This thesis gathers some contributions to statistical pattern recognition particularly targeted at problems in which the feature vectors are high-dimensional. Three pattern recognition scenarios are addressed, namely pattern classification, regression analysis and score fusion. For each of these, an algorithm for learning a statistical model is presented. In order to address the difficulty that is encountered when the feature vectors are high-dimensional, adequate models and objective functions are defined. The strategy of learning simultaneously a dimensionality reduction function and the pattern recognition model parameters is shown to be quite effective, making it possible to learn the model without discarding any discriminative information. Another topic that is addressed in the thesis is the use of tangent vectors as a way to take better advantage of the available training data. Using this idea, two popular discriminative dimensionality reduction techniques are shown to be effectively improved. For each of the algorithms proposed throughout the thesis, several data sets are used to illustrate the properties and the performance of the approaches. The empirical results show that the proposed techniques perform considerably well, and furthermore the models learned tend to be very computationally efficient. / Villegas Santamaría, M. (2011). Contributions to High-Dimensional Pattern Recognition [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/10939

El objetivo en esta Memoría ha sido el análisis de eficiencia de un determinado sector empresarial, teniendo en cuenta dos problemas casi siempre presentes, y de naturaleza muy diferente, por una parte, que los datos que se manejan pueden ser imprecisos y, por tanto, afectar al resultado de cualquier estudio de eficiencia y, por otra parte, el deseo de ordenar las empresas (Unidades De Toma de Decisión) atendiendo a la medición de su eficiencia. Para la medición de la eficiencia se ha recurrido a la metodología no paramétrica del Análisis Envolvente de datos (DEA) aplicandola a empresas del sector textil muy cercanas a nosotros. Ahora bien, dado que consideramos que siempre existe alguna incertidumbre o un posible error en la medición de algunos datos (inputs y outputs), introducimos la limitación de la certeza con el tratamiento fuzzy de los datos, métodos que no requieren conocer ni aplicar hipótesis sobre distribuciones de probabilidad de esos datos, que dicho sea de paso, podría no ser fáctible bajo determinados supuestos de incertidumbre. Pero además de la medir la eficiencia pretendemos proporcionar más información que la mera separación dicotómica entre empresas eficientes o no eficientes. Para ello desarrollamos y aplicamos los modelos de super-efficiencyfuzzy y cross-efficiency-fuzzy, que nos permiten establecer una ordenación bajo incertidumbre. Con este trabajo hemos realizado un estudio amplio de la eficiencia bajo incertidumbre. Se observa que los resultados obtenidos aplicando los distintos métodos son similares. Además, estos métodos proporcionan más información sobre las unidades estudiadas que las que proporciona un solo índice de eficiencia. Estos métodos pueden ser aplicables a otros tipos de empresas, aportando nueva información que puede ayudar u orientar en la toma de decisiones de sus gestores / Pla Ferrando, ML. (2013). Modelos flexibles para la valoración de la eficiencia [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/31521

