Submitted by Geyciane Santos (geyciane_thamires@hotmail.com) on 2015-11-12T20:17:58Z No. of bitstreams: 1 Tese - Márcia Henke.pdf: 2984974 bytes, checksum: a103355c1a7895956d40d4fa9422347a (MD5) / Approved for entry into archive by Divisão de Documentação/BC Biblioteca Central (ddbc@ufam.edu.br) on 2015-11-16T18:36:36Z (GMT) No. of bitstreams: 1 Tese - Márcia Henke.pdf: 2984974 bytes, checksum: a103355c1a7895956d40d4fa9422347a (MD5) / Approved for entry into archive by Divisão de Documentação/BC Biblioteca Central (ddbc@ufam.edu.br) on 2015-11-16T18:43:03Z (GMT) No. of bitstreams: 1 Tese - Márcia Henke.pdf: 2984974 bytes, checksum: a103355c1a7895956d40d4fa9422347a (MD5) / Made available in DSpace on 2015-11-16T18:43:03Z (GMT). No. of bitstreams: 1 Tese - Márcia Henke.pdf: 2984974 bytes, checksum: a103355c1a7895956d40d4fa9422347a (MD5) Previous issue date: 2015-03-30 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Electronic messages (emails) are still considered the most significant tools in business and personal applications due to their low cost and easy access. However, e-mails have become a major problem owing to the high amount of junk mail, named spam, which fill the e-mail boxes of users. Among the many problems caused by spam messages, we may highlight the fact that it is currently the main vector for the spread of malicious activities such as viruses, worms, trojans, phishing, botnets, among others. Such activities allow the attacker to have illegal access to penetrating data, trade secrets or to invade the privacy of the sufferers to get some advantage. Several approaches have been proposed to prevent sending unsolicited e-mail messages, such as filters implemented in e-mail servers, spam message classification mechanisms for users to define when particular issue or author is a source of spread of spam and even filters implemented in network electronics. In general, e-mail filter approaches are based on analysis of message content to determine whether or not a message is spam. A major problem with this approach is spam detection in the presence of concept drift. The literature defines concept drift as changes occurring in the concept of data over time, as the change in the features that describe an attack or occurrence of new features. Numerous Intrusion Detection Systems (IDS) use machine learning techniques to monitor the classification error rate in order to detect change. However, when detection occurs, some damage has been caused to the system, a fact that requires updating the classification process and the system operator intervention. To overcome the problems mentioned above, this work proposes a new changing detection method, named Method oriented to the Analysis of the Development of Attacks Characteristics (MECA). The proposed method consists of three steps: 1) classification model training; 2) concept drift detection; and 3) transfer learning. The first step generates classification models as it is commonly conducted in machine learning. The second step introduces two new strategies to avoid concept drift: HFS (Historical-based Features Selection) that analyzes the evolution of the features based on over time historical; and SFS (Similarity-based Features Selection) that analyzes the evolution of the features from the level of similarity obtained between the features vectors of the source and target domains. Finally, the third step focuses on the following questions: what, how and when to transfer acquired knowledge. The answer to the first question is provided by the concept drift detection strategies that identify the new features and store them to be transferred. To answer the second question, the feature representation transfer approach is employed. Finally, the transfer of new knowledge is executed as soon as changes that compromise the classification task performance are identified. The proposed method was developed and validated using two public databases, being one of the datasets built along this thesis. The results of the experiments shown that it is possible to infer a threshold to detect changes in order to ensure the classification model is updated through knowledge transfer. In addition, MECA architecture is able to perform the classification task, as well as the concept drift detection, as two parallel and independent tasks. Finally, MECA uses SVM machine learning algorithm (Support Vector Machines), which is less adherent to the training samples. The results obtained with MECA showed that it is possible to detect changes through feature evolution monitoring before a significant degradation in classification models is achieved. / As mensagens eletrônicas (e-mails) ainda são consideradas as ferramentas de maior prestígio no meio empresarial e pessoal, pois apresentam baixo custo e facilidade de acesso. Por outro lado, os e-mails tornaram-se um grande