Conselho Nacional de Desenvolvimento CientÃfico e TecnolÃgico / nÃo hà / A Flutamida à um medicamento anti-androgÃnico nÃo esterÃide aprovado para o tratamento do cÃncer de prÃstata. Seu efeito hepatotÃxico foi conhecido desde o inÃcio do perÃodo de comercializaÃÃo, hà mais de 10 anos. Em marÃo e junho de 2002, a Unidade de FarmacovigilÃncia da ANVISA recebeu as primeiras comunicaÃÃes de Ãbitos por hepatite fulminante associados ao uso da substÃncia (medicamento manipulado) em jovens do sexo feminino. A literatura mundial nunca registrara caso semelhante. O objetivo deste trabalho foi revisar e acrescentar informaÃÃes a respeito da seguranÃa da Flutamida quando utilizadas por pacientes do sexo feminino para o tratamento de hirsutismo, acne e alopecia androgenÃtica. Foram realizadas pesquisas para conhecer a indicaÃÃo aprovada em outros paÃses e as notificaÃÃes de reaÃÃes adversas graves em mulheres no banco de dados da OMS. AlÃm disso, por meio de busca ativa, tentou-se identificar outros casos no Brasil; rever as informaÃÃes das bulas de todos os medicamentos comercializados e analisar se as informaÃÃes obrigatÃrias e relevantes para diminuir o risco de reaÃÃes hepÃticas graves estavam presentes. Por fim, foi realizada pesquisa com os prescritores, por meio de questionÃrio eletrÃnico, para conhecer as informaÃÃes de que dispunham sobre a utilizaÃÃo do medicamento em mulheres e comparar estas informaÃÃes com uma revisÃo crÃtica da literatura no que tange à seguranÃa e eficÃcia. Os resultados de todas estas estratÃgias mostraram que a substÃncia tambÃm à utilizada off label em mulheres em outros paÃses, e tambÃm causou reaÃÃes adversas graves, mas nenhum Ãbito no sexo feminino fora relatado. Durante o trabalho, outros casos de Ãbito no Brasil foram identificados. Todas as bulas comercializadas no Brasil mencionavam que a substÃncia nÃo devia ser utilizada por mulheres, assim como mencionavam as reaÃÃes hepÃticas graves. As respostas dos prescritores ao questionÃrio permitem afirmar que a Flutamida se constitui em importante arsenal terapÃutico, e que o risco de seu uso vem sendo subestimado. O trabalho permite concluir que a Flutamida mostra-se eficaz para as condiÃÃes em que vem sendo utilizada em mulheres. Entretanto, os riscos evidenciam uma relaÃÃo benefÃcio/risco inaceitÃvel para as condiÃÃes estÃticas em que vem sendo empregada. / Flutamide is a non steroid anti-androgenic drug used to treat prostate cancer patients. When it became commercialized, more than 10 years ago, the hepatotoxic effect of flutamide was already recognized. In March and July 2002, the first cases of deaths in young females caused by fulminant hepatitis associated to off label use of this substance (magistral drug) was reported to the Pharmacovigilance Office/ANVISA. Up to now, similar cases were not described in the international literature. This work aimed to review and add information about the adequate use of flutamide by female patients to treat hirsutism, acne and androgenetic alopecia. To identify the standard prescriptions and the reported serious reactions in females worldwide, an extensive research through data provided by OMS was conducted. Moreover, using active search we looked for new local cases and review the information from package insert of all commercialized drugs in Brazil to analyze if the required and relevant information warning patients against the serious hepatic risks were provided. Finally, a survey with the prescribers was carried on through the web to identify the available information about the use of this drug in females and to make comparisons with data from the literature about the safety and efficacy of the use of this substance. The results showed that this substance is also used off label by women from other countries, who also presented serious adverse reactions. However, it was not reported any case of female death. Over the study, new cases of obituary occurred in Brazil, despite all analyzed package insert explain that this substance should not be used by females and mention serious hepatic reactions. The answers of the prescribers to the survey indicated that flutamide is an important therapeutic arsenal and that the risk of the use of this substance seem to have been underestimated. We concluded that flutamide is effective for the women treatment which it has been used. However, the threats of this drug point to the unacceptable use of this drug when applied for the esthetic finality.

