[pt] Esta dissertação faz uma comparação de três métodos de classificação em cascata de imagens multitemporais. Os classificadores se baseiam nas seguintes técnicas: (1) Máquina de Suporte Vetorial (SVM), (2) Modelos Ocultos de Markov (HMM) e (3) Cadeias de Markov Nebulosas(FMC). Para verificar a robustez dos modelos de classificação, introduziram-se nos dados de entrada outliers, avaliando-se assim, a robustez dos classificadores. Adicionalmente, avaliou-se o desempenho dos métodos quando a proporção de ocorrências de cada transição de classe no conjunto de treinamento difere da proporção no conjunto de teste. Determinou-se também qual o benefício do uso de conhecimento a priori sobre as transições possíveis. A análise experimental foi realizada sobre dois conjuntos de imagens de diferentes características, um par de imagens IKONOS do Rio de Janeiro, Brasil e um par de imagens LANDSAT7 de Alcinópolis, Mato Grosso do Sul. O estudo revelou que acurácia global das três abordagens tem um comportamento similar nos diferentes experimentos. Mostrou também que todas as três abordagens multitemporais apresentam desempenho superior aos seus homólogos monotemporais. / [en] This dissertation compares three cascade multitemporal image classification methods based on: (1) Support Vector Machines (SVM), (2) Hidden Markov Models (HMM) and (3) Fuzzy Markov Chains (FMC). The robustness of the classification models is verified, by introducing outliers in the data set. Additionally, performance of each method is evaluated when the number of occurrences of each class transition is different in the training and in the testing set. The gain of exploiting a prior knowledge regarding the admissible transitions in each target site is also investigated. The experimental analysis is conducted over two data sets with different characteristics; specifically a pair of IKONOS images of Rio de Janeiro and a pair of LANDSAT7 images of Alcinópolis, Mato Grosso do Sul. This study has concluded that the overall accuracy of the three approaches are similar through all experiments. The superiority ofthe multitemporal approaches over the monotemporal counterparts was confirmed.

L’analyse de la stabilité posturale chez l’homme a fait l’objet, ces dernières années, d’un intérêt grandissant au sein de la communauté scientifique. Le système postural permet de maintenir la stabilité du corps humain en posture statique ou dynamique. Cette capacité à maintenir cette stabilité devient critique dans le cas des sujets Parkinsoniens. La maladie de Parkinson a en effet une forte incidence sur la stabilité posturale. Un moyen efficace pour évaluer l’équilibre postural consiste à analyser les déplacements dans le plan horizontal du centre de pression du corps humain en posture orthostatique ; les trajectoires mesurées dans la direction medio-latérale (ML) et la direction Antéro-postérieure (AP) sont appelées signaux stabilométriques. Dans cette thèse, nous visons le développement de méthodes efficaces pour l’analyse de l’équilibre en posture orthostatique sous différentes conditions liées à l’entrée visuelle (yeux ouverts/yeux fermés), la position des pieds (pieds joints/pieds écartés) et en considérant d’autres facteurs comme le genre et l’âge. Dans ce cadre, nous proposons, tout d’abord, une méthode exploitant la variante EEMD (Ensemble Empirical Mode Decomposition) de la décomposition en modes empiriques (EMD) et l’analyse de la diffusion du stabilogramme. Dans le contexte du diagnostic de la maladie de Parkinson, la discrimination entre sujets sains et sujets Parkinsoniens est très importante, de même que l’évaluation du stade de la maladie pour les sujets atteints. Dans ce cadre, deux méthodes sont proposées. La première consiste tout d’abord en une extraction et sélection de caractéristiques temporelles et spectrales, à partir des signaux stabilométriques brutes ou des modes de fonctions intrinsèques dérivés de la décomposition EEMD. Des méthodes standards de type KNN, CART, RF et SVM sont ensuite appliquées pour reconnaitre les sujets Parkinsoniens. La deuxième méthode proposée, est une approche de classification qui repose sur l’emploi de HMMs construits en utilisant les signaux stabilométriques brutes dans les directions ML, AP et ML/AP. Enfin, une dernière méthode est proposée pour la segmentation automatique des signaux stabilométriques sous différentes conditions (entrée visuelle, position des pieds). Pour ce faire, un modèle de régression régi par une chaine de Markov cachée (HMMR) est utilisé pour détecter automatiquement les variations des structures des signaux stabilométriques entre ces conditions. Les résultats obtenus montrent clairement la supériorité des performances des méthodes proposées par rapport aux approches standards, aussi bien, en termes d’analyse de l’équilibre postural que de diagnostic de sujets Parkinsoniens / Recently, human balance control analysis has received an increasing interest from the research community. The human postural system maintains the stability of the body both in the static posture (quiet standing) and during locomotion. This ability to maintain stability becomes hard with aging and Parkinson's disease (PD) subjects. PD has a strong effect on postural stability during quiet standing situations, and during locomotion. One effective way to assess human stability is to analyze the center of pressure (CoP) displacements of the human body during quiet standing. The recorded CoP displacements in quiet standing are called stabilometric signals. This thesis aims to develop efficient approaches to analyze the human postural stability in quiet standing under visual and feet position conditions, as well as under age and gender. This is achieved using Empirical Mode Decomposition (EMD) method and stabilogram-diffusion technique. In the other part, the