problema devido à elevada quantidade de mensagens não desejadas, denominadas spam, que lotam as caixas de emails dos usuários. Dentre os diversos problemas causados pelas mensagens spam, destaca-se o fato de ser atualmente o principal vetor de propagação de atividades maliciosas como vírus, worms, cavalos de Tróia, phishing, botnets, dentre outros. Tais atividades permitem ao atacante acesso indevido a dados sigilosos, segredos de negócios ou mesmo invadir a privacidade das vítimas para obter alguma vantagem. Diversas abordagens, comerciais e acadêmicas, têm sido propostas para impedir o envio de mensagens de e-mails indesejados como filtros implementados nos servidores de e-mail, mecanismos de classificação de mensagens de spam para que os usuários definam quando determinado assunto ou autor é fonte de propagação de spam e até mesmo filtros implementados em componentes eletrônicos de rede. Em geral, as abordagens de filtros de e-mail são baseadas na análise do conteúdo das mensagens para determinar se tal mensagem é ou não um spam. Um dos maiores problemas com essa abordagem é a deteção de spam na presença de concept drift. A literatura conceitua concept drift como mudanças que ocorrem no conceito dos dados ao longo do tempo como a alteração das características que descrevem um ataque ou ocorrência de novas características. Muitos Sistemas de Deteção de Intrusão (IDS) usam técnicas de aprendizagem de máquina para monitorar a taxa de erro de classificação no intuito de detetar mudança. Entretanto, quando a deteção ocorre, algum dano já foi causado ao sistema, fato que requer atualização do processo de classificação e a intervenção do operador do sistema. Com o objetivo de minimizar os problemas mencionados acima, esta tese propõe um método de deteção de mudança, denominado Método orientado à Análise da Evolução das Características de Ataques (MECA). O método proposto é composto por três etapas: 1) treino do modelo de classificação; 2) deteção de mudança; e 3) transferência do aprendizado. A primeira etapa emprega modelos de classificação comumente adotados em qualquer método que utiliza aprendizagem de máquina. A segunda etapa apresenta duas novas estratégias para contornar concept drift: HFS (Historical-based Features Selection) que analisa a evolução das características com base no histórico ao longo do tempo; e SFS (Similarity based Features Selection) que observa a evolução das características a partir do nível de similaridade obtido entre os vetores de características dos domínios fonte e alvo. Por fim, a terceira etapa concentra seu objetivo nas seguintes questões: o que, como e quando transferir conhecimento adquirido. A resposta à primeira questão é fornecida pelas estratégias de deteção de mudança, que identificam as novas características e as armazenam para que sejam transferidas. Para responder a segunda questão, a abordagem de transferência de representação de características é adotada. Finalmente, a transferência do novo conhecimento é realizada tão logo mudanças que comprometam o desempenho da tarefa de classificação sejam identificadas. O método MECA foi desenvolvido e validado usando duas bases de dados públicas, sendo que uma das bases foi construída ao longo desta tese. Os resultados dos experimentos indicaram que é possível inferir um limiar para detetar mudanças a fim de garantir o modelo de classificação sempre atualizado por meio da transferência de conhecimento. Além disso, um diferencial apresentado no método MECA é a possibilidade de executar a tarefa de classificação em paralelo com a deteção de mudança, sendo as duas tarefas independentes. Por fim, o MECA utiliza o algoritmo de aprendizagem de máquina SVM (Support Vector Machines), que é menos aderente às amostras de treinamento. Os resultados obtidos com o MECA mostraram que é possível detetar mudanças por meio da evolução das características antes de ocorrer uma degradação significativa no modelo de classificação utilizado.

Submitted by Regis Albuquerque (regis.albuquerque1@gmail.com) on 2018-06-20T21:40:28Z No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) dissertacao_regis_corrigida_final.pdf: 2557634 bytes, checksum: b48eb7c37fd9dd633c4489a7f0f041a4 (MD5) / Approved for entry into archive by Secretaria PPGI (secretariappgi@icomp.ufam.edu.br) on 2018-06-20T21:52:37Z (GMT) No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) dissertacao_regis_corrigida_final.pdf: 2557634 bytes, checksum: b48eb7c37fd9dd633c4489a7f0f041a4 (MD5) / Approved for entry into archive by Divisão de Documentação/BC Biblioteca Central (ddbc@ufam.edu.br) on 2018-06-21T13:29:00Z (GMT) No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) dissertacao_regis_corrigida_final.pdf: 2557634 bytes, checksum: b48eb7c37fd9dd633c4489a7f0f041a4 (MD5) / Made available in DSpace on 2018-06-21T13:29:01Z (GMT). No. of bitstreams: 2 