Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal

Dissertation (D.M.A.)--The University of North Carolina at Greensboro, 2010. / Directed by John Locke; submitted to the School of Music. Title from PDF t.p. (viewed Jul. 7, 2010). Includes bibliographical references (p. 43-45).

Submitted by Lucelia Lucena (lucelia.lucena@ufpe.br) on 2015-03-09T19:29:39Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) DISSERTAÇÃO Luíz Vieira e Silva Filho.pdf: 2176053 bytes, checksum: 4882a96e67804421bca22e07debc49da (MD5) / Made available in DSpace on 2015-03-09T19:29:39Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) DISSERTAÇÃO Luíz Vieira e Silva Filho.pdf: 2176053 bytes, checksum: 4882a96e67804421bca22e07debc49da (MD5) Previous issue date: 2014-02-17 / Sistemas de Múltiplos Classificadores (Multiple Classifiers Systems - MCS) se baseiam na ideia de que combinar a opinião de vários especialistas pode produzir melhores resultados do que quando se usa apenas um especialista. Diversas técnicas de MCS foram desenvolvidas, apresentando pontos fortes e fracos, a depender do contexto em que são aplicadas. Este trabalho propõe uma arquitetura para MCS que visa potencializar a complementaridade entre essas técnicas, possuindo dois objetivos principais: i) a combinação de métodos de amostragem tradicionais, visando a geração de classificadores de melhor desempenho que componham um pool de classificadores; ii) a aplicação de um algoritmo de poda para remover do pool aqueles classificadores incompetentes para lidar com o problema em questão, considerando os critérios de seleção adotados. A arquitetura proposta foi avaliada em uma aplicação de credit-scoring. Os métodos de amostragem usados foram o Bagging e o Random Subspace com classificadores-base sendo árvores-de-decisão, construídas com base no algoritmo CART. Para o processamento da poda foi usado o algoritmo Orientation Ordering, e para combinação das saídas dos classificadores do ensemble adotou-se o método Majority Vote. Os experimentos realizados mostraram que a arquitetura proposta alcançou taxas de acerto similares ou superiores às atingidas pelos métodos apresentados na literatura. Esses resultados ainda foram obtidos com ensembles cujos tamanhos eram da ordemde 20% dos pools originais gerados na fase de treinamento.

The composition is scored for the following instruments: 2 flutes, 2 oboes, 2 clarinets, 2 bassoons and a large percussion section requiring 7 performers. Beneath the Dancing Moon is a programmatic piece in one movement form composed of 5 continuous sections. It depicts a night scene when the elves begin to dance beneath the moon. Later, the moaning ghosts from the dark forest and the witches with brooms come to join them. They dance furiously until the moon disappears, the sea stops dead and all the dancers suddenly vanish. The approximate performance time is 17 minutes.