discrimination between healthy and PD subjects is very important for diagnosing Parkinson's disease, as well as for evaluating the disease level of the patient. In this context, two approaches are proposed; the first approach consists of an EMD-based temporal and spectral feature extraction from the stabilometric signals. The second approach is based on a Hidden Markov Model (HMM) using the raw stabilometric signals. The HMM model is an efficient tool to analyze temporal and sequential data. Another approach is proposed in order to segment the stabilometric signals according to the visual and feet position conditions. This is achieved using a Hidden Markov Model Regression (HMMR)-based approach. This study help clinicians to better understand the motor strategies used by the subjects during quiet standing and may guide the rehabilitation process. The obtained results clearly show high performances of the proposed approaches with respect to other standard approaches in both postural stability analysis and discrimination healthy from PD subjects

In last years Hidden Markov Models (HMMs) have received significant attention in the task off-line handwritten text recognition (HTR). As in automatic speech recognition (ASR), HMMs are used to model the probability of an observation sequence, given its corresponding text transcription. However, in contrast to what happens in ASR, in HTR there is no standard set of local features being used by most of the proposed systems. In this thesis we propose the use of raw binary pixels as features, in conjunction with models that deal more directly with the binary data. In particular, we propose the use of Bernoulli HMMs (BHMMs), that is, conventional HMMs in which Gaussian (mixture) distributions have been replaced by Bernoulli (mixture) probability functions. The objective is twofold: on the one hand, this allows us to better modeling the binary nature of text images (foreground/background) using BHMMs. On the other hand, this guarantees that no discriminative information is filtered out during feature extraction (most HTR available datasets can be easily binarized without a relevant loss of information). In this thesis, all the HMM theory required to develop a HMM based HTR toolkit is reviewed and adapted to the case of BHMMs. Specifically, we begin by defining a simple classifier based on BHMMs with Bernoulli probability functions at the states, and we end with an embedded Bernoulli mixture HMM recognizer for continuous HTR. Regarding the binary features, we propose a simple binary feature extraction process without significant loss of information. All input images are scaled and binarized, in order to easily reinterpret them as sequences of binary feature vectors. Two extensions are proposed to this basic feature extraction method: the use of a sliding window in order to better capture the context, and a repositioning method in order to better deal with vertical distortions. Competitive results were obtained when BHMMs and proposed methods were applied to well-known HTR databases. In particular, we ranked first at the Arabic Handwriting Recognition Competition organized during the 12th International Conference on Frontiers in Handwriting Recognition (ICFHR 2010), and at the Arabic Recognition Competition: Multi-font Multi-size Digitally Represented Text organized during the 11th International Conference on Document Analysis and Recognition (ICDAR 2011). In the last part of this thesis we propose a method for training BHMM classifiers using In last years Hidden Markov Models (HMMs) have received significant attention in the task off-line handwritten text recognition (HTR). As in automatic speech recognition (ASR), HMMs are used to model the probability of an observation sequence, given its corresponding text transcription. However, in contrast to what happens in ASR, in HTR there is no standard set of local features being used by most of the proposed systems. In this thesis we propose the use of raw binary pixels as features, in conjunction with models that deal more directly with the binary data. In particular, we propose the use of Bernoulli HMMs (BHMMs), that is, conventional HMMs in which Gaussian (mixture) distributions have been replaced by Bernoulli (mixture) probability functions. The objective is twofold: on the one hand, this allows us to better modeling the binary nature of text images (foreground/background) using BHMMs. On the other hand, this guarantees that no discriminative information is filtered out during feature extraction (most HTR available datasets can be easily binarized without a relevant loss of information). In this thesis, all the HMM theory required to develop a HMM based HTR toolkit is reviewed and adapted to the case of BHMMs. Specifically, we begin by defining a simple classifier based on BHMMs with Bernoulli probability functions at the states, and we end with an embedded Bernoulli mixture HMM recognizer for continuous HTR. Regarding the binary features, we propose a simple binary feature extraction process without significant loss of information. All input images are scaled and binarized, in order to easily reinterpret them as sequences of binary feature vectors. Two extensions are proposed to this basic feature extraction method: the use of a sliding window in order to better capture the context, and a repositioning method in order to better deal with vertical distortions. Competitive results were obtained when BHMMs and proposed methods were applied to well-known HTR databases. In particular, we ranked first at the Arabic Handwriting Recognition Competition organized during the 12th International Conference on Frontiers in Handwriting Recognition (ICFHR 2010), and at the Arabic Recognition Competition: Multi-font Multi-size Digitally Represented Text organized during