L'objectif de cette thèse, menée dans un cadre industriel, est d'apparier des contenus textuels médiatiques. Plus précisément, il s'agit d'apparier à des articles de presse en ligne des vidéos pertinentes, pour lesquelles nous disposons d'une description textuelle. Notre problématique relève donc exclusivement de l'analyse de matériaux textuels, et ne fait intervenir aucune analyse d'image ni de langue orale. Surviennent alors des questions relatives à la façon de comparer des objets textuels, ainsi qu'aux critères mobilisés pour estimer leur degré de similarité. L'un de ces éléments est selon nous la similarité thématique de leurs contenus, autrement dit le fait que deux documents doivent relater le même sujet pour former une paire pertinente. Ces problématiques relèvent du domaine de la recherche d'information (ri), dans lequel nous nous ancrons principalement. Par ailleurs, lorsque l'on traite des contenus d'actualité, la dimension temporelle est aussi primordiale et les problématiques qui l'entourent relèvent de travaux ayant trait au domaine du topic detection and tracking (tdt) dans lequel nous nous inscrivons également.Le système d'appariement développé dans cette thèse distingue donc différentes étapes qui se complètent. Dans un premier temps, l'indexation des contenus fait appel à des méthodes de traitement automatique des langues (tal) pour dépasser la représentation classique des textes en sac de mots. Ensuite, deux scores sont calculés pour rendre compte du degré de similarité entre deux contenus : l'un relatif à leur similarité thématique, basé sur un modèle vectoriel de ri; l'autre à leur proximité temporelle, basé sur une fonction empirique. Finalement, un modèle de classification appris à partir de paires de documents, décrites par ces deux scores et annotées manuellement, permet d'ordonnancer les résultats.L'évaluation des performances du système a elle aussi fait l'objet de questionnements dans ces travaux de thèse. Les contraintes imposées par les données traitées et le besoin particulier de l'entreprise partenaire nous ont en effet contraints à adopter une alternative au protocole classique d'évaluation en ri, le paradigme de Cranfield. / The goal of this thesis, conducted within an industrial framework, is to pair textual media content. Specifically, the aim is to pair on-line news articles to relevant videos for which we have a textual description. The main issue is then a matter of textual analysis, no image or spoken language analysis was undertaken in the present study. The question that arises is how to compare these particular objects, the texts, and also what criteria to use in order to estimate their degree of similarity. We consider that one of these criteria is the topic similarity of their content, in other words, the fact that two documents have to deal with the same topic to form a relevant pair. This problem fall within the field of information retrieval (ir) which is the main strategy called upon in this research. Furthermore, when dealing with news content, the time dimension is of prime importance. To address this aspect, the field of topic detection and tracking (tdt) will also be explored.The pairing system developed in this thesis distinguishes different steps which complement one another. In the first step, the system uses natural language processing (nlp) methods to index both articles and videos, in order to overcome the traditionnal bag-of-words representation of texts. In the second step, two scores are calculated for an article-video pair: the first one reflects their topical similarity and is based on a vector space model; the second one expresses their proximity in time, based on an empirical function. At the end of the algorithm, a classification model learned from manually annotated document pairs is used to rank the results.Evaluation of the system's performances raised some further questions in this doctoral research. The constraints imposed both by the data and the specific need of the partner company led us to adapt the evaluation protocol traditionnal used in ir, namely the cranfield paradigm. We therefore propose an alternative solution for evaluating the system that takes all our constraints into account.

