalization performance at low time and space requirements. Further, the scalability experiments done with large training data sets have demonstrated that the proposed schemes scale well. A novel soft clustering scheme called Rough Support Vector Clustering (RSVC) employing the idea of Soft Minimum Enclosing Ball Problem (SMEB) is another contribution discussed in this thesis. Experiments done with a synthetic data set and the real world data set namely IRIS, have shown that RSVC finds meaningful soft cluster abstractions.
37
Asharaf, S. "Efficient Kernel Methods For Large Scale Classification." Thesis, 2007. http://hdl.handle.net/2005/1076.
Full textAbstract:
Classification algorithms have been widely used in many application domains. Most of these domains deal with massive collection of data and hence demand classification algorithms that scale well with the size of the data sets involved. A classification algorithm is said to be scalable if there is no significant increase in time and space requirements for the algorithm (without compromising the generalization performance) when dealing with an increase in the training set size. Support Vector Machine (SVM) is one of the most celebrated kernel based classification methods used in Machine Learning. An SVM capable of handling large scale classification problems will definitely be an ideal candidate in many real world applications. The training process involved in SVM classifier is usually formulated as a Quadratic Programing(QP) problem. The existing solution strategies for this problem have an associated time and space complexity that is (at least) quadratic in the number of training points. This makes the SVM training very expensive even on classification problems having a few thousands of training examples.
This thesis addresses the scalability of the training algorithms involved in both two class and multiclass Support Vector Machines. Efficient training schemes reducing the space and time requirements of the SVM training process are proposed as possible solutions. The classification schemes discussed in the thesis for handling large scale two class classification problems are a) Two selective sampling based training schemes for scaling Non-linear SVM and b) Clustering based approaches for handling unbalanced data sets with Core Vector Machine. To handle large scale multicalss classification problems, the thesis proposes Multiclass Core Vector Machine (MCVM), a scalable SVM based multiclass classifier. In MVCM, the multiclass SVM problem is shown to be equivalent to a Minimum Enclosing Ball (MEB) problem and is then solved using a fast approximate MEB finding algorithm. Experimental studies were done with several large real world data sets such as IJCNN1 and Acoustic data sets from LIBSVM page, Extended USPS data set from CVM page and network intrusion detection data sets of DARPA, US Defense used in KDD 99 contest. From the empirical results it is observed that the proposed classification schemes achieve good generalization performance at low time and space requirements. Further, the scalability experiments done with large training data sets have demonstrated that the proposed schemes scale well. A novel soft clustering scheme called Rough Support Vector Clustering (RSVC) employing the idea of Soft Minimum Enclosing Ball Problem (SMEB) is another contribution discussed in this thesis. Experiments done with a synthetic data set and the real world data set namely IRIS, have shown that RSVC finds meaningful soft cluster abstractions.
38
Zhang, Zhanyang. "Customizing kernels in Support Vector Machines." Thesis, 2007. http://hdl.handle.net/10012/3063.
Full textAbstract:
Support Vector Machines have been used to do classification and
regression analysis. One important part of SVMs are the kernels.
Although there are several widely used kernel functions, a carefully
designed kernel will help to improve the accuracy of SVMs. We
present two methods in terms of customizing kernels: one is
combining existed kernels as new kernels, the other one is to do feature selection.
We did theoretical analysis in the interpretation of
feature spaces of combined kernels. Further an experiment on a
chemical data set showed improvements of a linear-Gaussian combined
kernel over single kernels. Though the improvements are not
universal, we present a new idea of creating kernels in SVMs.
39
Chiang, Tse-Yu, and 江則佑. "Abbe-SVD: Compact Abbe’s Kernel Generation for Microlithography Aerial Image Simulation using Singular-Value Decomposition Method." Thesis, 2008. http://ndltd.ncl.edu.tw/handle/17771551777073303525.