license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) dissertacao_regis_corrigida_final.pdf: 2557634 bytes, checksum: b48eb7c37fd9dd633c4489a7f0f041a4 (MD5) Previous issue date: 2018-06-08 / FAPEAM - Fundação de Amparo à Pesquisa do Estado do Amazonas / Many machine learning applications have to deal with classification problems in dynamic environments. This type of environment may be affected by concept drift, which may reduce the accuracy of classification systems significantly. In this context, methods using ensemble of classifiers are interesting due to the fact that ensembles of classifiers allow the design of strategies for drift detection and reaction more accurate and robust to changes. A classification system based on ensemble of classifiers may be divided into three main phases: classifier generation; single classifier or subset of classifier selection; and classifier fusion. The selection phase may be performed as a dynamic process. In this case, for each unknown sample, the individual classifier or classifier ensemble most likely to be correct is chosen to assign a label to the sample. In this work, it is proposed a method for concept drift detection and reaction based on dynamic classifier ensemble selection. The proposed method choses the expert classifier ensemble according to diversity and accuracy values. Focusing on evaluating the impact of dynamic ensemble selection guided by diversity and accuracy in terms of concept drift detection and reaction, four series of experiments were carried in this work using both synthetic and real datasets. In addition, since the proposed method is broken down into four phases: pool of ensemble classifiers generation; dynamic ensemble selection; drift detection; and drift reaction, different versions of the proposed method were investigated by varying the parameters of each phase. The results show that, in general, all these different versions attain very similar accuracy values. Besides, when compared to two baselines: (1) DDM - single classifier-based; and (2) Leveraging Bagging - classifier ensemble-based, our method outperforms both baselines since it achieved higher accuracy, lower detection delay and false detection rates, and it did not present missing detection. However, both baselines present lower time complexity. Therefore, this work shows that dynamic classifier ensemble selection guided by diversity and accuracy helps to improve detection precision and the general accuracy of classification systems employed in problems with concept drift. / Muitas aplicações de aprendizado de máquina estão relacionadas com problemas de classificação em ambientes dinâmicos. Mudança de conceito figura nesse tipo de ambiente e pode prejudicar muito a acurácia de sistemas de classificação. Nesse contexto, a utilização de comitês de classificadores é interessante porque possibilita a implementação de processos de detecção e de reação à mudança mais acurados e robustos. Sistemas de classificação que utilizam comitês podem possuir três grandes fases: geração; seleção; e integração de classificadores. A etapa de seleção pode ser feita de forma dinâmica, isto é, para cada instância desconhecida, o classificador ou comitê de classificadores com maior probabilidade de acerto é escolhido para atribuir uma classe à essa instância. Neste trabalho, é proposto um método para detecção e reação à mudança de conceito que utiliza seleção dinâmica de comitês de classificadores. O método proposto escolhe o comitê especialista com base nos valores de diversidade e de acurácia de cada comitê candidato. A fim de avaliar o impacto do uso de seleção dinâmica guiada por diversidade e acurácia nas tarefas de detecção e reação a mudança de conceito, foram realizadas quatro séries de experimentos com bases sintéticas e reais. Além disso, como o método proposto é dividido em quatro fases: geração da população de comitês; seleção dinâmica do comitê especialista; detecção de mudanças; e reação à mudança, diferentes versões desse método foram investigadas em função da definição de parâmetros de cada fase. Os resultados dos experimentos mostraram que, de maneira geral, as versões estudadas são bem equivalentes em termos de acurácia média final. Adicionalmente, quando comparado a dois baselines: (1) DDM - que utiliza um único classificador; e (2) Leveraging Bagging - que utiliza um comitê de classificadores, o método proposto alcançou melhores taxas de acurácia, menores taxas de atraso de detecção, não deixou de detectar as mudanças conhecidas nas bases e produziu reduzidas taxas de falsa detecção, apesar de apresentar maior complexidade computacional. Portanto, o trabalho mostra que o uso de seleção dinâmica guiada por diversidade e acurácia melhora a precisão de detecção, bem como a acurácia geral de sistemas de classificação utilizados em problemas que apresentam mudança de conceitos.