Learning and reasoning with common sense is a challenging problem in Artificial Intelligence (AI). Humans have the remarkable ability to interpret images and text from different perspectives in multiple modalities, and to use large amounts of commonsense knowledge while performing visual or textual tasks. Inspired by that ability, we approach commonsense learning as leveraging perspectives from multiple modalities for images and text in the context of vision and language tasks. Given a target task (e.g., textual reasoning, matching images with captions), our system first represents input images and text in multiple modalities (e.g., vision, text, abstract scenes and facts). Those modalities provide different perspectives to interpret the input images and text. And then based on those perspectives, the system performs reasoning to make a joint prediction for the target task. Surprisingly, we show that interpreting textual assertions and scene descriptions in the modality of abstract scenes improves performance on various textual reasoning tasks, and interpreting images in the modality of Visual Question Answering improves performance on caption retrieval, which is a visual reasoning task. With grounding, imagination and question-answering approaches to interpret images and text in different modalities, we show that learning commonsense knowledge from multiple modalities effectively improves the performance of downstream vision and language tasks, improves interpretability of the model and is able to make more efficient use of training data. Complementary to the model aspect, we also study the data aspect of commonsense learning in vision and language. We study active learning for Visual Question Answering (VQA) where a model iteratively grows its knowledge through querying informative questions about images for answers. Drawing analogies from human learning, we explore cramming (entropy), curiosity-driven (expected model change), and goal-driven (expected error reduction) active learning approaches, and propose a new goal-driven scoring function for deep VQA models under the Bayesian Neural Network framework. Once trained with a large initial training set, a deep VQA model is able to efficiently query informative question-image pairs for answers to improve itself through active learning, saving human effort on commonsense annotations. / Ph. D. / Designing systems that learn and reason with common sense is a challenging problem in Artificial Intelligence (AI). Humans have the remarkable ability to interpret images and text from different perspectives in multiple modalities, and to use large amounts of commonsense knowledge while performing visual or textual tasks. Inspired by that ability, we approach commonsense learning as leveraging perspectives from multiple modalities for images and text in the context of vision and language tasks. Given a target task, our system first represents the input information (e.g., images and text) in multiple modalities (e.g., vision, text, abstract scenes and facts). Those modalities provide different perspectives to interpret the input information. Based on those perspectives, the system performs reasoning to make a joint prediction to solve the target task. Perhaps surprisingly, we show that imagining (generating) abstract scenes behind input textual scene descriptions improves performance on various textual reasoning tasks such as answering fill-in-the-blank and paraphrasing questions, and answering questions about images improves performance on retrieving image captions. Through the use of perspectives from multiple modalities, our system also makes use of training data more efficiently and has a reasoning process that is easy to understand. Complementary to the system design aspect, we also study the data aspect of commonsense learning in vision and language. We study active learning for Visual Question Answering (VQA). VQA is the task of answering open-ended natural language questions about images. In active learning for VQA, a model iteratively grows its knowledge through querying informative questions about images for answers. Inspired by human learning, we explore cramming (entropy), curiosity-driven (expected model change), and goal-driven (expected error reduction) active learning approaches, and propose a new goal-driven query selection function. We show that once initialized with a large training set, a VQA model is able to efficiently query informative question-image pairs for answers to improve itself through active learning, saving human effort on commonsense annotations.