Métodos tradicionais de aprendizado supervisionado, chamados de aprendizado monorrótulo, consideram que cada exemplo do conjunto de dados rotulados está associado a um único rótulo. No entanto, existe uma crescente quantidade de aplicações que lidam com exemplos que estão associados a múltiplos rótulos. Essas aplicações requerem métodos de aprendizado multirrótulo. Esse cenário de aprendizado introduz novos desafios que demandam abordagens diferentes daquelas tradicionalmente utilizadas no aprendizado monorrótulo. O custo associado ao processo de rotulação de exemplos, um problema presente em aprendizado monorrótulo, é ainda mais acentuado no contexto multirrótulo. O desenvolvimento de métodos para reduzir esse custo representa um desafio de pesquisa nessa área. Além disso, novos métodos de aprendizado também devem ser desenvolvidos para, entre outros objetivos, considerar a dependência de rótulos: uma nova característica presente no aprendizado multirrótulo. Há um consenso na comunidade de que métodos de aprendizado multirrótulo têm a capacidade de usufruir de melhor eficácia preditiva quando considerada a dependência de rótulos. Os principais objetivos deste trabalho estão relacionados a esses desafios: reduzir o custo do processo de rotulação de exemplos; e desenvolver métodos de aprendizado que explorem a dependência de rótulos. No primeiro caso, entre outras contribuições, um novo método de aprendizado ativo, chamado score dev, é proposto para reduzir os custos associados ao processo de rotulação multirrótulo. Resultados experimentais indicam que o método score dev é superior a outros métodos em vários domínios. No segundo caso, um método para identificar dependência de rótulos, chamado UBC, é proposto, bem como o BR+, um método para explorar essa característica. O método BR+ apresenta resultados superiores a métodos considerados estado da arte / Traditional supervised learning methods, called single-label learning, consider that each example from a labeled dataset is associated with only one label. However, an increasing number of applications deals with examples that are associated with multiple labels. These applications require multi-label learning methods. This learning scenario introduces new challenges and demands approaches that are different from those traditionally used in single-label learning. The cost of labeling examples, a problem in single-label learning, is even higher in the multi-label context. Developing methods to reduce this cost represents a research challenge in this area. Moreover, new learning methods should also be developed to, among other things, consider the label dependency: a new characteristic present in multi-label learning problems. Furthermore, there is a consensus in the community that multi-label learning methods are able to improve their predictive performance when label dependency is considered. The main aims of this work are related to these challenges: reducing the cost of the labeling process; and developing multi-label learning methods to explore label dependency. In the first case, as well as other contributions, a new multi-label active learning method, called score dev, is proposed to reduce the multi-labeling processing costs. Experimental results show that score dev outperforms other methods in many domains. In the second case, a method to identify label dependency, called UBC, is proposed, as well as BR+, a method to explore this characteristic. Results show that the BR+ method outperforms other state-of-the-art methods

O E. grandis x E. camaldulensis possui características favoráveis de adaptação à seca, conferidas pelo E. camaldulensis e qualidade da madeira para papel e celulose, conferida pelo E. grandis. Esta adaptação à seca está relacionada a fatores fisiológicos e também moleculares, expressos em sua proteoma e metaboloma, que se alteram na presença do estresse. Objetiva-se neste trabalho estudar as respostas fisiológicas, proteômicas e metabolômicas (metabólitos primários) diferencialmente expressos em folhas de Eucalyptus submetidas ao déficit hídrico. Dois genótipos de E. grandis x E. camaldulensis, sendo um tolerante (T) e um susceptível (S) ao déficit hídrico foram submetidos à 100% e 30% da capacidade de campo (CC), para as plantas bem irrigadas e as em déficit hídrico, respectivamente. Os tratamentos foram chamados de T100, T30, S100 e S30 para os diferentes genótipos, T e S, submetidos a diferentes CC, 100% e 30%. Estas plantas foram avaliadas fisiologicamente com auxílio do equipamento Infrared Gas Analyzer (IRGA). Foram empregadas técnicas de proteômica quantitativa, label-free e shotgun, através do uso de UPLC-MSE. O estudo de metabolômica ocorreu através da utilização do GC x GC-TOF/MS. Os dados de proteômica foram processados no programa Protein Lynx Global Server (PLGS) e ExpressionE, através das análises