Um dos problemas enfrentados pelos cientistas para o confinamento de plasma em Tokamaks, para se obter fusão termonuclear controlada, é o transporte radial de partículas pela borda do plasma. Nessa dissertação, estudamos o transporte através de um modelo que relaciona as flutuações eletrostáticas na borda do plasma às ondas de deriva. Essas ondas criam no plasma regiões de fluxo convectivo, formando ilhas que são, eventualmente, separadas por barreiras. Para apenas uma onda, o sistema é integrável e todas as trajetórias do plano de fase são curvas invariantes que, se não existirem barreiras, estão em ilhas divididas por separatrizes. Foi verificado que, quando uma segunda onda com velocidade de fase diferente da primeira é utilizada, o sistema não é mais integrável e a região anteriormente ocupada pelas separatrizes torna-se caótica. Com a quebra de separatrizes ocorre o transporte caótico de partículas. Quando uma separatriz é quebrada, surge em seu lugar uma estrutura que ainda preserva algumas características da separatriz, mas se modifica no espaço de fases ao longo do tempo. Essa estrutura é conhecida como Estrutura Lagrangiana Coerente (ELC). Nessa dissertação verificamos que as ELCs, por um lado, funcionam como barreiras de transporte, pois nenhuma trajetória consegue atravessa-la e, por outro lado, criam regiões no espaço de fases onde o transporte é alto, pois trajetórias próximas a elas tendem a ser aceleradas. Uma das principais contribuições obtidas ao se estudar ELCs no problema de duas ondas de deriva, aplicado ao confinamento de plasmas em Tokamaks, é a possibilidade de se prever a existência de ilhas, que funcionem como barreiras de transporte, no plano de fases que, por sua vez, são um importante mecanismo de aprisionamento de partículas. / One of the problems facing scientists in the confinement of plasma in tokamaks, to obtain controlled thermonuclear fusion, is the radial transport of particles at the plasma edge. In this dissertation, we study particle transport through a model that relates the electrostatic fluctuations at the edge of the plasma with drift waves. These waves create regions inside the plasma with convective flow, forming islands that are eventually separated by barriers. For one wave, the system is integrable and all the trajectories of phase space are invariant curves that are divided by separatrices. It was found that when a second wave with phase velocity different from the first is used, the system is no longer integrable and the region previously occupied by the separatrix becomes chaotic. With the destruction of the separatrix the transport of particles is chaotic. When a separatrix is broken, appears in its place a structure that preserves some features of the separatrix, but it is changing in phase space over time. This structure is known as Lagrangian Coherent Structure (LCS). In this dissertation we found that the LCSs, on the one hand, act as transport barriers, since no trajectory can cross it and, moreover, creates regions in phase space where particle transport is high, because trajectories close to them tend to be accelerated. One of the main contributions obtained by studying LCSs in the problem of two drift waves, applied to the confinement of plasma in tokamaks, is the ability to predict the existence of islands, which act as transport barriers, which are an important mechanism of trapping particles.

Aprender conceitos provenientes de fluxos de dados é uma tarefa significamente diferente do aprendizado tradicional em lote. No aprendizado em lote, existe uma premissa implicita que os conceitos a serem aprendidos são estáticos e não evoluem significamente com o tempo. Por outro lado, em fluxos de dados os conceitos a serem aprendidos podem evoluir ao longo do tempo. Esta evolução é chamada de mudança de conceito, e torna a criação de um conjunto fixo de treinamento inaplicável neste cenário. O aprendizado incremental é uma abordagem promissora para trabalhar com fluxos de dados. Contudo, na presença de mudanças de conceito, conceitos desatualizados podem causar erros na classificação de eventos. Apesar de alguns métodos incrementais baseados no modelo de misturas gaussianas terem sido propostos na literatura, nota-se que tais algoritmos não possuem uma política explicita de descarte de conceitos obsoletos. Nesse trabalho um novo algoritmo incremental para fluxos de dados com mudanças de conceito baseado no modelo de misturas gaussianas é proposto. O método proposto é comparado com vários algoritmos amplamente utilizados na literatura, e os resultados mostram que o algoritmo proposto é competitivo com os demais em vários cenários, superando-os em alguns casos. / Learning concepts from data streams differs significantly from traditional batch learning. In batch learning there is an implicit assumption that the concept to be learned is static and does not evolve significantly over time. On the other hand, in data stream learning the concepts to be learned may evolve over time. This evolution is called concept drift, and makes the creation of a fixed training set be no longer applicable. Incremental learning paradigm is a promising approach for learning in a data stream setting. However, in the presence of concept drifts, out dated concepts can cause misclassifications. Several incremental Gaussian mixture models methods have been proposed in the literature, but these algorithms lack an explicit policy to discard outdated concepts. In this work, a new incremental algorithm for data stream with concept drifts based on Gaussian Mixture Models is proposed. The proposed methodis compared to various algorithms widely used in the literature, and the results show that it is competitive with them invarious scenarios, overcoming them in some cases.