Full textAbstract:
碩士<br>國立臺灣大學<br>電子工程學研究所<br>96<br>At the present days, the key and critical part of industrial IC manufacture is the optical lithography technology which can duplicate the design patterns on the mask onto wafer by light exposure. However, when the mask patterns are too small which are approaching light wave length, the image quality and resolution on the wafer are getting worse owing to diffraction effect. Therefore, some necessary resolution enhancement techniques are proposed with remarkable skills and algorithms, and need to be verified by simulations or experimental results. The most direct and accurate simulation is the imaging of patterns on wafer. Accurate imaging simulation can show exposed and unexposed regions after photolithography by computer. Existing commercial and academic OPC simulators which compute in frequency domain with Abbe''s method applied on partial coherent light source take several days for computation with hundreds of computers working together at the same time. Hence, we propose to generate a compact Abbe''s kernel for microlithography aerial image simulation using singular-value decomposition method. The advantages of this approach are as follows: First, since not all the Abbe''s kernels have critical effects on aerial image, we can eliminate them to generate a compact one with SVD. Therefore, we can speed up simulation time, and furthermore keep the accuracy user specified. Second, with advanced concentric circles source discretization, equivalent kernels with higher precision is produced. Finally, we can use compact Abbe''s kernel to build LUT to speed up simulation time.
In this thesis, we introduce some basic knowledge of optical lithography in chapter 1 and some coherent light in optics with analytical solution in chapter 2. Then, partially coherent light concept, advanced illumination aperture and Abbe''s method are introduced in chapter 3 and our Abbe-SVD algorithm and advanced source discretization will also be derived. Experimental result and some comparisons will be shown in chapter 4 and finally conclusion will be made in chapter 5.
40
Chiang, Tse-Yu. "Abbe-SVD: Compact Abbe's Kernel Generation for Microlithography Aerial Image Simulation using Singular-Value Decomposition Method." 2008. http://www.cetd.com.tw/ec/thesisdetail.aspx?etdun=U0001-2306200804053500.
Full text
41
Kunc, Vladimír. "Comparison of different models for forecasting of Czech electricity market." Master's thesis, 2017. http://www.nusl.cz/ntk/nusl-367836.
Full textAbstract:
There is a demand for decision support tools that can model the electricity markets and allows to forecast the hourly electricity price. Many different ap- proach such as artificial neural network or support vector regression are used in the literature. This thesis provides comparison of several different estima- tors under one settings using available data from Czech electricity market. The resulting comparison of over 5000 different estimators led to a selection of several best performing models. The role of historical weather data (temper- ature, dew point and humidity) is also assesed within the comparison and it was found that while the inclusion of weather data might lead to overfitting, it is beneficial under the right circumstances. The best performing approach was the Lasso regression estimated using modified Lars. 1
42
(11393945), Shaben Kayamboo. "Proactive Fault Detection using Machine Learning to Aid Predictive Maintenance in Photovoltaic Systems." Thesis, 2024. https://figshare.com/articles/thesis/Proactive_Fault_Detection_using_Machine_Learning_to_Aid_Predictive_Maintenance_in_Photovoltaic_Systems/26548420.
Full textAbstract:
In recent history, photovoltaic (PV) systems as a means of energy generation have risen in popularity due to the world’s decreasing reliance on fossil fuels and a stronger focus on combating the adverse effects of climate change. While PV systems have immense potential, their vulnerability to faults substantially threatens their efficiency and reliability, potentially reducing their positive impact on the environment and the world economy. Current PV system maintenance strategies are either reactive or preventive, with a limited focus on predictive methods that leverage advanced machine learning models for fault detection. This thesis addresses this research gap, focusing on the development and optimisation of machine learning algorithms for proactive PV system fault detection. This is accomplished through the analysis of various PV system data parameters such as voltage, current, power, energy delivered or received, performance ratio, and meteorological data, among others.
This research investigation started with a data collection process from the Desert Knowledge Australia Solar Centre (DKASC), a facility dedicated to solar energy research. After collecting data from 10 of the most fault-prone sites, rigorous pre-processing steps, including cleaning, transforming, and balancing, were employed. Particular attention was given to inverter failures and inverter intermittent issues, as they were identified as the most common faults, significantly influencing PV system performance and reliability. A variety of machine learning algorithms were employed, including deep learning methods such as Artificial Neural Networks and Recurrent Neutral Networks. However, Kernel SVM and K Nearest Neighbours were found to be most effective in predicting the specific individual faults, inverter failures and inverter intermittent issues, respectively. Subsequent parameter optimisation efforts, including adjusting fault occurrence window sizes, running summary days, classifier hyperparameters, and validation methods, enabled differentiation between those two fault types in a combined faults dataset using the K Nearest Neighbours model.