the 11th International Conference on Document Analysis and Recognition (ICDAR 2011). In the last part of this thesis we propose a method for training BHMM classifiers using In last years Hidden Markov Models (HMMs) have received significant attention in the task off-line handwritten text recognition (HTR). As in automatic speech recognition (ASR), HMMs are used to model the probability of an observation sequence, given its corresponding text transcription. However, in contrast to what happens in ASR, in HTR there is no standard set of local features being used by most of the proposed systems. In this thesis we propose the use of raw binary pixels as features, in conjunction with models that deal more directly with the binary data. In particular, we propose the use of Bernoulli HMMs (BHMMs), that is, conventional HMMs in which Gaussian (mixture) distributions have been replaced by Bernoulli (mixture) probability functions. The objective is twofold: on the one hand, this allows us to better modeling the binary nature of text images (foreground/background) using BHMMs. On the other hand, this guarantees that no discriminative information is filtered out during feature extraction (most HTR available datasets can be easily binarized without a relevant loss of information). In this thesis, all the HMM theory required to develop a HMM based HTR toolkit is reviewed and adapted to the case of BHMMs. Specifically, we begin by defining a simple classifier based on BHMMs with Bernoulli probability functions at the states, and we end with an embedded Bernoulli mixture HMM recognizer for continuous HTR. Regarding the binary features, we propose a simple binary feature extraction process without significant loss of information. All input images are scaled and binarized, in order to easily reinterpret them as sequences of binary feature vectors. Two extensions are proposed to this basic feature extraction method: the use of a sliding window in order to better capture the context, and a repositioning method in order to better deal with vertical distortions. Competitive results were obtained when BHMMs and proposed methods were applied to well-known HTR databases. In particular, we ranked first at the Arabic Handwriting Recognition Competition organized during the 12th International Conference on Frontiers in Handwriting Recognition (ICFHR 2010), and at the Arabic Recognition Competition: Multi-font Multi-size Digitally Represented Text organized during the 11th International Conference on Document Analysis and Recognition (ICDAR 2011). In the last part of this thesis we propose a method for training BHMM classifiers using discriminative training criteria, instead of the conventionalMaximum Likelihood Estimation (MLE). Specifically, we propose a log-linear classifier for binary data based on the BHMM classifier. Parameter estimation of this model can be carried out using discriminative training criteria for log-linear models. In particular, we show the formulae for several MMI based criteria. Finally, we prove the equivalence between both classifiers, hence, discriminative training of a BHMM classifier can be carried out by obtaining its equivalent log-linear classifier. Reported results show that discriminative BHMMs clearly outperform conventional generative BHMMs. / Giménez Pastor, A. (2014). Bernoulli HMMs for Handwritten Text Recognition [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/37978 / TESIS

[pt] Com o crescimento gigantesco da Internet e dos sistemas de comunicações móveis celulares, as aplicações de processamento de voz nessas redes têm despertado grande interesse . Um problema particularmente importante nessa área consiste no reconhecimento de voz em um sistema servidor, baseado nos parâmetros acústicos calculados e quantizados no terminal do usuário (Reconhecimento de Voz Distribuído). Como em geral estes parâmetros não são os mais indicados como atributos de voz para o sistema de reconhecimento remoto, é importante que sejam examinadas diferentes transformações dos parâmetros, que permitam um melhor desempenho do reconhecedor. Esta dissertação trata da extração de atributos de reconhecimento eficientes a partir dos parâmetros dos codificadores utilizados em redes móveis celulares e em redes IP. Além disso, como a taxa dos parâmetros fornecidos ao reconhecedor de voz é normalmente superior àquela com a qual os codificadores geram os parâmetros, é importante analisar o efeito da interpolação dos parâmetros sobre o desempenho do sistema de reconhecimento, bem como o melhor domínio sobre o qual esta interpolação deve ser realizada. Estes são outros tópicos apresentados nesta dissertação. / [en] The huge growth of the Internet and cellular mobile communication systems has stimulated a great interest in the applications of speech processing in these networks. An important problem in this field consists in speech recognition in a server system, based on the acoustic parameters calculated and quantized in the user terminal (Distributed Speech Recognition). Since these parameters are not the most indicated ones for the remote recognition system, it is important to examine different transformations of these parameters, in order to allow a better performance of the recogniser. This dissertation is concerned with the extraction of efficient recognition features from the coder parameters used in cellular mobile networks and IP networks. In addition, as the rate that parameters supplied for the speech recogniser must be usually higher than that generated by the codec, it is important to analyze the effect of the interpolation of the parameters over the performance of the recognition system. Moreover, it is paramount to establish the best domain over which this interpolation must be carried out. These are other topics presented in this dissertation.