Die Supply Chain von Algen in Indonesien, insbesondere Karrageen und Agarprodukte, ist mit Störungen innerhalb der Unternehmen und in externen Netzwerken konfrontiert wie z. B. Quantität, Qualität, Preis und Infrastruktur. Sobald eine Störung in einem Teil der Supply Chain auftritt, kann diese die gesamte Wertschöpfungskette beeinflussen. Um eine nachhaltige Algenindustrie sicher zu stellen, bedarf es einer langzeitigen Planung. Hierbei kann das Supply Chain Risk Management (SCRM) helfen. SCRM umfasst vier kritische Stufen: Identifikation der Algen-Supply Chains, Identifikation und Kategorisierung der Risiken, Einschätzen der Risiken und Vermeidung der Risiken. Um die Algen-Supply Chains zu identifizieren, wurden Feldforschung, Tiefeninterviews und Literaturrecherchen durchgeführt. Die Feldstudie wurde in den Provinzen Süd-Sulawesi, West-Java, Ost-Java, Banten und West-Nusa Tenggara durchgeführt. Anschließend wurden die Algen-Supply Chains mit der Software Umberto NXT Universal 7.0 modelliert, um ein besseres Verständnis von den Material- und Energieflüssen zwischen den Hauptakteuren zu erhalten. Um die Risiken zu identifizieren und zu kategorisieren wurden Literaturrecherchen durchgeführt und die Delphi Methode angewandt, um potentielle Quellen der Risiken, ihre Gründe und ihre Effekte zu analysieren. Zur Einschätzung der Risiken wurde ein semi-quantitativer Ansatz gewählt, welcher auf Face-to-face Interviews zurückgreift. Dementsprechend wurde eine Risikokarte erstellt, welche die Wahrscheinlichkeiten und Effekte von ungünstigen Ereignissen widerspiegelt. Basierend auf den vorherigen Ergebnissen wurde die Intensität der Risiken in die folgenden fünf Kategorien unterteilt: vernachlässigbare, geringe, kritische, sehr kritische und katastrophale Risiken. Drei alternative Strategien zur Vermeidung der Risiken werden vorgeschlagen: der Bau einer kleinen Algen-Fabrik, große Algen-Herstellung und industrielle Algen-Cluster, welche direkt am Algenanbau angegliedert sein müssen. Um die Strategien zu bewerten wurde die Preference Ranking Organization Method for Enrichment Evaluation (PROMETHEE) angewandt, welche der multikriteriellen Entscheidungsanalyse zuzurechnen ist. Die Vermeidungsstrategien berücksichtigen  Nachhaltigkeitskriterien (Ökologie, Wirtschaft, Soziales) und Risikokriterien. Die Ergebnisse zeigen, dass Algen-Supply Chains mit vertikaler Kooperation, aus Algen-Lieferanten (Farmer, lokale Händler, Großhändler und Exporteure) und Algen-Herstellern (Carrageen- und Agarunternehmen) bestehen. Die meisten Algen-Farmer sind unabhängige Farmer, die die Möglichkeit haben die Algen entweder an lokale Händler oder andere je nach Präferenz verkaufen können. Lokale Händler spielen eine entscheidende Rolle hinsichtlich der finanziellen Unterstützung der Farmer, der technischen Informationen und des Marktzugangs. Indonesische Carrageen- und Agarhersteller weisen ein kontinuierliches Marktwachstum auf, insbesondere als Familienunternehmen. Die Risiken der Algen-Supply Chain werden in zwei Hauptkategorien unterteilt: interne und externe Risiken. Interne Risiken werden weiterhin in zwei Gruppen untergliedert: interne Unternehmensrisiken, welche aus Prozess- und Steuerungsrisiken bestehen, sowie externe Unternehmensrisiken, die Supply- und Nachfragerisiken umfassen. Externe Supply Chain Risiken, die aus externen Netzwerkketten resultieren, sind Risiken hinsichtlich regulatorischer Fragen, der Finanzierung, Infrastruktur als auch soziale und umweltbezogene Risiken. Als kritischstes Risiko für die Carrageen-Supply Chain wurde die mangelnde Qualität von E.cottonii rohgetrockneten Algen identifiziert. Weitere kritische Risiken der Agar-Supply Chain sind Quantitätsunsicherheiten, die Knappheit von Gracilaria roh getrockneten Algen sowie negative Umweltauswirkungen des Abwassers. Aus  der Fallstudie zu halb-raffinierten Karrageen resultierte das Clustern von Algen-Verarbeitungsbetrieben als ist die beste Risikovermeidungsstrategie sowohl für große als auch kleine Unternehmen. Nichtsdestotrotz kann sich das Clustern negativ auf die lokale Umwelt auswirken.

