abstract: Internet and social media devices created a new public space for debate on political and social topics (Papacharissi 2002; Himelboim 2010). Hotly debated issues span all spheres of human activity; from liberal vs. conservative politics, to radical vs. counter-radical religious debate, to climate change debate in scientific community, to globalization debate in economics, and to nuclear disarmament debate in security. Many prominent ’camps’ have emerged within Internet debate rhetoric and practice (Dahlberg, n.d.). In this research I utilized feature extraction and model fitting techniques to process the rhetoric found in the web sites of 23 Indonesian Islamic religious organizations, later with 26 similar organizations from the United Kingdom to profile their ideology and activity patterns along a hypothesized radical/counter-radical scale, and presented an end-to-end system that is able to help researchers to visualize the data in an interactive fashion on a time line. The subject data of this study is the articles downloaded from the web sites of these organizations dating from 2001 to 2011, and in 2013. I developed algorithms to rank these organizations by assigning them to probable positions on the scale. I showed that the developed Rasch model fits the data using Andersen’s LR-test (likelihood ratio). I created a gold standard of the ranking of these organizations through an expertise elicitation tool. Then using my system I computed expert-to-expert agreements, and then presented experimental results comparing the performance of three baseline methods to show that the Rasch model not only outperforms the baseline methods, but it was also the only system that performs at expert-level accuracy. I developed an end-to-end system that receives list of organizations from experts, mines their web corpus, prepare discourse topic lists with expert support, and then ranks them on scales with partial expert interaction, and finally presents them on an easy to use web based analytic system. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2016

Dissertação (mestrado)-Universidade de Brasília, Departamento de Ciências Contábeis e Atuariais, 2010. / Submitted by Débora Amorim Romcy Pereira (deboraromcy@bce.unb.br) on 2011-06-21T13:13:08Z No. of bitstreams: 1 2010_DarleneLeiteSilva.pdf: 534840 bytes, checksum: 9a481906ba839e102021bb248ad84045 (MD5) / Approved for entry into archive by Daniel Ribeiro(daniel@bce.unb.br) on 2011-06-21T15:11:41Z (GMT) No. of bitstreams: 1 2010_DarleneLeiteSilva.pdf: 534840 bytes, checksum: 9a481906ba839e102021bb248ad84045 (MD5) / Made available in DSpace on 2011-06-21T15:11:41Z (GMT). No. of bitstreams: 1 2010_DarleneLeiteSilva.pdf: 534840 bytes, checksum: 9a481906ba839e102021bb248ad84045 (MD5) / As mudanças ocorridas nas últimas décadas com a globalização influenciaram o desempenho e o processo de gestão das organizações. Na realidade, as organizações buscam combinar oportunidade de desenvolvimento com um estilo de liderança próprio, buscando atingir boa relação com empregados, que, consequentemente, obterão melhores performances, por razões como: alta qualidade do ambiente de trabalho, além do aumento de sua vantagem competitiva. Esta pesquisa consiste em analisar se existe diferença significante no desempenho econômico-financeiro das melhores empresas para se trabalhar (MET) indicadas pela Revista Exame – Você S/A de 2007, 2008 e 2009, e as não integrantes deste ranking, no tocante a: liquidez corrente (LC), margem bruta (MB), margem líquida (ML), retorno sobre os ativos (ROA) e produtividade (PROD). Este estudo buscou basear-se no estudo de Fulmer, Gearhart e Scott (2003), que testaram as “100 melhores empresas para se trabalhar nos EUA”, publicado na Revista Fortune. Para tanto, a metodologia adotada constitui-se de um estudo empírico-analítico, realizado com empresas brasileiras selecionadas a partir do banco de dados da Revista Exame – Você S/A, sendo selecionadas apenas as empresas de capital aberto e o mesmo quantitativo de empresas não integrantes deste ranking para o grupo de controle, que tiveram seus demonstrativos contábeis publicado na CVM. Na análise dos dados foram utilizadas medidas de tendência central: média e desvio padrão. Na decisão da aceitação ou rejeição das hipóteses, foi utilizado o teste de Mann-Whitney para amostras independentes. Como resultado, comprovou-se não haver diferença significante entre os dois grupos de empresas: ranqueadas (R) e não ranqueadas (NR), o que conduz à conclusão de que as melhores empresas para se trabalhar no Brasil, com ênfase nas boas práticas e políticas de RH, não apresentam desempenho econômico-financeiro superiores às empresas não integrantes do ranking. _________________________________________________________________________________ ABSTRACT / The changes happened in the last decades with the globalization influenced the performance and the process of management of the organizations. Actuality, the organizations seek to combine development opportunity with an own leadership style, looking for reaching good relation with employees, that, consequently, will obtain better performances, for reasons as: high labor quality of the environment, besides the increase of their competitive advantage. This study consists of analyzing if significant difference exists in the economical-financial acting of the best companies to work (MET) recommended for the Exame – Você S/A Magazine 2007, 2008 and 2009, and the not integrant of this ranking, concerning to: average liquidity (LC), gross margin (MB) liquid margin (ML), return on assets (ROA) and productivity (PROD). This study looked for to base in the study of Fulmer, Gearhart and Scott (2003), that tested the “100 best companies to work in the USA”, published in the Fortune Magazine. For that, the adopted methodology is constituted of an empiric-analytical study, accomplished with Brazilian companies selected from the Exame – Você S/A database system, being selected only the open companies and the same quantitative of companies not integrant of this ranking for the control group, that had their accounting demonstratives published in CVM. In the analysis of the data measure of central trend were used: average and standard deviation. In the decision of the acceptance or rejection of the hypotheses, the test of Mann-Whitney was used for independent samples. As result, it was proved that there’s no significant difference among the two groups of companies: ranked (R) and not ranked (NR), what conducts to the conclusion that the best companies to work in Brazil, with emphasis in the good practices and policies of human resources, don’t show superior economical-financial performance to the companies not integrant of the ranking.

