The scale of engineering problems has sharply increased over the last twenty years. Larger coupled systems, increasing complexity, and limited resources create a need for methods that automatically decompose problems into manageable sub-problems by discovering and leveraging problem structure. The ability to learn the coupling (inter-dependence) structure and reorganize the original problem could lead to large reductions in the time to analyze complex problems. Such decomposition methods could also provide engineering insight on the fundamental physics driving problem solution. This work forwards the current state of the art in engineering decomposition through the application of techniques originally developed within computer science and information theory. The work describes the current state of automatic problem decomposition in engineering and utilizes several promising ideas to advance the state of the practice. Mutual information is a novel metric for data dependence and works on both continuous and discrete data. Mutual information can measure both the linear and non-linear dependence between variables without the limitations of linear dependence measured through covariance. Mutual information is also able to handle data that does not have derivative information, unlike other metrics that require it. The value of mutual information to engineering design work is demonstrated on a planetary entry problem. This study utilizes a novel tool developed in this work for planetary entry system synthesis. A graphical method, force-based clustering, is used to discover related sub-graph structure as a function of problem structure and links ranked by their mutual information. This method does not require the stochastic use of neural networks and could be used with any link ranking method currently utilized in the field. Application of this method is demonstrated on a large, coupled low-thrust trajectory problem. Mutual information also serves as the basis for an alternative global optimizer, called MIMIC, which is unrelated to Genetic Algorithms. Advancement to the current practice demonstrates the use of MIMIC as a global method that explicitly models problem structure with mutual information, providing an alternate method for globally searching multi-modal domains. By leveraging discovered problem inter-dependencies, MIMIC may be appropriate for highly coupled problems or those with large function evaluation cost. This work introduces a useful addition to the MIMIC algorithm that enables its use on continuous input variables. By leveraging automatic decision tree generation methods from Machine Learning and a set of randomly generated test problems, decision trees for which method to apply are also created, quantifying decomposition performance over a large region of the design space.

No description available.

In the human visual system, one of the most prominent functions of the extensive feedback from the higher brain areas within and outside of the visual cortex is attentional modulation. The feedback helps the brain to concentrate its resources on visual features that are relevant for recognition, i. e. it iteratively selects certain aspects of the visual scene for refined processing by the lower areas until the inference process in the higher areas converges to a single hypothesis about this scene. In order to minimize a number of required selection-refinement iterations, one has to find a short sequence of maximally informative portions of the visual input. Since the feedback is not static, the selection process is adapted to a scene that should be recognized. To find a scene-specific subset of informative features, the adaptive selection process on every iteration utilizes results of previous processing in order to reduce the remaining uncertainty about the visual scene. This phenomenon inspired us to develop a computational algorithm solving a visual classification task that would incorporate such principle, adaptive feature selection. It is especially interesting because usually feature selection methods are not adaptive as they define a unique set of informative features for a task and use them for classifying all objects. However, an adaptive algorithm selects features that are the most informative for the particular input. Thus, the selection process should be driven by statistics of the environment concerning the current task and the object to be classified. Applied to a classification task, our adaptive feature selection algorithm favors features that maximally reduce the current class uncertainty, which is iteratively updated with values of the previously selected features that are observed on the testing sample. In information-theoretical terms, the selection criterion is the mutual information of a class variable and a feature-candidate conditioned on the already selected features, which take values observed on the current testing sample. Then, the main question investigated in this thesis is whether the proposed adaptive way of selecting features is advantageous over the conventional feature selection and in which situations. Further, we studied whether the proposed adaptive information-theoretical selection scheme, which is a computationally complex algorithm, is utilized by humans while they perform a visual classification task. For this, we constructed a psychophysical experiment where people had to select image parts that as they think are relevant for classification of these images. We present the analysis of behavioral data where we investigate whether human strategies of task-dependent selective attention can be explained by a simple ranker based on the mutual information, a more complex feature selection algorithm based on the conventional static mutual information and the proposed here adaptive feature selector that mimics a mechanism of the iterative hypothesis refinement. Hereby, the main contribution of this work is the adaptive feature selection criterion based on the conditional mutual information. Also it is shown that such adaptive selection strategy is indeed used by people while performing visual classification.