Diese Dissertation besteht aus fünf empirischen Studien. Zwei Studien befassen sich mit passiv gemanagten Finanzprodukten. Sie untersuchen den Replikationsprozess von Exchange Traded Funds (ETFs) und vergleichen hierbei die Replikationsgüte von synthetischen und physischen ETFs. Oftmals wird darauf verwiesen, dass synthetische ETFs eine höhere Replikationsgüte besitzen als physische ETFs. Dies lässt sich für Renten-ETFs bestätigen, allerdings nicht für Aktien-ETFs. Zudem wird gezeigt, dass ETFs und Indexzertifikate, die sich im direkten Wettbewerb befinden, im Hinblick auf ihre Geldmittelflüsse komplementär, allerdings nicht perfekt komplementär zueinander sind. Dieser Effekt lässt sich mithilfe der Replikationsgüte und einer Zuordnung beider Indexprodukte in verschiedene Marktnischen erklären. Weitere drei Studien befassen sich mit aktiv gemanagten Finanzprodukten. Sie widmen sich der Frage, ob mithilfe von Fondsbewertungen wie dem Feri Trust Rating, der Finanztest-Bewertung und der FondsNote die zukünftige Performance deutscher Aktienfonds prognostiziert werden kann. Hintergrund ist, dass Investoren Fondsbewertungen in ihre Anlageentscheidung einbeziehen. Sie investieren vor allem in Fonds, die eine Top-Bewertung aufweisen. Die Prognosefähigkeit von Fondsbewertungen kann sich allerdings stark voneinander unterscheiden. Die Ergebnisse zeigen, dass mithilfe der FondsNote am besten zwischen sich zukünftig besser und schlechter entwickelnden Fonds differenziert werden kann. Die Prognosefähigkeit lässt sich durch Kombination der drei Fondsbewertungen sogar erhöhen. Dies hängt allerdings von der Kombination und dem verwendeten Performancemaß/-zeitraum ab. Zudem werden Faktoren untersucht, die einen Einfluss auf die Prognosefähigkeit haben können. Es wird gezeigt, dass qualitative Bewertungsfaktoren nicht zu einer Erhöhung der Prognosefähigkeit beitragen. Stattdessen weisen die Fondskosten und das Verhalten der Investoren einen signifikanten Einfluss auf. / This thesis consists of five empirical studies that deal with actively and passively managed financial products. The first two studies focus on the replication process of exchange-traded funds (ETFs) and compare the tracking ability of ETFs based on physical replication of their benchmark indices with those of synthetic ETFs. Contrary to conventional wisdom, synthetic equity ETFs are not different in terms of tracking errors from their physical counterparts. However, synthetic fixed-income ETFs have lower tracking errors than physical fixed-income ones. Moreover, the second study examines the coexistence of ETFs and index certificates within one market by analyzing the relationship between their money flows. Evidence shows that ETFs and index certificates complement each other, but not in a perfect way. This effect can be explained by similar tracking abilities and a segmentation of investors into different market niches. The other three studies address the question if fund ratings like the Feri Trust rating, the Finanztest-Bewertung, and the FondsNote can predict the future performance of German equity mutual funds. The reason is that investors include fund ratings in their decision-making. They primarily invest in funds which have the best fund rating. However, fund rating predictability can significantly differ among fund ratings. Results indicate that the FondsNote can best distinguish between well and poorly performing funds. Predictability can be enhanced by a combination of fund ratings. However, it depends on the particular fund rating combination, the chosen performance measure, and the post-rating period. Moreover, these three studies analyze factors that could influence the predictability of fund ratings. It is shown that qualitative factors can hardly improve the predictability. By contrast, the costs of funds and the behavior of investors with regard to fund ratings significantly influence the ability to predict future performance.