comparativas S100 vs S30 e T100 vs T30, e dos metabólitos primários nos programas ChromaTOF e MetaboAnalyst. Foi possível observar que o T100 apresentou menor taxa fotossintética e condutância estomática do que o S100. Ambos os genótipos apresentaram taxas fotossintéticas e condutância estomática muito menores a 30% da CC do que a 100% da CC. A análise proteômica identificou um total de 397, 305, 366, 309 proteínas nos tratamentos S100, S30, T100 e T30 respectivamente. As análises comparativas por PLGS constataram que houve um aumento no número de proteínas diferencialmente expressas na presença do déficit hídrico. Cinco processos biológicos que apresentaram um aumento no número de proteínas diferencialmente expressas na presença do déficit hídrico foram: homeostase celular, fotossíntese, resposta ao estímulo abiótico, resposta ao estresse e morte celular. Três vias biológicas que apresentaram a participação de muitas enzimas identificadas, relacionadas a processos fotossintéticos, foram: fixação de carbono em organismos fotossintéticos, ciclo TCA e glicólise/gluconeogênese. O déficit hídrico diminuiu o número de proteínas diferencialmente expressas relacionadas ao processo metabólico de compostos contendo bases nucleares, regulação biológica e processo biossintético, que estão relacionados ao crescimento, desenvolvimento e manutenção dos processos vitais das plantas. Em relação à análise metabolômica foram identificados um total de 93, 94, 90 e 91 metabólitos primários nos tratamentos S100, S30, T100 e T30, respectivamente. Utilizando o programa Metaboanalyst, foi possível identificar os 15 metabólitos que mais contribuíram para a separação dos tratamentos, com maiores \"VIP scores\", sendo alguns responsivos ao déficit hídrico. A via da purina e arginina foi identificada como a mais frequente dentre os metabólitos identificados com VIP score ≥ 1,5. / The E. grandis x E. camaldulensis has favourable characteristics of adaptation to drought, conferred by E. camaldulensis and quality of wood for pulp and paper, conferred by E. grandis. This adaptation to drought is related to physiological factors and also expressed in their molecular proteome and metabolome, which change in the presence of stress. The aim of this work was to study the physiological responses, proteomics and metabolomics (primary metabolites) differentially expressed in leaves of Eucalyptus under drought. Two genotypes of E. grandis x E. camaldulensis, a tolerant (T) and a susceptible (S) to drought stress, were subjected to 100% and 30% of field capacity (FC), for the well-watered plants and drought stressed plants, respectively. The treatments were called T100, T30, S100 and S30 for different genotypes, T and S, submitted to different FC, 100% and 30%. These plants were evaluated physiologically using the Infrared Gas Analyzer (IRGA). Label-free and shotgun quantitative proteomics were realized using UPLC-MSE. The metabolomics study was carried out using GC x GC-TOF/MS mass spectrometer. The proteomics data were processed using the Protein Lynx Global Server program (PLGS) and ExpressionE program, through comparative analyses S100 vs S30 and T100 vs S30, and primary metabolites in ChromaTOF and MetaboAnalyst programs. It was observed that T100 had lower photosynthetic rate and stomatal conductance than S100. Both genotypes showed stomatal conductance and photosynthetic rates lower at 30% of FC than at 100% of FC. The proteomic analysis identified a total of 397, 305, 366, 309 proteins in the treatments S100, S30, T100 and T30 respectively. Comparative PLGS analyses showed an increase in the number of differentially expressed proteins under drought stress. The five biological processes that showed an increase in the number of differentially expressed proteins under drought stress were: cellular homeostasis, photosynthesis, response to abiotic stimulus, response to stress and cell death. The three biological pathways that had the participation of many identified enzymes, which are related to photosynthetic processes, were: carbon fixation in photosynthetic organisms, TCA cycle and glycolysis / gluconeogenesis. The drought reduced the number of differentially expressed proteins related to the metabolism of compounds containing nuclear bases, biological regulation and biosynthetic process, which were related to growth, development and maintenance of the vital processes of plants. The metabolomic analysis identified a total of 93, 94, 90 and 91 primary metabolites in the treatments S100, S30, T100 and T30, respectively. Using Metaboanalyst program, it was possible to identify 15 metabolites that contributed to the separation of treatments with higher \"VIP scores\", some of these are responsive to drought. The purine and arginine pathway was identified as the most frequent among the metabolites identified with VIP score ≥ 1.5.