Longshore sediment transport rate was measured during energetic sea breeze activity, on intermediate-to-reflective sandy beaches in Southwestern Australia. Estimates of suspended load were obtained using backscatter sensors, current meters and streamer traps. Total load was determined using fluorescent tracer sand and an impoundment study. The measurementsw ere cross-compareda nd usedt o evaluates everalw idely-used longshore transport equations. The streamer trap measurement revealed an exponential distribution of the suspended sediment flux with vertical mixing decreasing in the onshore direction. A continuous time series of the longshore suspended sediment flux across the surf zone was obtained by combining the streamer trap measurements with data collected using surf zone instruments. Comparison of the suspended longshore flux with the total longshore flux derived from the dispersal of the sand tracer indicated that the relative contribution of the suspendedlo ad to the total load was at least 59 %. The movement of sandt racer on four different beaches demonstrated that nearshore sediments were transported obliquely across the surf zone, challenging our conventional view of dividing nearshore sediment transport into cross-shore and longshore components. Furthermore, tracer was found to move from the outer surf zone to the swash zone and vice versa, indicating a cross-shore sediment exchange. The contribution of the swash zone to the total longshore flux was estimated around 30-40 %. Despite large differences in the temporal and spatial scales of the measurement techniques, the littoral drift rates are comparable, suggesting a northward transport rate of 138,000-200,000 m3 year-1. Longshore sediment transport during sea breezes is mainly the result of a high longshore energy flux exerted by wind waves. This is accurately predicted by the equations of Inman and Bagnold (1963) and CERC (1984). The bimodal wave field, characteristic of Southwestern Australia, renders the Kamphuis (1991b) formula unsuitable in this instance.

Le transport des particules et de la chaleur dans la zone de bord des tokamaks joue un rôle déterminant à la fois sur les performances du plasma confiné et sur l’extraction de la puissance et ainsi la durée de vie des composants face au plasma. C’est dans ce contexte que s’inscrit ce travail de thèse, qui porte sur le rôle joué par les écoulements transverses au champ magnétique dans l’équilibre entre dynamique parallèle et dynamique perpendiculaire qui régit la région périphérique d’un tokamak. Ces écoulements peuvent produire des asymétries poloïdales du dépôt de chaleur et de particules sur les composants face au plasma, et plus généralement des asymétries des diverses quantités dans le plasma. Les vitesses de dérive radiale sont d’origine électrique (liées à la présence d’un champ électrique radial résultant de l’équilibre des charges) ou liées aux effets de la géométrie toroïdale induisant une inhomogénéité du champ magnétique (vitesse de gradient-courbure). Pour progresser dans la compréhension de ces phénomènes, la modélisation numérique du transport et de la turbulence en géométrie complexe est indispensable. En complément, des outils de diagnostic synthétique permettant de modéliser les processus de mesure dans les plasmas numériques sont développés pour permettre une comparaison réaliste entre modèles et expériences. La modélisation des vitesses de dérive perpendiculaire a été introduite dans le code SOLEDGE2D décrivant le transport de la densité, quantité de mouvement et énergie d’un plasma de tokamak. Nous avons d’abord étudié l’impact d’un champ électrique prescrit sur les équilibres plasma, pour comprendre les mécanismes à l’origine des asymétries du plasma et étudier l’établissement d’écoulement parallèle et d’asymétrie du dépôt de chaleur sur les composants face au plasma. Nous avons ensuite implémenté un modèle auto-consistant de résolution du potentiel électrique dans les équations fluides de SOLEDGE2D afin de comprendre l’équilibre du champ électrique et d’étudier l’effet de la configuration magnétique du tokamak et de la vitesse de gradient-courbure sur ce dernier. Dans la deuxième partie de cette thèse, un diagnostic synthétique permettant de modéliser les mesures expérimentales de rétro-diffusion Doppler a été développé et testé en vue d’être appliqué aux simulations du code fluide 3D turbulent, TOKAM3X. Ce diagnostic permet de mesurer la vitesse perpendiculaire du plasma à partir du mouvement des fluctuations de densité. Il a été utilisé ici pour comparer les asymétries de vitesse observées expérimentalement aux asymétries mesurées dans les simulations numériques. / The transport of heat and particles in the edge of tokamaks plays a key role in both the performance of the confined plasma and the extraction of power and thus the lifetime of the plasma facing components. It’s in this context that this thesis is inscribed, which focuses on the role played by the transverse magnetic field flows in the balance between parallel and perpendicular dynamic that governs the edge region of a tokamak. These flows can produce poloidal asymmetries of heat and particles deposit on plasma facing components and generally asymmetries of various amounts in plasma. The radial drift velocities are due to the presence of a radial electric field resulting from charge balance (electric drift velocity) or related to effects of the toroidal geometry inducing a magnetic field inhomogeneity (curvature drift velocity). To advance the understanding of these phenomena, numerical modeling of transport and turbulence in complex geometries is essential. In addition, synthetic diagnostic tools for modeling the measurement process in numerical plasmas are developed to enable a realistic comparison between models and experiments. Modeling of perpendicular drift velocities was introduced into the SOLEDGE2D code describing the transport of the density, momentum and energy of a tokamak plasma. We first studied the impact of a prescribed electric field on plasma equilibrium to understand the mechanisms behind plasma asymmetries and study the establishment of parallel flows and asymmetry of the heat flux on plasma facing components. Then we implemented a self-consistent model solving the electric potential in SOLEDGE2D fluid equations to understand the equilibrium of the electric field and to study the effect of the magnetic configuration of the tokamak and the curvature drift velocity on it. In the second part of this thesis, a synthetic diagnosis modeling the experimental measurements of Doppler backscattering was developed and tested in order to be applied to simulations of 3D turbulent fluid code TOKAM3X. This diagnosis measures the perpendicular velocity of the plasma from the movement of the density fluctuations. It was used to compare the perpendicular velocity asymmetries observed experimentally to asymmetries measured in numericalsimulations.