La mise en oeuvre de systèmes avancés d’aide à la conduite (ADAS) basée vision, est une tâche complexe et difficile surtout d’un point de vue robustesse en conditions d’utilisation réelles. Une des fonctionnalités des ADAS vise à percevoir et à comprendre l’environnement de l’ego-véhicule et à fournir l’assistance nécessaire au conducteur pour réagir à des situations d’urgence. Dans cette thèse, nous nous concentrons sur la détection et la reconnaissance des objets mobiles car leur dynamique les rend plus imprévisibles et donc plus dangereux. La détection de ces objets, l’estimation de leurs positions et la reconnaissance de leurs catégories sont importants pour les ADAS et la navigation autonome. Par conséquent, nous proposons de construire un système complet pour la détection des objets en mouvement et la reconnaissance basées uniquement sur les capteurs de vision. L’approche proposée permet de détecter tout type d’objets en mouvement en fonction de deux méthodes complémentaires. L’idée de base est de détecter les objets mobiles par stéréovision en utilisant l’image résiduelle du mouvement apparent (RIMF). La RIMF est définie comme l’image du mouvement apparent causé par le déplacement des objets mobiles lorsque le mouvement de la caméra a été compensé. Afin de détecter tous les mouvements de manière robuste et de supprimer les faux positifs, les incertitudes liées à l’estimation de l’ego-mouvement et au calcul de la disparité doivent être considérées. Les étapes principales de l’algorithme sont les suivantes : premièrement, la pose relative de la caméra est estimée en minimisant la somme des erreurs de reprojection des points d’intérêt appariées et la matrice de covariance est alors calculée en utilisant une stratégie de propagation d’erreurs de premier ordre. Ensuite, une vraisemblance de mouvement est calculée pour chaque pixel en propageant les incertitudes sur l’ego-mouvement et la disparité par rapport à la RIMF. Enfin, la probabilité de mouvement et le gradient de profondeur sont utilisés pour minimiser une fonctionnelle d’énergie de manière à obtenir la segmentation des objets en mouvement. Dans le même temps, les boîtes englobantes des objets mobiles sont générées en utilisant la carte des U-disparités. Après avoir obtenu la boîte englobante de l’objet en mouvement, nous cherchons à reconnaître si l’objet en mouvement est un piéton ou pas. Par rapport aux algorithmes de classification supervisée (comme le boosting et les SVM) qui nécessitent un grand nombre d’exemples d’apprentissage étiquetés, notre algorithme de boosting semi-supervisé est entraîné avec seulement quelques exemples étiquetés et de nombreuses instances non étiquetées. Les exemples étiquetés sont d’abord utilisés pour estimer les probabilités d’appartenance aux classes des exemples non étiquetés, et ce à l’aide de modèles de mélange de gaussiennes après une étape de réduction de dimension réalisée par une analyse en composantes principales. Ensuite, nous appliquons une stratégie de boosting sur des arbres de décision entraînés à l’aide des instances étiquetées de manière probabiliste. Les performances de la méthode proposée sont évaluées sur plusieurs jeux de données de classification de référence, ainsi que sur la détection et la reconnaissance des piétons. Enfin, l’algorithme de détection et de reconnaissances des objets en mouvement est testé sur les images du jeu de données KITTI et les résultats expérimentaux montrent que les méthodes proposées obtiennent de bonnes performances dans différents scénarios de conduite en milieu urbain. / Vision-based Advanced Driver Assistance Systems (ADAS) is a complex and challenging task in real world traffic scenarios. The ADAS aims at perceiving andunderstanding the surrounding environment of the ego-vehicle and providing necessary assistance for the drivers if facing some emergencies. In this thesis, we will only focus on detecting and recognizing moving objects because they are more dangerous than static ones. Detecting these objects, estimating their positions and recognizing their categories are significantly important for ADAS and autonomous navigation. Consequently, we propose to build a complete system for moving objects detection and recognition based on vision sensors. The proposed approach can detect any kinds of moving objects based on two adjacent frames only. The core idea is to detect the moving pixels by using the Residual Image Motion Flow (RIMF). The RIMF is defined as the residual image changes caused by moving objects with compensated camera motion. In order to robustly detect all kinds of motion and remove false positive detections, uncertainties in the ego-motion estimation and disparity computation should also be considered. The main steps of our general algorithm are the following : first, the relative camera pose is estimated by minimizing the sum of the reprojection errors of matched features and its covariance matrix is also calculated by using a first-order errors propagation strategy. Next, a motion likelihood for each pixel is obtained by propagating the uncertainties of the ego-motion and disparity to the RIMF. Finally, the motion likelihood and the depth gradient are used in a graph-cut-based approach to obtain the moving objects segmentation. At the same time, the bounding boxes of moving object are generated based on the U-disparity map. After obtaining the bounding boxes of the moving object, we want to classify the moving objects as a pedestrian or not. Compared to supervised classification algorithms (such as boosting and SVM) which require a large amount of labeled training instances, our proposed semi-supervised boosting algorithm is trained with only a few labeled instances and many unlabeled instances. Firstly labeled instances are used to estimate the probabilistic class labels of the unlabeled instances using Gaussian Mixture Models after a dimension reduction step performed via Principal Component Analysis. Then, we apply a boosting strategy on decision stumps trained using the calculated soft labeled instances. The performances of the proposed method are evaluated on several state-of-the-art classification datasets, as well as on a pedestrian detection and recognition problem.Finally, both our moving objects detection and recognition algorithms are tested on the public images dataset KITTI and the experimental results show that the proposed methods can achieve good performances in different urban scenarios.