A diploma thesis deals with food quality labels in the Czech Republic. Quality labels are granted by the Department of Agriculture, private companies or by retail chains. The thesis describes various food quality labels granted by various organizations. A research question is: "Prefer costumers buying food with a quality label?" An author conducted a questionnaire research focused on the quality label Klasa, which is granted by the Department of Agriculture of the Czech Republic. Results of the research were compared with a research focused on the other quality label - Regionální potravina. In the following part of the thesis are compared systems of food quality labels in the Czech Republic and in foreign countries. The problem of costs connected with working of private labels is also mentioned in the thesis. Special attention is paid to private labels of food retail chains, because they are very important and economically strong players on the food market in the Czech Republic.

Multiprotocol Label Switching (MPLS) fuses the intelligence of routing with the performance of switching and provides significant benefits to networks with a pure IP architecture as well as those with IP and ATM or a mix of ther Layer 2 technologies. MPLS technology is key to scalable virtual private networks (VPNs) and end-to-end quality of service (QoS), enabling efficient utilization of existing networks to meet future growth. The technology also helps to deliver highly scalable, differentiated end-to-end IP services with simpler configuration, management, and provisioning for both Internet providers and end-users. However, MPLS is a connection-oriented architecture. In case of failure MPLS first has to establish a new label switched path (LSP) and then forward the packets to the newly established LSP. For this reason MPLS has a slow restoration response to a link or node failure on the LSP.The thesis provides a description of MPLS-based architecture as a preferred technology for integrating ATM and IP technologies, followed by a discussion of the motivation for the fast and reliable restoration mechanism in an MPLS network. In this thesis first we address the fast rerouting mechanisms for MPLS networks and then we focus on the problem of packet loss, packet reordering and packet delay for protected LSP in MPLS-based network for a single node/link failure. In order to deliver true service assurance for guaranteed traffic on a protected LSP we use the fast rerouting mechanism with a preplanned alternative LSP. We propose enhancements to current proposals described in extant literature. Our fast rerouting mechanism avoids packet disorder and significantly reduces packet delay during the restoration period.An extension of the Fast Rerouting proposal, called Reliable and Fast Rerouting (RFR), provides some preventive actions for the protected LSP against packet loss during a failure. RFR maintains the same advantages of Fast Rerouting while eliminating packet losses, including those packet losses due to link or node failure (circulating on the failed links), which were considered to be "inevitable" up to now.For the purpose of validating and evaluating the behavior of these proposals a simulation tool was developed. It is based on the NS, a well-known network simulator that is being used extensively in research work. An extension featuring the basic functionality of MPLS (MNS) is also available for the NS, and this is the basis of the developed simulation tool.Simulation results allow the comparison of Fast Rerouting and RFR with previous rerouting proposals.In addition to this we propose a mechanism for multiple failure recovery in an LSP. This proposal combines the path protection, segment protection and local repair methods. In addition to the multiple link/node failure protection, the multiple fault tolerance proposal provides a significant reduction of delay that the rerouted traffic can experience after a link failure, because the repair action is taken close to the point of failure.Then we proceed to address an inherent problem of the preplanned alternative LSP. As alternative LSPs are established together with the protected LSP it may happen that the alternative is not the optimal LSP at the time the failure occurs. To overcome this undesired behavior, we propose the Optimal and Guaranteed Alternative Path (OGAP). The proposal uses a hybrid of fast-rerouting and a dynamic approach to establish the optimal alternative LSP while rerouting the affected traffic using the preplanned alternative LSP. This hybrid approach provides the best of the fast rerouting and the dynamic approaches.At the same time we observed that the protection path becomes in fact unprotected from additional failures after the traffic is rerouted onto it.To address this we propose a guarantee mechanism for protection of the new protected LSP carrying the affected traffic, by establishing an alternative LSP for the rerouted traffic after a failure, avoiding the vulnerability problem for the protected traffic.Finally, we present a further optimization mechanism, adaptive LSP, to enhance the existing traffic engineering for Quality of Services (QoS)provision and improve network resource utilization. The adaptive LSP proposal allows more flexibility in network resource allocation and utilization by adapting the LSP to variations in all network loads,resulting in an enhancement of existing MPLS traffic engineering.