Cette thèse est composée de quatre chapitres. Le chapitre 1 analyse comment les différentes interventions publiques impactent l'octroi des microcrédits. Nous montrons que la garantie des prêts peut avoir un effet contreproductif en réduisant le nombre d'entrepreneurs bénéficiant de l'accompagnement offert par l'Institution de Microfinance (IMF). Alternativement, nous montrons que les subventions visant l'accompagnement peuvent être plus efficaces relativement à la garantie des prêts.Le chapitre 2 étudie comment les décisions d'une IMF concernant l'accompagnement peuvent impacter le comportement des emprunteurs. Nous montrons que l'asymétrie d'information renversée peut conduire à une relation non-monotone entre le type de l'emprunteur et l'offre de l'accompagnement. Cet équilibre apparaît suite à l'effet "soi-miroir". Notre modèle probit bivarié confirme l'existence d'un tel équilibre.Le chapitre 3 s'intéresse aux seuils de crédit imposés aux IMFs. Nous montrons que les seuils de crédit peuvent générer l'éloignement de la mission en facilitant le co-financement avec les banques classiques des projets les plus larges au détriment des projets les plus petits. Notre modèle probit différence-en-différences confirme l'existence de cet effet pervers à partir des données d'une IMF française.Le chapitre 4 compare les prêts octroyés aux entrepreneurs hommes et femmes par une IMF française avant et après l'introduction du seuil de crédit. Nous montrons que, sans le seuil, l'IMF choisit les femmes avec les demandes de crédit les plus élevées. Cependant, cela n'est plus le cas après l'introduction du seuil de crédits qui détériore la situation des entrepreneurs femmes. / This thesis is organized in four chapters.Chapter 1 theoretically analyses how various forms of state intervention impact microfinance institutions' (MFIs') lending behavior. We show that loan guarantees can have a counterproductive effect on financial inclusion triggered by unsubsidized business development services (BDS). Alternatively, we show that, BDS subsidization can do better in terms of financial inclusion than the loan guarantee. Chapter 2 analyses how decisions of an MFI on BDS provision can impact borrowers' behavior. We show that, reversed asymmetric information can lead to a non monotonic relationship between borrowers' type and assignment to BDS. In this equilibrium the MFI does not train the lowest and the highest type borrowers. This relationship occurs due to the "looking-glass self" effect. Our empirical bivariate probit model confirms the existence of such equilibrium. Chapter 3 tackles the issue of loan ceilings imposed to MFIs. We show that loan ceilings can trigger mission drift by facilitating the co-financing of large projects with regular banks at the expense of small projects. We test this prediction by exploiting the natural experiment of a French MFI. Difference-in-differences probit estimations show that the risk of the mission drift is real.Chapter 4 compares the loans granted to male and female entrepreneurs by a French MFI before and after the enforcement of the loan ceiling. We find that the ceiling free MFI selected women with larger requested amounts. However, under ceiling enforcement this was no longer the case, suggesting that female entrepreneurs are worse off after ceiling enforcement.