This research project makes two novel contributions to the field. First, it developed an adaptive method for predicting specific faults in PV systems. Second, through a parameter optimisation process, this research created an adaptive method for differentiation between two specific faults. Through these adaptive fault prediction methods, the most effective machine learning model can be selected to predict any particular fault or differentiate between any specific faults, enhancing their real-world utility and impact.
The findings from this research have considerable implications for future work in this domain. They serve as a guide for further research and development efforts to inform predictive maintenance strategies for PV systems. Future directions include the investigation of other types of faults, expanding the dataset to include more diverse fault scenarios, exploring advanced feature engineering and selection methods, integrating the developed fault prediction models with practical maintenance scheduling systems, and assessing the economic impacts of these models on the efficiency and cost-effectiveness of PV systems.
In summary, while this research does contribute to improving the reliability and efficiency of PV systems through enhanced fault prediction, it also provides direction for further research into developing robust predictive maintenance strategies. The findings of this research support the broader goal of making renewable energy more reliable, efficient, and cost-effective in pursuit of a more sustainable energy-driven future.
43
Vasilescu, M. Alex O. "A Multilinear (Tensor) Algebraic Framework for Computer Graphics, Computer Vision and Machine Learning." Thesis, 2012. http://hdl.handle.net/1807/65327.
Full textAbstract:
This thesis introduces a multilinear algebraic framework for computer graphics, computer vision, and machine learning, particularly for the fundamental purposes of image synthesis, analysis, and recognition. Natural images result from the multifactor interaction between the imaging process, the scene illumination, and the scene geometry. We assert that a principled mathematical approach to disentangling and explicitly representing these causal factors, which are essential to image formation, is through numerical multilinear algebra, the algebra of higher-order tensors.
Our new image modeling framework is based on(i) a multilinear generalization of principal components analysis (PCA), (ii) a novel multilinear generalization of independent components analysis (ICA), and (iii) a multilinear projection for use in recognition that maps images to the multiple causal factor spaces associated with their formation. Multilinear PCA employs a tensor extension of the conventional matrix singular value decomposition (SVD), known as the M-mode SVD, while our multilinear ICA method involves an analogous M-mode ICA algorithm.
As applications of our tensor framework, we tackle important problems in computer graphics, computer vision, and pattern recognition; in particular, (i) image-based rendering, specifically introducing the multilinear synthesis of images of textured surfaces under varying view and illumination conditions, a new technique that we call
``TensorTextures'', as well as (ii) the multilinear analysis and recognition of facial images under variable face shape, view, and illumination conditions, a new technique that we call ``TensorFaces''. In developing these applications, we introduce a multilinear image-based rendering algorithm and a multilinear appearance-based recognition algorithm. As a final, non-image-based application of our framework, we consider the analysis, synthesis and recognition of human motion data using multilinear methods, introducing a new technique that we call ``Human Motion Signatures''.
44
Ghous, Hamid. "Building a robust clinical diagnosis support system for childhood cancer using data mining methods." Thesis, 2016. http://hdl.handle.net/10453/90061.
Full textAbstract:
University of Technology Sydney. Faculty of Engineering and Information Technology.<br>Progress in understanding core pathways and processes of cancer requires thorough analysis of many coding and noncoding regions of the genome. Data mining and knowledge discovery have been applied to datasets across many industries, including bioinformatics. However, data mining faces a major challenge in its application to bioinformatics: the diversity and dimensionality of biomedical data. The term ‘big data’ was applied to the clinical domain by Yoo et al. (2014), specifically referring to single nucleotide polymorphism (SNP) and gene expression data. This research thesis focuses on three different types of data: gene-annotations, gene expression and single nucleotide polymorphisms.