L'objectif de cette thèse est de créer un nouveau système qui est capable de gérer des auto-formations intelligentes, interactives et adaptatives aux besoins, styles d'apprentissage et à l'état des connaissances de chaque apprenant. Ce système est nommé IWEBISE (Interactive and Intelligent System for Education). Il est le résultat de l'accouplement entre cinq grands domaines : l'ingénierie des connaissances, l'interaction homme-machine, la psychologie cognitive, l'intelligence artificielle et la psychopédagogie. IWEBISE est constitué de cinq parties :<br> Modèle de l'apprenant : Cette partie détaille comment le style d'apprentissage des apprenants est modélisé selon le modèle Felder-Silverman, qui dépend de nombreux paramètres : Nombre d'exemples, nombre d'exercices, le lieu des exemples avant ou après le contenu et le lieu des exercices avant ou après le contenu. Les connaissances des étudiants sont exprimées dans le système en utilisant le modèle de recouvrement (Overlay), qui les considère comme une partie du domaine de la connaissance. L' IWEBISE utilise également un modèle ouvert de l'apprenant . Celui-ci permet aux étudiants de changer par eux-mêmes leurs états de connaissances relatives à chaque concept, ce qui leur permet d'étudier à grands pas sans se sentir bloqués par leurs processus d'apprentissage. Le modèle de l'apprenant est composé de deux parties: La partie statique qui stocke les renseignements personnels de l'apprenant. La partie dynamique qui garde les interactions des apprenants avec le système. Elles sont présentées par plusieurs paramètres: nombre de réponses correctes (NCA), nombre de réponses incorrectes (NICA), temps passé à résoudre une question (TSSQ), temps consacré à la lecture ou à l'interaction avec un concept spécifique (TSR) et le nombre de tentatives pour répondre à une question (NAAQ). Une fois qu'un apprenant a passé une séance de pré-test, la partie dynamique est lancée en utilisant ces paramètres. Six méthodes sont employées pour symboliser l'état des connaissances des apprenants dans six niveaux (excellent, très bien, bien, plutôt bien, faibles et très faibles) dans le but de déterminer la meilleure pour être utilisée plus tard dans le nouveau IWEBISE. Ces méthodes sont: FBAM, ART2, Fuzzy-ART2, HMM et NN / HMM. F-mesure métrique est employé pour mesurer la performance des méthodes mentionnées. Les résultats montrent que Fuzzy-ART2 donne la meilleure qualité de catégorisation (0.281 ), qui est considérée comme un facteur très important pour s''assurer qu'une carte de concepts appropriée est affichée à l'apprenant en fonction de son état de connaissances. <br> Modèle de tuteur : Cette partie détaille la façon de modéliser les stratégies pédagogiques utilisées par les enseignants pendant la présentation du contenu des cours aux étudiants. Elles sont modélisées par une table composée de neuf champs qui permettent de stocker les couleurs utilisées pour présenter l'état des connaissances des apprenants dans le plan du parcours et la possibilité de montrer ou de cacher un concept d'apprentissage. Le modèle de tuteur se concentre également sur un algorithme de prédiction pour prévoir les concepts suivants qui pourraient être visités par les apprenants. Le processus de prédiction est réalisé en suivant trois phases: Phase d'initialisation: Pour chaque apprenant un HMM (λ) est construit à base de sa précédente séquence d'accès aux concepts. Phase d'ajustement: Étant donnée une nouvelle séquence observée et un HMM (λ), l'algorithme de Baum-Welch est utilisé pour ajuster les HMM initialisés et de maximiser la nouvelle séquence observée. Phase de prédiction: l'algorithme Forward est appliqué pour déterminer la distribution de probabilité de chaque concept dans le cours. La plus haute valeur représente le concept suivant qui sera visité par l'apprenant. L'exactitude de prédiction est évaluée par deux critères, la Sensibilité et la Précision. La sensibilité est définie comme le nombre de concepts prédits correctement (vrais positifs) divisé par le nombre de concepts visités (positifs réels). La précision est le pourcentage de prédictions positives qui sont correctes. Les résultats montrent que HMM génère une plus grande précision en utilisant une séquence de concepts plus large qui varie de 20% à 99% et quand la précision est élevée, la sensibilité est élevée aussi.