Ce travail de thèse traite de l’optimisation des analyses et des prétraitements des images radar pour la détection des nappes d'huile en domaine océanique (communément appelés "oil slicks" en anglais) ainsi que la localisation des sources de suintements d’huiles d'origine naturelle ("oil seeps") sur le plancher océanique. Moyens, méthodes et difficultés des divers traitements y sont exposés. Il se compose des trois axes de recherche distincts expliqués et détaillés ci-dessous :1- Une approche stochastique pour le prétraitement et l'amélioration des images radar en bande C afin de détecter automatiquement les nappes d'huile.2- Une approche stochastique utilisant une grande quantité d'images radar pour évaluer l'influence de la vitesse du vent et les différents modes de l'instrument (SAR) pour l'optimisation de la détection des nappes d'hydrocarbures.3- La localisation précise de la source des émissions d'hydrocarbures marins à l'aide d'un nouveau modèle de dérive verticale, appliqué au Golfe du Mexique (USA).En premier, nous nous sommes intéressés à l'optimisation des prétraitements et l'amélioration des images radar en bande C par des méthodes stochastiques. La méthodologie proposée comprend trois niveaux de traitement: prétraitement, seuillage et nettoyage binaire. Le premier niveau s’attèle à corriger l'hétérogénéité de la luminosité dans les images radar due à la réflexion non lambertienne du signal radar sur la surface de la mer. Le deuxième niveau consiste en une étape de seuillage qui vise à produire des objets noirs aussi proches que possible de l'ensemble de données d'apprentissage manuellement élaborées. Le troisième niveau, quant à lui, vise à nettoyer les images binaires de sortie des résidus de bruit. Plusieurs méthodes de prétraitement et de nettoyage ont été testées et évaluées par un moteur de qualification qui compare les objets détectés automatiquement avec les zones des objets noirs détectées manuellement. Par la suite, nous nous sommes penchés sur l'évaluation de l'influence de la vitesse du vent et des modes de l'instrument sur la détection des nappes d'hydrocarbures sur les images radar en utilisant une approche stochastique. Cette étude a été dictée par le besoin de définir les conditions météorologiques à même de permettre une détection optimale des nappes d’huiles, à partir des images radar. L’objectif a été de déterminer l’intervalle de vitesse du vent qui optimise la détection des nappes d'huiles dans toutes les images radar utilisant du BigData et une approche stochastique. Ce travail a également été une occasion de nous intéresser aux propriétés des modes d'acquisition radar employés dans la détection des nappes d'huile. Ainsi, un ordre de performance de 5 modes est établi (IW, APP, PRI, IMP et WSM) et montre que le mode IW (Sentinel-1), avec la meilleure résolution spatiale (supérieure à 5x20m), est la plus approprié pour détecter une nappe d'huile à forte vitesse du vent. Enfin, nous nous sommes focalisés sur l'estimation de la localisation des sources de pétrole naturel marin à l'aide d'un nouveau modèle de dérive verticale. Les manifestations de suintements d'hydrocarbures sur la surface de la mer sont généralement décalées de leur source sur les fonds marins de plusieurs centaines de mètres ou même de kilomètres. Ce décalage est fonction de la vitesse ascensionnelle et des courants marins le long de la colonne d'eau. Dans cette étude, le diamètre des gouttelettes ne nous est pas connu à priori. Pour combler ce manque d’information, on a appliqué une nouvelle méthode appelée «le chemin des sources». Si ces trois études peuvent être prises chacune indépendamment des autres, elles sont solidement interconnectées et complémentaires. Elles forment une sorte de processus allant de l'optimisation de la détection d’une nappe (les moyens et les outils les plus adéquats pour une meilleure détection) jusqu’à la localisation de sa source sur le plancher océanique / This thesis deals with the preprocessing of radar images and their optimization for the analyzes in order to detect natural marine oil slicks (Sea surface Outbreak/SSO) as well as better determine their source location at the Sea Floor Source (SFS). We explained herein means, methods and difficulties encountered. This thesis consists of the following three distinct research axes represented by three submitted papers :1- A stochastic approach for pre-processing and improvement of C-band radar images to automatically detect oil slicks;2- A stochastic approach using a large quantity of radar images to evaluate the influence of wind speed and the different modes of the instrument (SAR) on the delectability of marine oil slicks ;3- Accurate location of the Sea Floor Source of marine hydrocarbon emissions using a new vertical drift model within the water column, applied to the northern Gulf of Mexico (southern USA).So first, we focused on the optimization of pre-processing and the improvement of C-band radar images by stochastic methods to automatically detect oil slicks. The proposed methodology includes three processing levels : preprocessing, thresholding, and binary cleaning. The first level consists of correcting the heterogeneity of the luminosity in the radar images resulting from the non-Lambertian reflection of the radar signal on the sea surface. The second level consists of a thresholding step which aims to produce dark objects as close as possible to the manually developed training data set. The third level consists of cleaning the output binary images from the noise residuals. Several preprocessing and cleaning methods have been tested and evaluated by a qualification engine that compares the objects automatically detected with the manual detection. Then, we focus in a second chapter in the evaluation of the influence of wind speed and instrument modes on the detection of oil slicks from radar images by using a stochastic approach. This study was dictated by the need to define the meteorological conditions capable for an optimal