Genetic association studies have led to the discovery of single genetic variants associated with common diseases. However, complex diseases are not caused by a single gene acting alone but are the result of complex linear and non-linear interactions among different types of microarray data. In this scenario, a single gene can have a small effect on disease but cannot be the major cause of the disease. For this reason there is a critical need to implement new approaches which take into account linear and non-linear gene-gene and patient-patient interactions that can eventually help in diagnosis and prognosis of complex diseases. Several computational methods have been developed to deal with gene annotations, gene expressions and SNP data of complex diseases. However, analysis of every gene expression and SNP profile, and finding gene-to-gene relationships, is computationally infeasible because of the high-dimensionality of data. In addition, many computational methods have problems with scaling to large datasets, and with overfitting. Therefore, there is growing interest in applying data mining and machine learning approaches to understand different types of microarray data.
Cancer is the disease that kills the most children in Australia (Torre et al., 2015). Within this thesis, the focus is on childhood Acute Lymphoblastic Leukaemia. Acute Lymphoblastic Leukaemia is the most common childhood malignancy with 24% of all new cancers occurring in children within Australia (Coates et al., 2001). According to the American Cancer Society (2016), a total of 6,590 cases of ALL have been diagnosed across all age groups in USA and the expected deaths are 1,430 in 2016.
The project uses different data mining and visualisation methods applied on different types of biological data: gene annotations, gene expression and SNPs.
This thesis focuses on three main issues in genomic and transcriptomic data studies:
(i) Proposing, implementing and evaluating a novel framework to find functional relationships between genes from gene-annotation data.
(ii) Identifying an optimal dimensionality reduction method to classify between relapsed and non-relapsed ALL patients using gene expression.
(iii) Proposing, implementing and evaluating a novel feature selection approach to identify related metabolic pathways in ALL
This thesis proposes, implements and validates an efficient framework to find functional relationships between genes based on gene-annotation data. The framework is built on a binary matrix and a proximity matrix, where the binary matrix contains information related to genes and their functionality, while the proximity matrix shows similarity between different features. The framework retrieves gene functionality information from Gene Ontology (GO), a publicly available database, and visualises the functional related genes using singular value decomposition (SVD). From a simple list of gene-annotations, this thesis retrieves features (i.e Gene Ontology terms) related to each gene and calculates a similarity measure based on the distance between terms in the GO hierarchy. The distance measures are based on hierarchical structure of Gene Ontology and these distance measures are called similarity measures. In this framework, two different similarity measures are applied:
(i) A hop-based similarity measure where the distance is calculated based on the number of links between two terms.
(ii) An information-content similarity measure where the similarity between terms is based on the probability of GO terms in the gene dataset.
This framework also identifies which method performs better among these two similarity measures at identifying functional relationships between genes. Singular value decomposition method is used for visualisation, having the advantage that multiple types of relationships can be visualised simultaneously (gene-to-gene, term-to-term and gene-to-term)
In this thesis a novel framework is developed for visualizing patient-to-patient relationships using gene expression values. The framework builds on the random forest feature selection method to filter gene expression values and then applies different linear and non-linear machine learning methods to them. The methods used in this framework are Principal Component Analysis (PCA), Kernel Principal Component Analysis (kPCA), Local Linear Embedding (LLE), Stochastic Neighbour Embedding (SNE) and Diffusion Maps. The framework compares these different machine learning methods by tuning different parameters to find the optimal method among them. Area under the curve (AUC) is used to rank the results and SVM is used to classify between relapsed and non-relapsed patients.
The final section of the thesis proposes, implements and validates a framework to find active metabolic pathways in ALL using single nucleotide polymorphism (SNP) profiles. The framework is based on the random forest feature selection method. A collected dataset of ALL patient and healthy controls is constructed and later random forest is applied using different parameters to find highly-ranked SNPs. The credibility of the model is assessed based on the error rate of the confusion matrix and kappa values. Selected high ranked SNPs are used to retrieve metabolic pathways related to ALL from the KEGG metabolic pathways database.
The methodologies and approaches presented in this thesis emphasise the critical role that different types of microarray data play in understanding complex diseases like ALL. The availability of flexible frameworks for the task of disease diagnosis and prognosis, as proposed in this thesis, will play an important role in understanding the genetic basis to common complex diseases.
This thesis contributes to knowledge in two ways:
(i) Providing novel data mining and visualisation frameworks to handle biological data.
(ii) Providing novel visualisations for microarray data to increase understanding of disease.