In this thesis two different problems in distributed sensor networks are considered. Part I involves optimal quantiser design for decentralised estimation of a two-state hidden Markov model with dual sensors. The notion of optimality for quantiser design is based on minimising the probability of error in estimating the hidden Markov state. Equations for the filter error are derived for the continuous (unquantised) sensor outputs (signals), which are used to benchmark the performance of the quantisers. Minimising the probability of filter error to obtain the quantiser breakpoints is a difficult problem therefore an alternative method is employed. The quantiser breakpoints are obtained by maximising the mutual information between the quantised signals and the hidden Markov state. This method is known to work well for the single sensor case. Cases with independent and correlated noise across the signals are considered. The method is then applied to Markov processes with Gaussian signal noise, and further investigated through simulation studies. Simulations involving both independent and correlated noise across the sensors are performed and a number of interesting new theoretical results are obtained, particularly in the case of correlated noise. In Part II, the focus shifts to the detection of faults in helicopter transmission systems. The aim of the investigation is to determine whether the acoustic signature can be used for fault detection and diagnosis. To investigate this, statistical change detection algorithms are applied to acoustic vibration data obtained from the main rotor gearbox of a Bell 206 helicopter, which is run at high load under test conditions.

Title: Analysis of Climate Model Outputs Author: RNDr. Zuzana Chládová E-mail: zuzana.chladova@gmail.com Department: Department of Meteorology and Environment Protection, Faculty of Mathematics and Physics, Charles University in Prague Supervisor: RNDr. Aleš Raidl, Ph.D. Supervisor's e-mail address: ales.raidl@mff.cuni.cz Consultant: doc. RNDr. Jaroslava Kalvová, CSc. Regional climate models are currently the most important tools regularly used for downscaling outputs of global climate models. This analysis compares control and future runs of the global climate models HadCM3, ECHAM5/OPYC3 and ARPÉGE/OPA and the regional climate models RCAO, RCA3, HIRHAM4, HIRHAM5 and ALADIN- CLIMATE/CZ with observed data and CRU data for the Czech Republic. In the period 1961-1990, the global climate models underestimated the air temperature in comparison with corresponding virtual time series representing real data; mean annual courses and variance of the temperature, on the other hand, were simulated satisfactorily. The results of the regional climate models showed overestimation of the model temperature in winter season, while in other seasons the model temperatures corresponded better with real values and the results of simulation were generally more accurate in comparison with global climate models. Concerning...