Made available in DSpace on 2015-03-05T19:14:41Z (GMT). No. of bitstreams: 0 Previous issue date: 29 / Nenhuma / O objetivo do estudo é a avaliação do desempenho econômico e financeiro de empresas calçadistas brasileiras. O estudo contextualiza-se no ambiente caracterizado a partir dos anos 1980, quando o mercado competitivo ultrapassou as fronteiras domésticas, exigindo das empresas maior eficiência na gestão de suas atividades, tendo na contabilidade uma das fontes de informações necessárias para orientar a adequada utilização dos recursos. A pesquisa se caracteriza como de natureza exploratória, quantitativa e documental. Para o desenvolvimento do estudo coletaram-se demonstrações contábeis correspondentes aos anos de 2000 a 2006 e relativas a 15 indústrias do setor calçadista brasileiro localizadas nas regiões Sul, Sudeste e Nordeste. O estudo, com base em 30 indicadores selecionados, possibilitou a análise do conjunto das empresas, por meio do índice-padrão de cada indicador. Em seguida, efetuou-se a análise dos indicadores por empresa. Visando testar o significado dos resultados da análise, foram aplicadas as técn / The objective of this study is to evaluate the economic and financial performance of the Footwear Industry from the 1980’s on, when the competitive market crossed the domestic frontier, demanding more efficiency from shoemaking companies regarding the management of their activities. Within this context, accountancy is seen as one of the necessary information sources to guide the proper use of resources. The study is exploratory, quantitative and documental in nature. Statements of financial accounting from 15 Brazilian shoemaking companies located in the South, Southeast and Northeast regions were collected from the year 2000 to 2006. The study was based on 30 selected indexes, which made possible to analyze the group of companies by the standard-index of each indicator. Following that, the index analysis of each company was made. In order to check the analysis results, typical value techniques were used as a comparative measure of the standard indexes and the statistical correlation tests. The companies ran

Submitted by Marcelo Bonança (marcelo.bonanca@vale.com) on 2014-10-31T17:02:59Z No. of bitstreams: 1 Dissertação_Marcelo Bonança.pdf: 792001 bytes, checksum: 5a68a41e5ddebd68c372493c95f262a8 (MD5) / Approved for entry into archive by Vitor Souza (vitor.souza@fgv.br) on 2014-11-25T17:15:41Z (GMT) No. of bitstreams: 1 Dissertação_Marcelo Bonança.pdf: 792001 bytes, checksum: 5a68a41e5ddebd68c372493c95f262a8 (MD5) / Approved for entry into archive by Vitor Souza (vitor.souza@fgv.br) on 2014-11-25T17:17:48Z (GMT) No. of bitstreams: 1 Dissertação_Marcelo Bonança.pdf: 792001 bytes, checksum: 5a68a41e5ddebd68c372493c95f262a8 (MD5) / Made available in DSpace on 2014-11-26T16:45:01Z (GMT). No. of bitstreams: 1 Dissertação_Marcelo Bonança.pdf: 792001 bytes, checksum: 5a68a41e5ddebd68c372493c95f262a8 (MD5) Previous issue date: 2014-05-19 / This study analyses the effects of diversification in risk and return among the major global mining players. To evaluate the diversification, firstly, we created a ranking that relates risk, degree of diversification and return of the companies. It was mandatory to create a data base with a series of revenues and proxies of cash generation by business, which is not fully provided by Bloomberg, Economática or other public data bases. We estimated the effects of diversification by using the CAPM to calculate the companies’ beta and basic statistics of the historical return series to find the Sharpe Index. We looked through the financial statements to understand more profoundly their businesses for a sample equivalent to 50% of the market capitalization of the top 100 mining companies of the world for the period from 2002 to 2014. Finally, we approached, as a study case, essential issues in the diversification as efficiency and discipline in capital allocation. Our study confirmed the presence of a negative correlation between risk and diversification degree. It also showed evidences which suggest that focused companies have better discipline in capital allocation, and hence, generate more value to their shareholders. / Neste trabalho, medimos os efeitos da diversificação sob o aspecto de risco e retorno entre os grandes players globais de mineração. Para avaliar a diversificação, primeiramente, foi criado um ranking que relaciona risco, grau de diversificação e retorno das empresas. Foi necessário montar um banco de dados com séries de receita e de proxies de geração de caixa por negócios, o qual não se encontra prontamente disponível na Bloomberg, Economática ou em outras bases de dados públicas. Estimamos os efeitos da diversificação fazendo o uso do CAPM para calcular o beta das empresas e das estatísticas básicas das séries históricas de retornos para encontrar o Índice de Sharpe. Examinamos os demonstrativos financeiros das empresas para entender com maior profundidade seus negócios para uma amostra equivalente a 50% do valor de mercado das top 100 empresas de mineração do mundo para o período entre 2002 e 2014. Por fim, abordamos, sob a forma de estudo de caso, questões relevantes na diversificação como a eficiência e a disciplina na alocação de capital. Nosso estudo confirmou que existe uma correlação negativa entre o risco e o grau de diversificação e que há sinais que sugerem que as companhias mais focadas possuem maior disciplina na alocação do capital, e, com isso geram mais valor para seus acionistas.