[pt] Sistemas de computação paralelos vêm se tornando pervasivos, sendo usados para interagir com o mundo físico e processar uma grande quantidade de dados de várias fontes. É essencial, portanto, a melhora contínua do desempenho computacional para acompanhar o ritmo crescente da quantidade de informações que precisam ser processadas. Algumas dessas aplicações admitem uma menor qualidade no resultado final em troca do aumento do desempenho de execução. Este trabalho tem por objetivo avaliar a viabilidade de usar métodos de aprendizagem supervisionada para garantir que a técnica de Sincronização Relaxada, utilizada para o aumento do desempenho de execução, forneça resultados dentro de limites aceitáveis de erro. Para isso, criamos uma metodologia que utiliza alguns dados de entrada para montar casos de testes que, ao serem executados, irão fornecer valores representativos de entrada para o treinamento de métodos de aprendizagem supervisionada. Dessa forma, quando o usuário utilizar a sua aplicação (no mesmo ambiente de treinamento) com uma nova entrada, o algoritmo de classificação treinado irá sugerir o fator de relaxamento de sincronização mais adequado à tripla aplicação/entrada/ambiente de execução. Utilizamos essa metodologia em algumas aplicações paralelas bem conhecidas e mostramos que, aliando a Sincronização Relaxada a métodos de aprendizagem supervisionada, foi possível manter a taxa de erro máximo acordada. Além disso, avaliamos o ganho de desempenho obtido com essa técnica para alguns cenários em cada aplicação. / [en] Parallel computing systems have become pervasive, being used to interact with the physical world and process a large amount of data from various sources. It is essential, therefore, the continuous improvement of computational performance to keep up with the increasing rate of the amount of information that needs to be processed. Some of these applications admit lower quality in the final result in exchange for increased execution performance. This work aims to evaluate the feasibility of using supervised learning methods to ensure that the Relaxed Synchronization technique, used to increase execution performance, provides results within acceptable limits of error. To do so, we have created a methodology that uses some input data to assemble test cases that, when executed, will provide input values for the training of supervised learning methods. This way, when the user uses his/her application (in the same training environment) with a new input, the trained classification algorithm will suggest the relax synchronization factor that is best suited to the triple application/input/execution environment. We used this methodology insome well-known parallel applications and showed that, by combining Relaxed Synchronization with supervised learning methods, it was possible to maintain the maximum established error rate. In addition, we evaluated the performance gain obtained with this technique for a number of scenarios in each application.

Le glioblastome multiforme (GBM) est une tumeur de grade IV représentant 49% de toutes les tumeurs cérébrales. Malgré des modalités de traitement agressives (radiothérapie, chimiothérapie et résection chirurgicale), le pronostic est mauvais avec une survie globale médiane de 12 à 14 mois. Les aractéristiques issues de la neuro imagerie des GBM peuvent fournir de nouvelles opportunités pour la classification, le pronostic et le développement de nouvelles thérapies ciblées pour faire progresser la pratique clinique. Cette thèse se concentre sur le développement de modèles pronostiques exploitant des caractéristiques de radiomique extraites des images multimodales IRM (T1 pré- et post-contraste, T2 et FLAIR). Le contexte méthodologique proposé consiste à i) recaler tous les volumes multimodaux IRM disponibles et en segmenter un volume tumoral unique, ii) extraire des caractéristiques radiomiques et iii) construire et valider les modèles pronostiques par l’utilisation d’algorithmes d’apprentissage automatique