<br> Modèle de domaine : Le contenu des cours est organisé en un réseau de concepts pour représenter les objectifs d'apprentissage. Un objectif d'apprentissage concerne plusieurs concepts qui sont classés en trois types: des concepts principaux, des concepts pré-requis et des sous-concepts. Chaque nœud interne dans le réseau représente un concept, tandis que les nœuds externes dans le niveau le plus bas symbolisent plusieurs types d'unités d'enseignement, qui sont sous la forme de fichiers multimédia Flash interactive, images, vidéos, textes, exercices, des exemples. Trois différentes méthodes sont utilisées pour décrire le modèle de domaine : HBAM (Hirarchical Bidirectional Associative Memory), une base de données MySql, et XML, Dans la première méthode, le domaine de connaissances est conçu et modélisé en utilisant un réseau de neurones BAM hiérarchique. Le premier BAM-1 associe les objectifs d'apprentissage avec les concepts, mais le second BAM-2 est utilisé pour attribuer des unités d'enseignement à chaque concept. La couche de sortie de BAM-1 est la couche d'entrée du BAM-2, qui peut être vu comme une couche intermédiaire de toute l'architecture. Nombre de nœuds de couche d'entrée, de milieu et de sortie représentent le nombre d'objectifs d'apprentissage, les concepts et les unités d'enseignement respectivement . Dans la seconde méthode une base de données relationnelle est utilisée pour représenter le domaine de connaissance. Il est composé d'onze tableaux (les catégories principales, les sous-catégories, les sujets, les objectifs d'apprentissage, les objectifs d'apprentissage pré-requis, les concepts, les concepts pré-requis, la relation d'un sous-concepts avec un concept, le contenu et les questions pour les post-tests et pré-tests). Un Document Type Définition (DTD) du fichier est construit dans la troisième méthode pour déterminer un ensemble de règles pour définir et décrire l'organisation des connaissances dans un fichier XML. La méthode de base de données relationnelle est sélectionnée pour être utilisée au sein de notre nouveau système IWEBISE car d'une part, certains concepteurs de cours préfèrent avoir leur contenu des cours confidentiels et protégés et de l'autre part, XML n'est pas capable de traiter le contenu de cours énormes et tous les types de données tels que des images et la vidéo. <br> Le moteur d'adaptation : Il est considéré comme l'une des parties les plus importantes de IWEBISE grâce aux trois raisons suivantes: Il relie les différentes parties du système, il génère la page d'un concept selon le style d'apprentissage de chaque apprenant et il adapte la carte de concepts en fonction de l'état des connaissances de chaque apprenants aussi. <br> L'interface utilisateur représente les moyens d'interaction disponible sur le système IWEBISE. Il est classé en quatre niveaux : Administrateur: Il permet aux administrateurs de créer une catégorie de cours, sous-catégorie, gérer les utilisateurs et les processus d'abonnement. Concepteur: Il permet aux concepteurs de cours de gérer les objectifs d'apprentissage, les concepts, les sous-concepts, les contenu des concepts et les questions des tests. Elle leur permet également d'exporter leurs cours sous la norme SCORM. Tuteur: Il permet aux enseignants de gérer leurs stratégies d'enseignement et de donner des conseils appropriés aux apprenants. Apprenant: Il permet aux apprenants de compléter leur processus d'apprentissage en utilisant le pré-test, les post-tests, le questionnaire «Index des styles d'apprentissage", les glossaires, le chat et le forum. Le nouveau système IWEBISE est évalué par des concepteurs de cours et par des étudiants, dans le but d'optimiser ses performances au cours de l'enseignement et les processus d'apprentissage. Douze critères sont utilisés pour l'évaluer: la Cohérence, l'évidence, la prévisibilité, la richesse, l'exhaustivité, la motivation, la structure d'Hypertext, l'autonomie, la facilité d'utilisation, l'esthétique, la collaboration et l'interactivité.<br> L'originalité de cette thèse est basée sur : 1.L'utilisation d'une nouvelle architecture d'un réseau de neurones appelé HBAM pour modéliser le domaine des connaissances d'un cours. Ce nouveau réseau peut être utilisé dans de nombreux autres domaines tels que: la reconnaissance des formes ; 2.L 'utilisation d'un nouvel algorithme hybride qui utilise un réseau de neurones (Fuzzy-ART2) et une méthode statistique (HMM) pour la modélisation des connaissances des apprenants ; 3.L 'utilisation de nombreux algorithmes d'apprentissage tels que: FBAM, ART2, Fuzzy-ART2 et une structure hybride Fuzzy-ART2/HMM, qui sont utilisés pour classer la réflexion des apprenants et leurs raisonnements en six niveaux ; 4.L'utilisation d'un HMM pour prédire le prochain concept, basé sur l'histoire des concepts visités par un apprenant ; 5.La définition des styles d'apprentissage des apprenants à l' Institut Supérieur des Langues (Université d'Alep) par rapport à l'apprentissage d'une langue. Ceci est fait en utilisant le modèle Felder et Silverman ; 6.Le comparaison d'IWEBISE avec d'autres systèmes éducatifs. 7.Le capacité d'l'IWEBISE à l'exportation et la réalisation des cours conformément à la norme SCORM avec l'objectif de les réutiliser dans d'autres plates-formes d'enseignement telle que: Moodle ; 8.La construction et la mise en œuvre d'un nouveau système intelligent et adaptatif pour l'éducation basé sur le Web pour l'enseignement de la grammaire anglaise.