[pt] Esta dissertação propõe uma metodologia, baseada em procedimentos quantitativos, para auxiliar o diagnóstico de indivíduos portadores de doença cardíaca. A metodologia proposta foi implementada e analisada em um grupo de indivíduos do banco de dados público intitulado Heart Disease Database (Base de Dados pública de Doença Cardíaca) (Aha, atualizado em 2001), diagnosticados nas cidades de Cleveland e Long Beach, nos Estados Unidos. Os resultados obtidos neste estudo foram comparados aos resultados de outros autores encontrados na literatura, de forma a se ter uma medida da qualidade dos resultados aqui obtidos. Foram utilizadas também outras técnicas de classificação de padrões conhecidas na literatura, denominadas Análise Discriminante e Algoritmo C4.5, de forma a estabelecer comparações com os resultados obtidos nesta dissertação utilizando Redes Neurais, e aplicar a metodologia sugerida na divisão dos conjuntos de treinamento/generalização. Os resultados obtidos foram satisfatórios. Um percentual de acerto médio de 91,0 % foi atingido, enquanto que outros resultados de estudos usando a mesma base de dados alcançaram percentuais de acerto médio de 83,0 % (Ho & Chou, 2001) e 83,5 % (Hu, Li, Cai & Xu, 2004). O desempenho da Rede Neural também foi melhor quando comparado ao da Análise Discriminante e do Algoritmo C4.5. A metodologia de divisão dos conjuntos de treinamento/generalização sugerida nesta dissertação promoveu melhorias em todas as três técnicas de classificação de padrões utilizadas. Acredita-se que os resultados obtidos poderão auxiliar as condutas médicas em relação ao diagnóstico de doença cardíaca, podendo, portanto, vir a ser úteis na prevenção e/ou tratamento de doenças cardíacas. / [en] This dissertation proposes a methodology, established in quantitative procedures, to assist the diagnostic of individuals with heart disease. The proposed methodology was implemented and analyzed in a group of individuals of the public database called Heart Disease Database (Aha, current in 2001), diagnosed in the cities of Cleveland and Long Beach, in the United States. The results gotten in this study had been compared with the results of other authors found in literature to have a measure of the quality of the results gotten here. Others techniques of classification of standards known in literature had also been used, called Discriminate Analysis and C4.5 Algorithm, to establish comparisons with the results gotten in this dissertation using Neural Networks, and to apply the methodology suggested in the division of the sets of training/generalization. The gotten results were satisfactory. A percentage of average rightness of 91.0 % was reached, whereas other results of studies using the same database had reached percentages of average rightness of 83.0 % (Ho & Chou, 2001) and 83.5 % (Hu, Li, Cai & Xu, 2004). The performance of the Neural Network was also better when compared with Discriminate Analysis and C4.5 Algorithm. The methodology of division of the sets of training/generalization suggested in this dissertation promoted improvements in all the three used techniques of classification of standards. It´s believable that the gotten results will be able to assist the medical behaviors in relation to the diagnostic of heart disease, becoming useful in the prevention and/or treatment of heart diseases.