Följande studie syftar till att undersöka hur en rödvinsflaskas etikett kan designas för att attrahera målgruppen millennials och samtidigt förmedla vinets smak. För att göra detta gjordes en undersökning av målgruppens preferenser genom en kvantitativ inledande studie, följt av en förstudie som undersökte tidigare studier och även designteori. I designprocessen användes den insamlade informationen från den inledande studien och förstudien för att ta designbeslut. Tre etikettalternativ togs fram som sedan validerades med målgruppen i form av en fokusgrupp. Fokusgruppen innefattade delar med kvantitativa enkäter samt smaktest. Resultatet av studien visar att ett av de tre alternativen tenderar att attrahera målgruppen mer och samtidigt förmedla vinets smak mer än de andra. Detta alternativ heter Text över hela och har vit text på en svart bakgrund över hela etiketten. Texten är vinets smaknoter, varumärkesnamn, sort, ursprung samt årtal. All text är lika stor med lika marginaler, dock är vissa ord satt i en högre vikt för att öka läsbarheten. Varumärket är även rött för att det ska bli tydligt vad som är namnet på vinet.

Tato práce se zabývá technologií Multiprotocol Label Switching a to zejména moderními metodami, které je možné použít v rámci této technolologie. Jako příklad lze uvést využití podpory kvality služeb při směrování. V práci jsou navrhnuty a simulovány různé topologie a scénáře, které ověřují možnosti využití MPLS v podpoře kvality služeb.

Uppsatsens titel: Leverantörens varumärkesstrategi – En studie om hur konsumenters beteende kan påverka en leverantörs varumärkesstrategi gällande private label och national brands Kurs: Examensarbete, Civilekonomexamen, Marknadsföring, 30hp (4FE62E) Lärosäte: Ekonomihögskolan vid Linnéuniversitetet, Kalmar Författare: Amanda Andersson och Rebecca Danielsson Handledare: Christine Tidåsen Examinator: Bertil Hultén Datum: 2014-05-26 Nyckelord: Private Label, National Brands, Varumärkesstrategi, Leverantör, Konsumentbeteende, Attitydslojalitet, Beteendelojalitet   Bakgrund: Varumärkesutbudet på hyllorna i dagligvaruhandeln har förändrats. Denna uppsats fokuserar på hur en konsuments beteende gällande varumärkena private label och national brands ser ut. Vidare utgår uppsatsen från hur konsumentens beteende kan påverka leverantörens varumärkesstrategi gällande private label och national brands.   Syfte: Examensarbetets syfte är att beskriva hur konsumenters köpbeteende och tankar kan påverka en leverantörs varumärkesstrategi gällande private label och national brands. Vidare avser uppsatsens författare generera slutsatser gällande hur leverantörer kan hantera sin varumärkesstrategi. Slutligen ämnar examensarbetets slutsatsers ligga till grund för rekommendationer till samarbetsföretaget och leverantören Nordic Stream.    Metod: Denna uppsats är en kvalitativ studie där konsumenters beteende och tankar beskrivits utifrån fyra fokusgruppintervjuer. I uppsatsen har även intervjuer genomförts med leverantören Nordic Stream och detaljisterna ICA och Bauhaus. Uppsatsen bygger på en abduktiv forskningsansats. Den teoretiska referensramen bygger på särskilda författare och teorier inom marknadsföring.   Slutsats: Attitydslojala konsumenter det vill säga de konsumenter som har en positiv attityd till en detaljists butiksvarumärke utryckte en mer positiv attityd mot private label. Beteendelojala konsumenter det vill säga de konsumenter som åter besöker en detaljist oberoende av attityden till detta butiksvarumärke utryckte en mer negativ attityd mot private label. Leverantörers varumärkesstrategier kan påverkas av konsumenters beteende både direkt och indirekt. Det som identifierats som intressant och påverkar leverantören är attityden och förväntningen på de olika varumärkena innan ett köpbeslut. / Titel: The supplier's brand strategy-a study of how consumer behaviour can affect a supplier's brand strategy regarding private label and national brands Course: Master Thesis, Business Administration and Economics, Marketing, 30 ETCS (4FE62E) Institution: School of Buisness and Economics at Linnaeus University, Kalmar Authors: Amanda Andersson and Rebecca Danielsson Supervisor: Christine Tidåsen Examiner: Bertil Hultén Date: 2014-05-26 Keywords: Private Label, National Brands, Brand Strategy, Supplier, Consumer Behaviour, Attitude loyalty, Behaviour loyalty   Background: The brand range on the grocery shelves in the stores has changed. This thesis focuses on describing consumers behaviour regarding the brands private label and national brands. Further more the thesis focus on how consumer behavior can affect the suppliers brand strategy regarding private label and national brands.   