No description available.

Artificial Neural Networks (ANNs) are known to be among the most effective and expressive machine learning models. Their impressive abilities to learn have been reflected in many broad application domains such as image recognition, medical diagnosis, online banking, robotics, dynamic systems, and many others. ANNs with multiple layers of complex non-linear transformations (a.k.a Deep ANNs) have shown recently successful results in the area of computer vision and speech recognition. ANNs are parametric models that approximate unknown functions in which parameter values (weights) are adapted during training. ANN’s weights can be large in number and thus render the trained model more complex with chances for “overfitting” training data. In this study, we explore the effects of network pruning on performance of ANNs and ranking of features that describe the data. Simplified ANN model results in fewer parameters, less computation and faster training. We investigate the use of Hessian-based pruning algorithms as well as simpler ones (i.e. non Hessian-based) on nine datasets with varying number of input features and ANN parameters. The Hessian-based Optimal Brain Surgeon algorithm (OBS) is robust but slow. Therefore a faster parallel Hessian- approximation is provided. An additional speedup is provided using a variant we name ‘Simple n Optimal Brain Surgeon’ (SNOBS), which represents a good compromise between robustness and time efficiency. For some of the datasets, the ANN pruning experiments show on average 91% reduction in the number of ANN parameters and about 60% - 90% in the number of ANN input features, while maintaining comparable or better accuracy to the case when no pruning is applied. Finally, we show through a comprehensive comparison with seven state-of-the art feature filtering methods that the feature selection and ranking obtained as a byproduct of the ANN pruning is comparable in accuracy to these methods.