exploitant des cohortes cliniques multicentriques de patients. Le coeur des méthodes développées est fondé sur l’extraction de radiomiques (incluant des paramètres d’intensité, de forme et de textures) pour construire des modèles pronostiques à l’aide de deux algorithmes d’apprentissage, les machines à vecteurs de support (support vector machines, SVM) et les forêts aléatoires (random forest, RF), comparées dans leur capacité à sélectionner et combiner les caractéristiques optimales. Les bénéfices et l’impact de plusieurs étapes de pré-traitement des images IRM (re-échantillonnage spatial des voxels, normalisation, segmentation et discrétisation des intensités) pour une extraction de métriques fiables ont été évalués. De plus les caractéristiques radiomiques ont été standardisées en participant à l’initiative internationale de standardisation multicentrique des radiomiques. La précision obtenue sur le jeu de test indépendant avec les deux algorithmes d’apprentissage SVM et RF, en fonction des modalités utilisées et du nombre de caractéristiques combinées atteignait 77 à 83% en exploitant toutes les radiomiques disponibles sans prendre en compte leur fiabilité intrinsèque, et 77 à 87% en n’utilisant que les métriques identifiées comme fiables.Dans cette thèse, un contexte méthodologique a été proposé, développé et validé, qui permet la construction de modèles pronostiques dans le cadre des GBM et de l’imagerie multimodale IRM exploitée par des algorithmes d’apprentissage automatique. Les travaux futurs pourront s’intéresser à l’ajout à ces modèles des informations contextuelles et génétiques. D’un point de vue algorithmique, l’exploitation de nouvelles techniques d’apprentissage profond est aussi prometteuse. / Glioblastoma multiforme (GBM) is a WHO grade IV tumor that represents 49% of ail brain tumours. Despite aggressive treatment modalities (radiotherapy, chemotherapy and surgical resections) the prognosis is poor, as médian overall survival (OS) is 12-14 months. GBM’s neuroimaging (non-invasive) features can provide opportunities for subclassification, prognostication, and the development of targeted therapies that could advance the clinical practice. This thesis focuses on developing a prognostic model based on multimodal MRI-derived (Tl pre- and post-contrast, T2 and FLAIR) radiomics in GBM. The proposed methodological framework consists in i) registering the available 3D multimodal MR images andsegmenting the tumor volume, ii) extracting radiomics iii) building and validating a prognostic model using machine learning algorithms applied to multicentric clinical cohorts of patients. The core component of the framework rely on extracting radiomics (including intensity, shape and textural metrics) and building prognostic models using two different machine learning algorithms (Support Vector Machine (SVM) and Random Forest (RF)) that were compared by selecting, ranking and combining optimal features. The potential benefits and respective impact of several MRI pre-processing steps (spatial resampling of the voxels, intensities quantization and normalization, segmentation) for reliable extraction of radiomics was thoroughly assessed. Moreover, the standardization of the radiomics features among methodological teams was done by contributing to “Multicentre Initiative for Standardisation of Radiomics”. The accuracy obtained on the independent test dataset using SVM and RF reached upto 83%- 77% when combining ail available features and upto 87%-77% when using only reliable features previously identified as robust, depending on number of features and modality. In this thesis, I developed a framework for developing a compréhensive prognostic model for patients with GBM from multimodal MRI-derived “radiomics and machine learning”. The future work will consists in building a unified prognostic model exploiting other contextual data such as genomics. In case of new algorithm development we look forward to develop the Ensemble models and deep learning-based techniques.