Orientador: Gabriela Castellano / Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Física Gleb Wataghin / Made available in DSpace on 2018-08-30T20:31:19Z (GMT). No. of bitstreams: 1 TapiaHerrera_LuisCarlos1982-_D.pdf: 17368597 bytes, checksum: b04bfdc96a80f7bba2cdca7390a9d09e (MD5) Previous issue date: 2016 / Resumo: Os neurônios são elementos que no cérebro trabalham em grupo e de forma organizada. A técnica de ressonância magnética funcional (fMRI) permite identificar redes corticais e subcorticais do cérebro quando ele desenvolve atividades cognitivas motoras ou perceptivas. No entanto, redes nomeadas de redes em estado de repouso, estão presentes em ausência de tarefas específicas. Alguns estudos modelaram redes funcionais do cérebro com a ajuda da teoria de grafos. Um dos objetivos deste trabalho foi analisar, utilizando teoria de grafos, dados funcionais do cérebro coletados com a técnica de fMRI, de 10 voluntários saudáveis, que participaram de dois protocolos: uma aquisição em estado de repouso e outra durante uma tarefa de produção de palavras. Outro objetivo do trabalho foi testar duas métricas matemáticas (correlação de Pearson e informação mútua), para determinar quais delas conseguem captar melhor diferenças entre as duas condições mencionadas. Também se objetivou comparar parâmetros termodinâmicos das redes de repouso obtidas por meio dos dados reais com os de redes simuladas computacionalmente via modelo de Ising. Finalmente, um último objetivo foi explorar os dados para ver que informação poderia ser obtida a partir dos mesmos, sem uso prévio de modelos sobre as tarefas realizadas. Utilizando a teoria de grafos, achamos diferenças entre as redes nas condições de repouso e de produção de palavras para os parâmetros grau médio e coeficiente de cluster. Adicionalmente foram comparadas as redes dos hemisférios direito e esquerdo nas redes geradas na condição de produção de palavras, e achamos que o grau médio das redes pode predizer a lateralização (dominância hemisférica para linguagem), também achada com análises padrões de fMRI. Relativo às métricas matemáticas, a correlação de Pearson e a informação mútua foram comparadas para determinar qual destas métricas captura melhor a similaridade ou sincronia entre duas séries temporais que contêm atividade hemodinâmica do cérebro. Concluímos que a correlação linear é uma medida capaz de caracterizar de forma satisfatória a sincronia entre duas séries desse tipo. Simulações computacionais do modelo de Ising foram desenvolvidas para posteriormente criar redes funcionais em três regimes diferentes: crítico, subcrítico e supercrítico. Esta abordagem do estado de repouso foi examinada em trabalhos prévios, e foi concluído que o cérebro como sistema dinâmico possui uma maior semelhança com o sistema simulado no regime crítico. Finalmente, uma metodologia independente de modelo foi implementada para detectar áreas ativas do cérebro em tarefas dirigidas. Esta metodologia foi testada nos dados na condição de produção de palavras, permitindo identificar as áreas envolvidas na execução da tarefa / Abstract: Neuronal elements in the brain are not isolated, they work together and work in an organized way. The functional magnetic resonance imaging (fMRI) technique allows identifying cortical networks when the brain develops a task. However, resting state brain networks are present in the absence of any task. Some studies have modeled the brain networks architecture with aid of graph theory. One of the main aims of this work was the analysis of resting state and language task fMRI data sets, of ten healthy subjects, using graph theory. In order to study the linear and nonlinear relationships between time series of cortical areas of the brain, two metrics were compared the Pearson correlation and the mutual information. Also, graphs parameters built from resting state data and graph parameters built using simulations of the Ising model were compared. Finally, we developed a methodology to study the time series of differents regions of the brain in order to obtain information of the task without using predefined models of the brain activity. We found differences in the mean degree and the cluster coefficient of the network between the two conditions. In addition, we compared the networks corresponding to the left and right hemispheres during the language task, and found that the mean degree of these networks can predict the language lateralization found with standard fMRI analysis in most cases. The mean degree of the network and the cluster coefficient shows differences for the two conditions. Relative to the comparison between the Pearson correlation and the mutual information, we conclude that the linear correlation is an efficient metric to characterize the synchrony between the haemodynamic time series of the brain. Computational simulations of the Ising model for three different phases were developed: in critical, subcritical and supercritical phases. This comparison was presented in a previous work, and it was concluded that the brain as a dynamical system has remarkable similarities with the computational model in the critical phase. Relatively to the model independent methodology developed, it was possible to identify brain areas engaged with the word production task / Doutorado / Física / Doutor em Ciências / 157356/2011-6 / CNPQ

This thesis extends a system of an artificial player of a word-association game Codenames to easy addition of support for new languages. The system is able to play Codenames in roles as a guessing player, a clue giver or, by their combination a Duet version player. For analysis of different languages a neural toolkit Stanza was used, which is language independent and enables automated processing of many languages. It was mainly about lemmatization and part of speech tagging for selection of clues in the game. For evaluation of word associations were several models tested, where the best results had a method Pointwise Mutual Information and predictive model fastText. The system supports playing Codenames in 36 languages comprising 8 different alphabets.

The work is mainly concerned with the possibilities of breaking the encryption algorithm AES with using of non-template attacks. In the introduction are listed techniques of differential analysis, which are using in the present, but for the sake of completeness is there mention about simple power analysis. In the next chapters are briefly described countermeasures against power analysis and further is described the AES algorithm. Most important parts are chapters where are described attack implementation on AES-128 through correlation power analysis and mutual information analysis. These attacks exploit power traces from www pages dedicated to book Power Analysis Attacks - Revealing the Secrets of Smartcards, http://DPAbook.org and especially to power traces from DPA Contest 4.2, http://www.dpacontest.org. In conclusion is comparison of methods based on the number of power traces needed for finding the key of secret message.

No description available.