In this thesis, we investigate how well we can reconstruct the best rank-? approximation of a large matrix from a small number of its entries. We show that even if a data matrix is of full rank and cannot be approximated well by a low-rank matrix, its best low-rank approximations may still be reliably computed or estimated from a small number of its entries. This is especially relevant from a statistical viewpoint: the best low-rank approximations to a data matrix are often of more interest than itself because they capture the more stable and oftentimes more reproducible properties of an otherwise complicated data-generating model. In particular, we investigate two agnostic approaches: the first is based on spectral truncation; and the second is a projected gradient descent based optimization procedure. We argue that, while the first approach is intuitive and reasonably effective, the latter has far superior performance in general. We show that the error depends on how close the matrix is to being of low rank. Our results can be generalized to the spectral and entrywise error and provide flexible tools for the error analysis of the follow-up computation. Moreover, we derive a high-order decomposition of the error. With an explicit expression of the main error source, we obtain an improved estimate of the linear form. Both theoretical and numerical evidence is presented to demonstrate the effectiveness of the proposed approaches.

In this thesis we consider a few results related to the Zermelo and Extended Zermelo Model as well as outline some partial results and open problems related thereto. First we will analyze a discrete dynamical system considering under what conditions the convergence of this dynamical system predicts the outcome of the Extended Zermelo Model. In the following chapter we will focus on the Zermelo Model by giving a method for simplifying the derivation of Zermelo ratings for tournaments in terms of specific types of strongly connected components. Following this, the idea of stability of a tournament will be discussed and an upper bound will be obtained on the stability of three-team tournaments. Finally, we will conclude with some partial results related to the topics presented in the previous chapters.

Arrow's Impossibility Theorem outlines the flaws that effect any voting system that attempts to order a set of objects. For its entire history, American college football has been determining its champion based on a voting system. Much of the literature has dealt with why the voting system used is problematic, but there does not appear to be a large collection of work done to create a better, mathematical process. More generally, the inadequacies of ranking in football are a manifestation of the problem of ranking a set of objects. Herein, principal component analysis is used as a tool to provide a solution for the problem, in the context of American college football. To show its value, rankings based on principal component analysis are compared against the rankings used in American college football. / Thesis / Master of Science (MSc) / The problem of ranking is a ubiquitous problem, appearing everywhere from Google to ballot boxes. One of the more notable areas where this problem arises is in awarding the championship in American college football. This paper explains why this problem exists in American college football, and presents a bias-free mathematical solution that is compared against how American college football awards their championship.

No description available.

No description available.

[pt] Considere uma rede com k nodos que pode apresentar falhas ao longo de sua operação. Além disso, assuma que é inviável verificar todos os nodos sempre que uma falha ocorre. Motivados por este cenário, propomos um método que usa aprendizado de máquina supervisionado para gerar rankings dos nodos mais prováveis por serem responsáveis pela falha. O método proposto é um metaclassificador que pode utilizar qualquer tipo de classificador internamente, onde o modelo gerado pelo metaclassificador é uma composição daqueles gerados pelos classificadores internos. Cada modelo interno é treinado com um subconjunto dos dados. Estes subconjuntos são criados sucessivamente a partir dos dados originais eliminando-se algumas instâncias. As instâncias eliminadas são aquelas cujas classes já foram colocadas no ranking. Métricas derivadas da Acurácia, Precision e Recall foram propostas e usadas para avaliar este método. Utilizando uma base de domínio público, verificamos que os tempos de treinamento e classificação do metaclassificador são maiores que os de um classificador simples. Entretanto ele atinge resultados melhores em alguns casos, como ocorre com as árvores de decisão, que superam a acurácia do benchmark por uma margem maior que 5 por cento. / [en] Consider a network with k nodes that may fail along its operation. Furthermore assume that it is impossible to check all nodes whenever a failure occurs. Motivated by this scenario, we propose a method that uses supervised learning to generate rankings of the most likely nodes responsible for the failure. The proposed method is a meta-classifier that is able to use any kind of classifier internally, where the model generated by the meta-classifier is a composition of those generated by the internal classifiers. Each internal model is trained with a subset of the data created from the elimination of instances whose classes were already put in the ranking. Metrics derived from Accuracy, Precision and Recall were proposed and used to evaluate this method. Using a public data set, we verified that the training and classification times of the meta-classifier were greater than those of a simple classifier. However it reaches better results in some cases, as with the decision trees, that exceeds the benchmark accuracy for a margin greater than 5 percent.