detection of oil slicks, from the radar images. The objective was to determine the wind speed range which optimizes the detection of oil slicks in all radar images using BigData and a stochastic approach. This work was also an opportunity to investigate the properties of the radar acquisition modes used in the detection of oil slicks. Thus, a 5-mode performance order is established (IW, APP, PRI, IMP and WSM) and shows that the IW (Sentinel-1) mode, with the best spatial resolution (greater than 5x20m) detects oil slicks at high wind speed. Finally, we focused on estimating the location of marine natural oil seeps sources using a new vertical drift model, applied in the Gulf of Mexico. Thus, we have developed a new method for detecting the source of oil seeps from natural sources on the seafloor according to the vertical drift model. Occurrences of oil seeps on the sea surface are generally offset from their sources on the seabed by several hundred meters or even kilometers. This deflection is dependent on the upward velocity of the oil and marine currents along the water column. In this study, the diameter of the droplets is not known to us a priori. To fill this gap, a new method called "the sources path" was applied herein that propose the Sea Floor Source taking into account the droplet size and the vertical drift within the water column before their Sea surface Outbreak (SSO).If these three studies can be taken independently of each other, they are firmly interconnected and complementary. They form a sort of process ranging from the optimization of the detection of an oil slick (the most appropriate means and tools for better detection) to the location of its source on the seafloor

In this thesis, we investigate examples of random walks on free products of cyclic groups. Free products are groups that contain words constructed by concatenation with possible simplifications[20]. Mairesse in [17] proved that the harmonic measure on the boundary of these random walks has a Markovian Multiplicative structure (this is a class of Markov measures which requires fewer parameters than the usual Markov measures for its description ), and also showed how in the case of the harmonic measure these parameters can be found from Traffic Equations. Then Mairesse and Math ́eus in [20] continued investigation of these random walks and the associated Traffic Equations. They introduced the Stationary Traffic Equations for the situation when the measure is shift-invariant in addition to being μ-invariant. In this thesis, we review these developments as well as explicitly describe several concrete examples of random walks on free products, some of which are new.

The notion of Strategic Fit has been and remains to be one of the most important arguements in the fields of business strategy and strategic management. This research study examines the key concept of ‘strategic fit’ and its associated theories and seeks to investigate the causes that have created ‘strategic drift’ in Abu Dhabi’s governmental agencies in the infrastructure sector with the aim of improving their performance. However, for organisations, public or private, it still remains to be the most important notion and one that cannot be ignored because it is about the success of the organisation in its external environment where all competitive activity takes place. Thus, there are many underlying factors such as organisational culture, structure and organisational history that impact, or influence, the level of fit that organisations achieve. Therefore it is imperative that research is further undertaken on a deeper and wider level to fully understand the concept and importance of strategic fit and how it can be achieved. Hence, a research study in this area, especially in developing cities such as Abu Dhabi, is well justified and needed. The lack of strategic fit that has been witnessed in the Abu Dhabi’s infrastructure sector over the past few years continues to be, and it is a clear indication of a developing mismatch between the government and its policies and the agencies that implement them. However, this is clearly a major issue for the government going forward, if polices are developed with no clear understanding of the available resources and capabilities. Similarly, the study also seeks to determine why a strategic fit has not been achievable by the Abu Dhabi government. Though the government has developed policies to better serve its people, it continuously faces issues of policies not being implemented, or being implemented too late and targets not been met according to the original brief. It has been identified that there are numerous deficiencies between the various government agencies in the infrastructure sector in terms of keeping pace with governmental policies - consequently, resulting in strategic gaps with an increasing possibility of a possible strategic drift, if these issues are not addressed effectively and in a timely manner. The study further wishes to explore if strategic gaps have occurred as a result of inherent cultural, historical, political and/or structural aspects of the Abu Dhabi government and its numerous agencies – simply because these issues can be seen as preventative to change and progress which, in turn, can lead to strategic gaps and strategic drift in the long term. The research paradigm selected for this research study is that of the pragmatist approach and subsequently the key methodology employed throughout this research is that of the mixed methods. That mixed methods research is an intellectual and practical synthesis based on qualitative and quantitative research. Also, This research provides a summary of the important findings, proved hypotheses, achieved aim and objectives, and significant contribution to the knowledge of strategic management through the development of the “Community Innovation Scheme”, the “3-Spectrums of Change”, the “4Ms Change Model”; and the resulting framework model; the “Government Strategy Model”.