Investigations into the origin and evolution of regulatory mechanisms require quantitative estimates of the abundance and co-occurrence of functional protein domains among distantly related genomes. The metabolic and regulatory capabilities of an organism are implicit in its protein content. Currently available methods suffer for strong ascertainment biases, requiring methods for unbiased approaches to protein domain contents at genome-wide scales. The discussion will be highlighted on large scale patterns of similarities and differences of domain contains between phylum-level or even higher level taxonomic groups. This provides insights into large-scale evolutionary trends. The complement of recognizable functional protein domains and their combinations convey essentially the same information and at the same time are much more readily accessible, although protein domain models trained for one phylogenetic group frequently fail on distantly related sequences. Transcription factors (TF) typically cooperate to activate or repress the expression of genes. They play a critical role in developmental processes. While Chromatin Regulation (CR) facilitates DNA organization and prevent DNA aggregation and tangling which is important for replication, segregation, and gene expression. To compare the set of TFs and CRs between species, the genome annotation of equal quality was employed. However, the existing annotation suffers from bias in model organism. The similar count of transcripts are expected to be similar in mammals, but model organism such as human has more annotated transcripts than non model such as gorilla. Moreover, closely related species (e.g, dolphin and human) show a dramatically different distribution of TFs and CRs. Within vertebrates, this is unreasonable and contradicts phylogenetic knowledge. To overcome this problem, performing gene prediction followed by the detection of functional domains via HMM-based annotation of SCOP domains were proposed. This methods was demonstrated to lead toward consistent estimates for quantitative comparison. To emphasize the applicability, the protein domain distribution of putative TFs and CRs by quantitative and boolean means were analyzed. In particular, systematic studies of protein domain occurrences and co-occurrences to study avoidance or preferential co-occurrence of certain protein domains within TFs and CRs were utilized. Pooling related domain models based on their GO-annotation in combination with de novo gene prediction methods provides estimates that seem to be less affected by phylogenetic biases. it was shown for 18 diverse representatives from all eukaryotic kingdoms that a pooled analysis of the tendencies for co-occurrence or avoidance of protein domains is indeed feasible. This type of analysis can reveal general large-scale patterns in the domain co-occurrence and helps to identify lineage-specific variations in the evolution of protein domains. Somewhat surprisingly, strong ubiquitous patterns governing the evolutionary behavior of specific functional classes were not found. Instead, there are strong variations between the major groups of Eukaryotes, pointing at systematic differences in their evolutionary constraints. Species-specific training is required, however, to account for the genomic peculiarities in many lineages. In contrast to earlier studies wide-spread statistically significant avoidance of protein domains associated with distinct functional high-level gene-ontology terms were found.

BACKGROUND: A long-standing challenge in the post-genomic era of Bioinformatics is the prediction of protein-protein interactions, and ultimately the prediction of protein functions. The problem is intrinsically harder, when only amino acid sequences are available, but a solution is more universally applicable. So far, the problem of uncovering protein-protein interactions has been addressed in a variety of ways, both experimentally and computationally. MOTIVATION: The central problem is: How can protein complexes with solved threedimensional structure be utilized to identify and classify protein binding sites and how can knowledge be inferred from this classification such that protein interactions can be predicted for proteins without solved structure? The underlying hypothesis is that protein binding sites are often restricted to a small number of residues, which additionally often are well-conserved in order to maintain an interaction. Therefore, the signal-to-noise ratio in binding sites is expected to be higher than in other parts of the surface. This enables binding site detection in unknown proteins, when homology based annotation transfer fails. APPROACH: The problem is addressed by first investigating how geometrical aspects of domain-domain associations can lead to a rigorous structural classification of the multitude of protein interface types. The interface types are explored with respect to two aspects: First, how do interface types with one-sided homology reveal convergently evolved motifs? Second, how can sequential descriptors for local structural features be derived from the interface type classification? Then, the use of sequential representations for binding sites in order to predict protein interactions is investigated. The underlying algorithms are based on machine learning techniques, in particular Hidden Markov Models. RESULTS: This work includes a novel approach to a comprehensive geometrical classification of domain interfaces. Alternative structural domain associations are found for 40% of all family-family interactions. Evaluation of the classification algorithm on a hand-curated set of interfaces yielded a precision of 83% and a recall of 95%. For the first time, a systematic screen of convergently evolved motifs in 102.000 protein-protein interactions with structural information is derived. With respect to this dataset, all cases related to viral mimicry of human interface bindings are identified. Finally, a library of 740 motif descriptors for binding site recognition - encoded as Hidden Markov Models - is generated and cross-validated. Tests for the significance of motifs are provided. The usefulness of descriptors for protein-ligand binding sites is demonstrated for the case of &amp;quot;ATP-binding&amp;quot;, where a precision of 89% is achieved, thus outperforming comparable motifs from PROSITE. In particular, a novel descriptor for a P-loop variant has been used to identify ATP-binding sites in 60 protein sequences that have not been annotated before by existing motif databases.