nÃo hà / This article analyzes the Brazilian stock mutual funds market, proposing dynamic rankings constructed from different risk-return performance metrics during the period from 1998 to 2009. We can evidence an unusual level of persistence, especially among the funds with better performance due to the expertise of managers. The quarterly rebalancing of portfolios based on these rankings allows us to infer that in scenarios characterized by economic boom, the equal weighted strategies in funds winners provide significant increases in average monthly earnings, risk reduction associated with diversification and consequent increase in performance in relation to market or sectoral benchmarks. This evidence is robust to the use of different performance metrics for the selection of funds, signaling that the shareholders of the winner funds require good performance not only in Sharpe, but also on other metrics, such as Treynor, Sortino and Calmar. In these optimistic scenarios, only the index of the industrial sector (INDX) provided values compatible with these portfolios of funds. However, in crisis scenarios, no strategy involving the funds can provide the levels characteristic of the hedging index of the electricity sector (IEE). It is possible to show that the majority of these strategies are dominated by the benchmarks on the riskreturn criteria, except for the value weighted portfolios composed by looser funds, a sign that the common shareholders of the big and passive funds may be experiencing higher levels of inertia. / Este artigo analisa o mercado de fundos de investimentos em aÃÃes no Brasil, propondo rankings dinÃmicos construÃdos a partir de diferentes mÃtricas de performances risco-retorno, durante o perÃodo de 1998 a 2009. à possÃvel evidenciar um nÃvel de persistÃncia incomum, principalmente dentre os fundos com melhor performance devido à expertise dos gestores. O rebalanceamento quadrimestral de portfolios com base nestes rankings permite inferir que em cenÃrios caracterizados por boom econÃmico ou recuperaÃÃo dos mercados financeiros, as estratÃgias com participaÃÃes iguais em fundos winners proporcionam aumentos significativos de ganhos mÃdios mensais, reduÃÃo de risco associado à diversificaÃÃo e consequente aumento de performance em relaÃÃo a benchmarks de mercado ou setoriais. Esta evidÃncia à robusta ao uso de diferentes mÃtricas de performance para seleÃÃo dos fundos, sinalizando que os cotistas dos fundos winners ativos exigem boa performance nÃo somente em Sharpe, mas tambÃm em outras mÃtricas, tais como os Ãndices de Treynor, Calmar e Sortino. Nestes cenÃrios otimistas, apenas o Ãndice do setor industrial (INDX) proporcionou valores compatÃveis aos destes portfolios de fundos. PorÃm, em cenÃrios de crise, nenhuma estratÃgia envolvendo os fundos consegue proporcionar os nÃveis de hedge caracterÃsticos do Ãndice do setor de energia elÃtrica (IEE), sendo possÃvel evidenciar que a maioria destas estratÃgias à dominada nos critÃrios de ganho-risco por Ãndices setorias ou de mercado, com exceÃÃo dos portfolios value weighted compostos por fundos loosers, um sinal de que os cotistas usuais dos grandes fundos passivos e indexados ao Ibovespa podem estar apresentando maior nÃvel de inÃrcia.