RNA Sequencing has recently emerged as a breakthrough technology for microRNA (miRNA) discovery. This technology has allowed the discovery of thousands of miRNAs in a large number of species. However, despite the benefits of this technology, it also carries its own limitations, including the need for sequencing read libraries and of the genome. Differently, ab initio computational methods need only the genome as input to search for genonic locus likely to give rise to novel miRNAs. In the core of most of these methods, there are predictive models induced by using data mining techniques able to distinguish between real (positive) and pseudo (negative) miRNA precursors (pre-miRNA). Nevertheless, the applicability of current literature ab initio methods have been compromised by high false detection rates and/or by other computational difficulties. In this work, we investigated how the main aspects involved in the induction of predictive models for pre-miRNA affect the predictive performance. Particularly, we evaluate the discriminant power of feature sets proposed in the literature, whose computational costs and composition vary widely. The computational experiments were carried out using sequence data from 45 species, which covered species from eight phyla. The predictive performance of the classification models induced using large training set sizes (≥ 1; 608) composed of instances extracted from real and pseudo human pre-miRNA sequences did not differ significantly among the feature sets that lead to the maximal accuracies. Moreover, the differences in the predictive performances obtained by these models, due to the learning algorithms, were neglectable. Inspired by these results, we obtained a feature set which can be computed 34 times faster than the less costly among those feature sets, producing the maximal accuracies, albeit the proposed feature set has achieved accuracy within 0.1% of the maximal accuracies. When classification models using the elements previously discussed were induced using small training sets (120) from 45 species, we showed that the feature sets that produced the highest accuracies in the classification of human sequences were also more likely to produce higher accuracies for other species. Nevertheless, we showed that the learning complexity of pre-miRNAs vary strongly among species, even among those from the same phylum. These results showed that the existence of specie specific features indicated in previous studies may be correlated with the learning complexity. As a consequence, the predictive accuracies of models induced with different species and same features and instances spaces vary largely. In our results, we show that the use of training examples from species phylogenetically more complex may increase the predictive performances for less complex species. Finally, by using ensembles of computationally less costly feature sets, we showed alternative ways to increase the predictive performance for many species while keeping the computational costs of the analysis lower than those using the feature sets from the literature. Since in miRNA discovery the number of putative miRNA loci is in the order of millions, the analysis of putative miRNAs using a computationally expensive feature set and or inaccurate models would be wasteful or even unfeasible for large genomes. In this work, we explore most of the learning aspects implemented in current ab initio pre-miRNA prediction tools, which may lead to the development of new efficient ab initio pre-miRNA discovery tools / O sequenciamento de pequenos RNAs surgiu recentemente como uma tecnologia inovadora na descoberta de microRNAs (miRNA). Essa tecnologia tem facilitado a descoberta de milhares de miRNAs em um grande número de espécies. No entanto, apesar dos benefícios dessa tecnologia, ela apresenta desafios, como a necessidade de construir uma biblioteca de pequenos RNAs, além do genoma. Diferentemente, métodos computacionais ab initio buscam diretamente no genoma regiões prováveis de conter miRNAs. A maioria desses métodos usam modelos preditivos capazes de distinguir entre os verdadeiros (positivos) e pseudo precursores de miRNA - pre-miRNA - (negativos), os quais são induzidos utilizando técnicas de mineração de dados. No entanto, a aplicabilidade de métodos ab initio da literatura atual é limitada pelas altas taxas de falsos positivos e/ou por outras dificuldades computacionais, como o elevado tempo necessário para calcular um conjunto de atributos. Neste trabalho, investigamos como os principais aspectos envolvidos na indução de modelos preditivos de pre-miRNA afetam o desempenho preditivo. Particularmente, avaliamos a capacidade discriminatória de conjuntos de atributos propostos na literatura, cujos custos computacionais e a composição variam amplamente. Os experimentos computacionais foram realizados utilizando dados de sequências positivas e negativas de 45 espécies, cobrindo espécies de oito filos. Os resultados mostraram que o desempenho preditivo de classificadores induzidos utilizando conjuntos de treinamento com 1608 ou mais vetores de atributos calculados de sequências humanas não diferiram significativamente, entre os conjuntos de atributos que produziram as maiores acurácias. Além disso, as diferenças entre os desempenhos preditivos de classificadores induzidos por diferentes algoritmos de aprendizado, utilizando um mesmo conjunto de atributos, foram pequenas ou não significantes. Esses resultados inspiraram a obtenção de um conjunto de atributos menor e que pode ser calculado até 34 vezes mais rapidamente do que o conjunto de atributos menos custoso produzindo máxima acurácia, embora a acurácia produzida pelo conjunto proposto não difere em mais de 0.1% das acurácias máximas. Quando esses experimentos foram executados utilizando vetores de atributos calculados de sequências de outras 44 espécies, os resultados mostraram que os conjuntos de atributos que produziram modelos com as maiores acurácias utilizando vetores calculados de sequências humanas também produziram as maiores acurácias quando pequenos conjuntos de treinamento (120) calculados de exemplos de outras espécies foram utilizadas. No entanto, a análise destes modelos mostrou que a complexidade de aprendizado varia amplamente entre as espécies, mesmo entre aquelas pertencentes a um mesmo filo. Esses resultados mostram que a existência de características espécificas em pre-miRNAs de certas espécies sugerida em estudos anteriores pode estar correlacionada com a complexidade de aprendizado. Consequentemente, a acurácia de modelos induzidos utilizando um mesmo conjunto de atributos e um mesmo algoritmo de aprendizado varia amplamente entre as espécies. i Os resultados também mostraram que o uso de exemplos de espécies filogeneticamente mais complexas pode aumentar o desempenho preditivo de espécies menos complexas. Por último, experimentos computacionais utilizando técnicas de ensemble mostraram estratégias alternativas para o desenvolvimento de novos modelos para predição de pre-miRNA com maior probabilidade de obter maior desempenho preditivo do que estratégias atuais, embora o custo computacional dos atributos seja inferior. Uma vez que a descoberta de miRNAs envolve a análise de milhares de regiões genômicas, a aplicação prática de modelos preditivos de baixa acurácia e/ou que dependem de atributos computacionalmente custosos pode ser inviável em análises de grandes genomas. Neste trabalho, apresentamos e discutimos os resultados de experimentos computacionais investigando o potencial de diversas estratégias utilizadas na indução de modelos preditivos para predição ab initio de pre-miRNAs, que podem levar ao desenvolvimento de ferramentas ab initio de maior aplicabilidade prática

Dans ce travail nous nous intéressons aux problèmes liés à la combinaison d'informations en provenance de sources multiples. Nous proposons de représenter les informations en provenance de la théorie des ensembles flous (FST) et de la théorie de l'évidence (DST) dans le cadre unificateur des ensembles aléatoires (RST). Le processus de combinaison fait face à deux problématiques majeures : (1) une explosion du temps de calcul dû au grand nombre d'éléments focaux, et (2) la combinaison d'informations en conflit total. Nous proposons dans un premier temps de réduire le temps de calcul du processus de combinaison, en appliquant une approximation directe aux informations de la FST qui s'avère très efficace lorsque la cardinalité du cadre de discernement est élevée. Dans un deuxième temps nous proposons une formulation générale pour les règles de combinaison de la RST, ainsi qu'une nouvelle classe de règles adaptatives qui a l'avantage de (a) prendre en compte de manière automatique la fiabilité des sources, (b) combiner des informations définies sur des cadres de discernement différents et homogènes. Elle possède un comportement similaire à la règle conjonctive lorsque les sources sont en accord et un comportement similaire à la règle disjonctive lorsque les sources sont en désaccord.