Les caractéristiques des rayonnements solaires dépendent fortement de certains événements météorologiques non observés comme fréquence, taille et type des nuages et leurs propriétés optiques (aérosols atmosphériques, al- bédo du sol, vapeur d’eau, poussière et turbidité atmosphérique) tandis qu’une séquence du rayonnement solaire peut être observée et mesurée à une station donnée. Ceci nous a suggéré de modéliser les processus de rayonnement solaire (ou d’indice de clarté) en utilisant un modèle Markovien caché (HMM), paire corrélée de processus stochastiques. Notre modèle principal est un HMM à temps continu (Xt, yt)t_0 est tel que (yt), le processus observé de rayonnement, soit une solution de l’équation différentielle stochastique (EDS) : dyt = [g(Xt)It − yt]dt + _(Xt)ytdWt, où It est le rayonnement extraterrestre à l’instant t, (Wt) est un mouvement Brownien standard et g(Xt), _(Xt) sont des fonctions de la chaîne de Markov non observée (Xt) modélisant la dynamique des régimes environnementaux. Pour ajuster nos modèles aux données réelles observées, les procédures d’estimation utilisent l’algorithme EM et la méthode du changement de mesures par le théorème de Girsanov. Des équations de filtrage sont établies et les équations à temps continu sont approchées par des versions robustes. Les modèles ajustés sont appliqués à des fins de comparaison et classification de distributions et de prédiction. / Characteristics of solar radiation highly depend on some unobserved meteorological events such as frequency, height and type of the clouds and their optical properties (atmospheric aerosols, ground albedo, water vapor, dust and atmospheric turbidity) while a sequence of solar radiation can be observed and measured at a given station. This has suggested us to model solar radiation (or clearness index) processes using a hidden Markov model (HMM), a pair of correlated stochastic processes. Our main model is a continuous-time HMM (Xt, yt)t_0 is such that the solar radiation process (yt)t_0 is a solution of the stochastic differential equation (SDE) : dyt = [g(Xt)It − yt]dt + _(Xt)ytdWt, where It is the extraterrestrial radiation received at time t, (Wt) is a standard Brownian motion and g(Xt), _(Xt) are functions of the unobserved Markov chain (Xt) modelling environmental regimes. To fit our models to observed real data, the estimation procedures combine the Expectation Maximization (EM) algorithm and the measure change method due to Girsanov theorem. Filtering equations are derived and continuous-time equations are approximated by robust versions. The models are applied to pdf comparison and classification and prediction purposes.

Les maladies altérant le processus de la déglutition sont multiples, affectant la qualité de vie du patient et sa capacité de fonctionner en société. La nature exacte et la gravité des changements post/pré-traitement dépendent de la localisation de l’anomalie. Une réadaptation efficace de la déglutition, cliniquement parlant, dépend généralement de l’inclusion d’une évaluation vidéo-fluoroscopique de la déglutition du patient dans l’évaluation post-traitement des patients en risque de fausse route. La restriction de cette utilisation est due au fait qu’elle est très invasive, comme d’autres moyens disponibles, tels que la fibre optique endoscopique. Ces méthodes permettent d’observer le déroulement de la déglutition et d’identifier les lieux de dysfonctionnement, durant ce processus, avec une précision élevée. "Mieux vaut prévenir que guérir" est le principe de base de la médecine en général. C’est dans ce contexte que se situe ce travail de thèse pour la télésurveillance des malades et plus spécifiquement pour suivre l’évolution fonctionnelle du processus de la déglutition chez des personnes à risques dysphagiques, que ce soit à domicile ou bien en institution, en utilisant le minimum de capteurs non-invasifs. C’est pourquoi le principal signal traité dans ce travail est le son. La principale problématique du traitement du signal sonore est la détection automatique du signal utile du son, étape cruciale pour la classification automatique de sons durant la prise alimentaire, en vue de la surveillance automatique. L’étape de la détection du signal utile permet de réduire la complexité du système d’analyse sonore. Les algorithmes issus de l’état de l’art traitant la détection du son de la déglutition dans le bruit environnemental n’ont pas montré une bonne performance. D’où l’idée d’utiliser un seuil adaptatif sur le signal, résultant de la décomposition en ondelettes. Les problématiques liées à la classification des sons en général et des sons de la déglutition en particulier sont abordées dans ce travail avec une analyse hiérarchique, qui vise à identifier dans un premier temps les segments de sons de la déglutition, puis à le décomposer en trois sons caractéristiques, ce qui correspond parfaitement à la physiologie du processus. Le couplage est également abordé dans ce travail. L’implémentation en temps réel de l’algorithme de détection a été réalisée. Cependant, celle de l’algorithme de classification reste en perspective. Son utilisation en clinique est prévue. / The diseases affecting and altering the swallowing process are multi-faceted, affecting the patient’s quality of life and ability to perform well in society. The exact nature and severity of the pre/post-treatment changes depend on the location of the anomaly. Effective swallowing rehabilitation, clinically depends on the inclusion of a video-fluoroscopic evaluation of the patient’s swallowing in the post-treatment evaluation. There are other