On s'attache dans cette thèse à plusieurs aspects liés à la résolution de problèmes multi-objectifs, sans se limiter au cas biobjectif. Nous considérons la résolution exacte, dans le sens de la détermination de l'ensemble des points non dominés, ainsi que la résolution approchée dans laquelle on cherche une approximation de cet ensemble dont la qualité est garantie a priori.Nous nous intéressons d'abord au problème de la détermination d'une représentation explicite de la région de recherche. La région de recherche, étant donné un ensemble de points réalisables connus, exclut la partie de l'espace des objectifs que dominent ces points et constitue donc la partie de l'espace des objectifs où les efforts futurs doivent être concentrés dans la perspective de déterminer tous les points non dominés.Puis nous considérons le recours aux algorithmes de séparation et évaluation ainsi qu'aux algorithmes de ranking afin de proposer une nouvelle méthode hybride de détermination de l'ensemble des points non dominés. Nous montrons que celle-ci peut également servir à obtenir une approximation de l'ensemble des points non dominés. Cette méthode est implantée pour le problème de l'arbre couvrant de poids minimal. Les quelques propriétés de ce problème que nous passons en revue nous permettent de spécialiser certaines procédures et d'intégrer des prétraitements spécifiques. L'intérêt de cette approche est alors soutenu à l'aide de résultats expérimentaux. / This thesis deals with several aspects related to solving multi-objective problems, without restriction to the bi-objective case. We consider exact solving, which generates the nondominated set, and approximate solving, which computes an approximation of the nondominated set with a priori guarantee on the quality.We first consider the determination of an explicit representation of the search region. The search region, defined with respect to a set of known feasible points, excludes from the objective space the part which is dominated by these points. Future efforts to find all nondominated points should therefore be concentrated on the search region.Then we review branch and bound and ranking algorithms and we propose a new hybrid approach for the determination of the nondominated set. We show how the proposed method can be adapted to generate an approximation of the nondominated set. This approach is instantiated on the minimum spanning tree problem. We review several properties of this problem which enable us to specialize some procedures of the proposed approach and integrate specific preprocessing rules. This approach is finally supported through experimental results.

La première partie de cette thèse introduit de nouveaux algorithmes de décomposition parcimonieuse de signaux. Basés sur Matching Pursuit (MP) ils répondent au problème suivant : comment réduire le temps de calcul de l'étape de sélection de MP, souvent très coûteuse. En réponse, nous sous-échantillonnons le dictionnaire à chaque itération, en lignes et en colonnes. Nous montrons que cette approche fondée théoriquement affiche de bons résultats en pratique. Nous proposons ensuite un algorithme itératif de descente de gradient par blocs de coordonnées pour sélectionner des caractéristiques en classification multi-classes. Celui-ci s'appuie sur l'utilisation de codes correcteurs d'erreurs transformant le problème en un problème de représentation parcimonieuse simultanée de signaux. La deuxième partie expose de nouvelles inégalités de concentration empiriques de type Bernstein. En premier, elles concernent la théorie des U-statistiques et sont utilisées pour élaborer des bornes en généralisation dans le cadre d'algorithmes de ranking. Ces bornes tirent parti d'un estimateur de variance pour lequel nous proposons un algorithme de calcul efficace. Ensuite, nous présentons une version empirique de l'inégalité de type Bernstein proposée par Freedman [1975] pour les martingales. Ici encore, la force de notre borne réside dans l'introduction d'un estimateur de variance calculable à partir des données. Cela nous permet de proposer des bornes en généralisation pour l'ensemble des algorithmes d'apprentissage en ligne améliorant l'état de l'art et ouvrant la porte à une nouvelle famille d'algorithmes d'apprentissage tirant parti de cette information empirique. / The first part of this thesis introduces new algorithms for the sparse encoding of signals. Based on Matching Pursuit (MP) they focus on the following problem : how to reduce the computation time of the selection step of MP. As an answer, we sub-sample the dictionary in line and column at each iteration. We show that this theoretically grounded approach has good empirical performances. We then propose a bloc coordinate gradient descent algorithm for feature selection problems in the multiclass classification setting. Thanks to the use of error-correcting output codes, this task can be seen as a simultaneous sparse encoding of signals problem. The second part exposes new empirical Bernstein inequalities. Firstly, they concern the theory of the U-Statistics and are applied in order to design generalization bounds for ranking algorithms. These bounds take advantage of a variance estimator and we propose an efficient algorithm to compute it. Then, we present an empirical version of the Bernstein type inequality for martingales by Freedman [1975]. Again, the strength of our result lies in the variance estimator computable from the data. This allows us to propose generalization bounds for online learning algorithms which improve the state of the art and pave the way to a new family of learning algorithms taking advantage of this empirical information.