Purpose: This thesis purpose is to describe how consumers behaviour and thoughts can affect a suppliers brand strategy regarding private label and national brands. Furthermore the thesis generate conclusions regarding how suppliers can manage their brand strategy. Finally the thesis conclusions intends to be the basis for recommendations to the the supplier Nordic Stream.   Method: This thesis is a qualitative study where consumer behaviour and thoughts is described in terms of four focus group interviews. The thesis has also conducted interviews with the supplier Nordic Stream and the retailers ICA and Bauhaus. The thesis is based on an abductive research approach. The theoretical framework is based on specific authors and theories in marketing.   Conclusion: Attitude loyal consumers are consumers who have a positive attitude to the retailers store brand and are those who expressed a more positive attitude toward private label. Behaviour loyal consumers are consumers who revisits a retailer regardless of the attitude to this store brand and are those who expressed a more negative attitude toward private label. Suppliers brand strategies can be affected by consumer behaviour, both directly and indirectly. Consumers attitudes and expectations of different brands before making a purchasing decision affect suppliers brand strategy.

Les algorithmes d'inférence ou d'optimisation possèdent généralement des hyperparamètres qu'il est nécessaire d'ajuster. Nous nous intéressons ici à l'automatisation de cette étape d'ajustement et considérons différentes méthodes qui y parviennent en apprenant en ligne la structure du problème considéré.La première moitié de cette thèse explore l'ajustement des hyperparamètres en apprentissage artificiel. Après avoir présenté et amélioré le cadre générique de l'optimisation séquentielle à base de modèles (SMBO), nous montrons que SMBO s'applique avec succès à l'ajustement des hyperparamètres de réseaux de neurones profonds. Nous proposons ensuite un algorithme collaboratif d'ajustement qui mime la mémoire qu'ont les humains d'expériences passées avec le même algorithme sur d'autres données.La seconde moitié de cette thèse porte sur les algorithmes MCMC adaptatifs, des algorithmes d'échantillonnage qui explorent des distributions de probabilité souvent complexes en ajustant leurs paramètres internes en ligne. Pour motiver leur étude, nous décrivons d'abord l'observatoire Pierre Auger, une expérience de physique des particules dédiée à l'étude des rayons cosmiques. Nous proposons une première partie du modèle génératif d'Auger et introduisons une procédure d'inférence des paramètres individuels de chaque événement d'Auger qui ne requiert que ce premier modèle. Ensuite, nous remarquons que ce modèle est sujet à un problème connu sous le nom de label switching. Après avoir présenté les solutions existantes, nous proposons AMOR, le premier algorithme MCMC adaptatif doté d'un réétiquetage en ligne qui résout le label switching. Nous présentons une étude empirique et des résultats théoriques de consistance d'AMOR, qui mettent en lumière des liens entre le réétiquetage et la quantification vectorielle / Inference and optimization algorithms usually have hyperparameters that require to be tuned in order to achieve efficiency. We consider here different approaches to efficiently automatize the hyperparameter tuning step by learning online the structure of the addressed problem. The first half of this thesis is devoted to hyperparameter tuning in machine learning. After presenting and improving the generic sequential model-based optimization (SMBO) framework, we show that SMBO successfully applies to the task of tuning the numerous hyperparameters of deep belief networks. We then propose an algorithm that performs tuning across datasets, mimicking the memory that humans have of past experiments with the same algorithm on different datasets. The second half of this thesis deals with adaptive Markov chain Monte Carlo (MCMC) algorithms, sampling-based algorithms that explore complex probability distributions while self-tuning their internal parameters on the fly. We start by describing the Pierre Auger observatory, a large-scale particle physics experiment dedicated to the observation of atmospheric showers triggered by cosmic rays. The models involved in the analysis of Auger data motivated our study of adaptive MCMC. We derive the first part of the Auger generative model and introduce a procedure to perform inference on shower parameters that requires only this bottom part. Our model inherently suffers from label switching, a common difficulty in MCMC inference, which makes marginal inference useless because of redundant modes of the target distribution. After reviewing existing solutions to label switching, we propose AMOR, the first adaptive MCMC algorithm with online relabeling. We give both an empirical and theoretical study of AMOR, unveiling interesting links between relabeling algorithms and vector quantization.