available means such as endoscopic optical fibre. The drawback of these evaluation approaches is that they are very invasive. However, these methods make it possible to observe the swallowing process and identify areas of dysfunction during the process with high accuracy. "Prevention is better than cure" is the fundamental principle of medicine in general. In this context, this thesis focuses on remote monitoring of patients and more specifically monitoring the functional evolution of the swallowing process of people at risk of dysphagia, whether at home or in medical institutions, using the minimum number of non-invasive sensors. This has motivated the monitoring of the swallowing process based on the capturing only the acoustic signature of the process and modeling the process as a sequence of acoustic events occuring within a specific time frame. The main problem of such acoustic signal processing is the automatic detection of the relevent sound signals, a crucial step in the automatic classification of sounds during food intake for automatic monitoring. The detection of relevant signal reduces the complexity of the subsequent analysis and characterisation of a particular swallowing process. The-state-of-the-art algorithms processing the detection of the swallowing sounds as distinguished from environmental noise were not sufficiently accurate. Hence, the idea occured of using an adaptive threshold on the signal resulting from wavelet decomposition. The issues related to the classification of sounds in general and swallowing sounds in particular are addressed in this work with a hierarchical analysis that aims to first identify the swallowing sound segments and then to decompose them into three characteristic sounds, consistent with the physiology of the process. The coupling between detection and classification is also addressed in this work. The real-time implementation of the detection algorithm has been carried out. However, clinical use of the classification is discussed with a plan for its staged deployment subject to normal processes of clinical approval.

Investigations into the origin and evolution of regulatory mechanisms require quantitative estimates of the abundance and co-occurrence of functional protein domains among distantly related genomes. The metabolic and regulatory capabilities of an organism are implicit in its protein content. Currently available methods suffer for strong ascertainment biases, requiring methods for unbiased approaches to protein domain contents at genome-wide scales. The discussion will be highlighted on large scale patterns of similarities and differences of domain contains between phylum-level or even higher level taxonomic groups. This provides insights into large-scale evolutionary trends. The complement of recognizable functional protein domains and their combinations convey essentially the same information and at the same time are much more readily accessible, although protein domain models trained for one phylogenetic group frequently fail on distantly related sequences. Transcription factors (TF) typically cooperate to activate or repress the expression of genes. They play a critical role in developmental processes. While Chromatin Regulation (CR) facilitates DNA organization and prevent DNA aggregation and tangling which is important for replication, segregation, and gene expression. To compare the set of TFs and CRs between species, the genome annotation of equal quality was employed. However, the existing annotation suffers from bias in model organism. The similar count of transcripts are expected to be similar in mammals, but model organism such as human has more annotated transcripts than non model such as gorilla. Moreover, closely related species (e.g, dolphin and human) show a dramatically different distribution of TFs and CRs. Within vertebrates, this is unreasonable and contradicts phylogenetic knowledge. To overcome this problem, performing gene prediction followed by the detection of functional domains via HMM-based annotation of SCOP domains were proposed. This methods was demonstrated to lead toward consistent estimates for quantitative comparison. To emphasize the applicability, the protein domain distribution of putative TFs and CRs by quantitative and boolean means were analyzed. In particular, systematic studies of protein domain occurrences and co-occurrences to study avoidance or preferential co-occurrence of certain protein domains within TFs and CRs were utilized. Pooling related domain models based on their GO-annotation in combination with de novo gene prediction methods provides estimates that seem to be less affected by phylogenetic biases. it was shown for 18 diverse representatives from all eukaryotic kingdoms that a pooled analysis of the tendencies for co-occurrence or avoidance of protein domains is indeed feasible. This type of analysis can reveal general large-scale patterns in the domain co-occurrence and helps to identify lineage-specific variations in the evolution of protein domains. Somewhat surprisingly, strong ubiquitous patterns governing the evolutionary behavior of specific functional classes were not found. Instead, there are strong variations between the major groups of Eukaryotes, pointing at systematic differences in their evolutionary constraints. Species-specific training is required, however, to account for the genomic peculiarities in many lineages. In contrast to earlier studies wide-spread statistically significant avoidance of protein domains associated with distinct functional high-level gene-ontology terms were found.