.

This single movement work is written for 2 flutes, 2 oboes, 3 clarinets in Bb, bass clarinet in Bb, 2 bassoons, alto saxophone in Eb, 2 horns in F, 2 trumpets in Bb, trombone, euphonium, tuba, contra bass, and 3 percussion. The approximate length is eight minutes. Both traditional and proportional systems of notation are employed. The entire piece is freely chromatic with some implications of whole tone and other nondiatonic scales. The harmonies are tertian yet have no functional tonal basis. Changing meters with asymmetrical divisions are used in all sections except C and E, which have time indications (in seconds) for each measure with subdivisions to aid the conductor. There are seven major formal divisions: A B transition C retransition A' D E.

The increasing variety of data mining tools offers a large palette of types and representation formats for predictive models. Managing the models then becomes a big challenge, as well as reusing the models and keeping the consistency of model and data repositories. Sustainable access and quality assessment of these models become limited to researchers. The approach for the Data and Model Governance (DMG) makes easier to process and support complex solutions. In this thesis, contributions are proposed towards ensembles of models with a focus on model representation, comparison and usage. Predictive Toxicology was chosen as an application field to demonstrate the proposed approach to represent predictive models linked to data for DMG. Further analysing methods such as predictive models comparison and predictive models combination for reusing the models from a collection of models were studied. Thus in this thesis, an original structure of the pool of models was proposed to represent predictive toxicology models called Predictive Toxicology Markup Language (PTML). PTML offers a representation scheme for predictive toxicology data and models generated by data mining tools. In this research, the proposed representation offers possibilities to compare models and select the relevant models based on different performance measures using proposed similarity measuring techniques. The relevant models were selected using a proposed cost function which is a composite of performance measures such as Accuracy (Acc), False Negative Rate (FNR) and False Positive Rate (FPR). The cost function will ensure that only quality models be selected as the candidate models for an ensemble. The proposed algorithm for optimisation and combination of Acc, FNR and FPR of ensemble models using double fault measure as the diversity measure improves Acc between 0.01 to 0.30 for all toxicology data sets compared to other ensemble methods such as Bagging, Stacking, Bayes and Boosting. The highest improvements for Acc were for data sets Bee (0.30), Oral Quail (0.13) and Daphnia (0.10). A small improvement (of about 0.01) in Acc was achieved for Dietary Quail and Trout. Important results by combining all the three performance measures are also related to reducing the distance between FNR and FPR for Bee, Daphnia, Oral Quail and Trout data sets for about 0.17 to 0.28. For Dietary Quail data set the improvement was about 0.01 though, but this data set is well known as a difficult learning exercise. For five UCI data sets tested, similar results were achieved with Acc improvement between 0.10 to 0.11, closing more the gaps between FNR and FPR. As a conclusion, the results show that by combining performance measures (Acc, FNR and FPR), as proposed within this thesis, the Acc increased and the distance between FNR and FPR decreased.

The Supergen wind research consortium is a group of research centres which undertake research primarily aimed at reducing the cost of offshore wind farming. Research is undertaken to apply the WRF mesoscale NWP model to the field of offshore wind resource assessment to assess its potential as an operational tool. WRF is run in a variety of configurations for a number of locations to determine and optimise a level of performance and assess how accessible that performance might be to an end user. Three studies set out to establish a level of performance at two different sites and improve performance through optimisation of model setup and post processing techniques. WRF was found to simulate wind speed to an appreciable level by reference to similar studies, though performance was found to vary throughout the course of the model runs and depending on the location. An average correlation coefficient of 0.9 was found for the Shell Flats resource assessment at 6-hourly resolution with an RMSE of 1.7ms-1. Performance at Scroby Sands was not at as high a level as that seen for Shell Flats with an average correlation coefficient for wind speed of 0.64 with an RMSE of 2ms-1. A range of variables were simulated by the model in the Shell Flats investigation to test the flexibility of the model output. Wind direction was produced to a moderate level of accuracy at 10-minute resolution while aggregated stability statistics showed the model had a good appreciation of the frequency of cases observed. Areas of uncertainty in model performance were addressed through model optimisation techniques including the generation of two ensembles and observational nudging. Both techniques were found to add value to the model output as well as improving performance. The difference between performance observed at Shell Flats and Scroby Sands shows that while the model clearly has inherent skill it is sensitive to the environment to which it is applied. In order to maximise performance, as large a computing resource as possible is recommended with a concerted effort to optimise model setup with the aim of allowing it to perform to its best ability. There is room for improvement in the application of mesoscale NWP to the field of offshore wind resource assessment but these results confirm an inherent skill in model performance. With the addition of further validation, improvements to model setup on a case by case basis and the application of optimisation techniques, it is anticipated mesoscale NWP can perform to a level which would justify its adoption operationally by the industry. The flexibility which can be offered relating to spatial and temporal coverage as well as the range of variables which can be produced make it an attractive option to developers if performance of a consistently high level can be established.