Dans cette thèse, nous nous intéressons au problème spécifique de l'apprentissage de structure de modèles graphiques probabilistes, c'est-à-dire trouver la structure la plus efficace pour représenter une distribution, à partir seulement d'un ensemble d'échantillons D ∼ p(v). Dans une première partie, nous passons en revue les principaux modèles graphiques probabilistes de la littérature, des plus classiques (modèles dirigés, non-dirigés) aux plus avancés (modèles mixtes, cycliques etc.). Puis nous étudions particulièrement le problème d'apprentissage de structure de modèles dirigés (réseaux Bayésiens), et proposons une nouvelle méthode hybride pour l'apprentissage de structure, H2PC (Hybrid Hybrid Parents and Children), mêlant une approche à base de contraintes (tests statistiques d'indépendance) et une approche à base de score (probabilité postérieure de la structure). Dans un second temps, nous étudions le problème de la classification multi-label, visant à prédire un ensemble de catégories (vecteur binaire y P (0, 1)m) pour un objet (vecteur x P Rd). Dans ce contexte, l'utilisation de modèles graphiques probabilistes pour représenter la distribution conditionnelle des catégories prend tout son sens, particulièrement dans le but minimiser une fonction coût complexe. Nous passons en revue les principales approches utilisant un modèle graphique probabiliste pour la classification multi-label (Probabilistic Classifier Chain, Conditional Dependency Network, Bayesian Network Classifier, Conditional Random Field, Sum-Product Network), puis nous proposons une approche générique visant à identifier une factorisation de p(y|x) en distributions marginales disjointes, en s'inspirant des méthodes d'apprentissage de structure à base de contraintes. Nous démontrons plusieurs résultats théoriques, notamment l'unicité d'une décomposition minimale, ainsi que trois procédures quadratiques sous diverses hypothèses à propos de la distribution jointe p(x, y). Enfin, nous mettons en pratique ces résultats afin d'améliorer la classification multi-label avec les fonctions coût F-loss et zero-one loss / In this thesis, we address the specific problem of probabilistic graphical model structure learning, that is, finding the most efficient structure to represent a probability distribution, given only a sample set D ∼ p(v). In the first part, we review the main families of probabilistic graphical models from the literature, from the most common (directed, undirected) to the most advanced ones (chained, mixed etc.). Then we study particularly the problem of learning the structure of directed graphs (Bayesian networks), and we propose a new hybrid structure learning method, H2PC (Hybrid Hybrid Parents and Children), which combines a constraint-based approach (statistical independence tests) with a score-based approach (posterior probability of the structure). In the second part, we address the multi-label classification problem, which aims at assigning a set of categories (binary vector y P (0, 1)m) to a given object (vector x P Rd). In this context, probabilistic graphical models provide convenient means of encoding p(y|x), particularly for the purpose of minimizing general loss functions. We review the main approaches based on PGMs for multi-label classification (Probabilistic Classifier Chain, Conditional Dependency Network, Bayesian Network Classifier, Conditional Random Field, Sum-Product Network), and propose a generic approach inspired from constraint-based structure learning methods to identify the unique partition of the label set into irreducible label factors (ILFs), that is, the irreducible factorization of p(y|x) into disjoint marginal distributions. We establish several theoretical results to characterize the ILFs based on the compositional graphoid axioms, and obtain three generic procedures under various assumptions about the conditional independence properties of the joint distribution p(x, y). Our conclusions are supported by carefully designed multi-label classification experiments, under the F-loss and the zero-one loss functions