Submitted by Alison Vanceto (alison-vanceto@hotmail.com) on 2017-01-03T12:49:19Z No. of bitstreams: 1 TeseMCD.pdf: 1564245 bytes, checksum: fbb9eb1099a1b38351371c97e8e49bb4 (MD5) / Approved for entry into archive by Marina Freitas (marinapf@ufscar.br) on 2017-01-16T16:47:27Z (GMT) No. of bitstreams: 1 TeseMCD.pdf: 1564245 bytes, checksum: fbb9eb1099a1b38351371c97e8e49bb4 (MD5) / Approved for entry into archive by Marina Freitas (marinapf@ufscar.br) on 2017-01-16T16:47:38Z (GMT) No. of bitstreams: 1 TeseMCD.pdf: 1564245 bytes, checksum: fbb9eb1099a1b38351371c97e8e49bb4 (MD5) / Made available in DSpace on 2017-01-16T16:47:46Z (GMT). No. of bitstreams: 1 TeseMCD.pdf: 1564245 bytes, checksum: fbb9eb1099a1b38351371c97e8e49bb4 (MD5) Previous issue date: 2016-01-04 / Não recebi financiamento / NELL is a computer system that has the goal of learn to learn 24 hours per day, continuously and learn more an better than the last day, to perform the knowledge base (KB). NELL is running since January 12 of 2010. Furthermore, NELL goals is have hight precision to be able to continue the learning. NELL is developed in macro-reading context, because this NELL needs very much redundancy to run. The first step to run NELL is to have an big (all-pairs-data). An all-pairs-data is a preprocessed base using Natural Language Processing (NLP), that base has all sufficient statistics about a corpus of web pages. The proposal of this project was to create a instance of NELL (currently in English) in Portuguese. For this, the first goal was the developing an all-pairs-data in Portuguese. The second step was to create a new version of Portuguese NELL. And finally, the third goal was to develop a coreference resolution hybrid method focused in features semantics and morphologics. This method is not dependent of a specific language, it is can be applied for another languages with the same alphabet of Portuguese language. The NELL in Portuguese was developed, but the all-pairs-data is not big enough. Because it Portuguese NELL is not running for ever, like the English version. Even so, this project present the steps about how to develop a NELL in other language and some ideas about how to improve the all-pairs-data. By the way, this project present a coreference resolution hybrid method with good results to NELL. / A NELL é um sistema de computador que possui o objetivo de executar 24 horas por dia, 7 dias por semana, sem parar. A versão atual da NELL foi iniciada em 12 de Janeiro de 2010 e continua ativa. Seu objetivo é aprender cada vez mais fatos da web para popular sua base de conhecimento (Knowlegde Base - KB). Além de aprender cada vez mais, a NELL também objetiva alcançar alta confiança no aprendizado para garantir a continuidade do aprendizado. A NELL foi desenvolvida e atua no contexto da macroleitura, no qual é necessária uma grande quantidade e redundância de dados. Para que o sistema possa aprender, o primeiro passo é criar uma base preprocessada (all-pairs-data) a partir do uso de técnicas linguísticas. O all-pairs-data deve possuir todas as estatísticas suficientes para a execução da NELL e também deve ser de um tamanho suficientemente grande para que o aprendizado possa ocorrer. Neste projeto, foi proposta a criação de uma nova instância da NELL em português. Inicialmente foi proposta a criação de um all-pairs-data e, em seguida, a criação de uma abordagem híbrida para a resolução de correferências independente de língua por base em características semânticas e morfológicas. A proposta híbrida objetivou aperfeiçoar o processo atual de tratamento de correferências na NELL, melhorando assim a confiabilidade no aprendizado. Todas as propostas foram desenvolvidas e a NELL em português obteve bons resultados. Tais resultados evidenciam que a leitura da web em português poderá se tornar um sistema de aprendizado sem-fim. Para que isso ocorra são também apresentadas as futuras abordagens e propostas. Além disso, este projeto apresenta a metodologia de criação da instância da NELL em português, uma proposta de resolução de correferência que explora atributos linguisticos,bem como a ontologia da NELL, além de apontar trabalhos futuros, nos quais inclui-se processos de adição de outras línguas na NELL, principalmente para aquelas que possuem poucas páginas web disponíveis para o aprendizado.

Le suivi, le contrôle et la surveillance (MCS) des pêches marines sont des problèmes essentiels pour la gestion durable des ressources halieutiques. Dans cette thèse, nous étudions le suivi spatial des activités des navires de pêche en utilisant les données de trajectoire du système de surveillance des navires (VMS) dans le cadre du projet INDESO (2013-2017). Notre objectif général est de développer une chaîne de traitement des données VMS afin de: i) effectuer un suivi de l'effort de pêche des flottilles de palangriers indonésiens, ii) détecter les activités de pêche illégales et évaluer leur importance. L'approche proposée repose sur des modèles de mélange gaussien (GMM) et les modèles de Markov cachés (HMM), en vue d'identifier les comportements élémentaires des navires de pêche, tels que les voyages, la recherche et les activités de pêche, dans un cadre non supervisé. Nous considérons différentes paramétrisations de ces modèles avec une étude particulière des palangriers indonésiens, pour lesquels nous pouvons bénéficier de données d'observateurs embarqués afin de procéder à une évaluation quantitative des modèles proposés et testés.Nous exploitons ensuite ces modèles statistiques pour deux objectifs différents: a) la discrimination des différents flottilles de pêche à partir des trajectoires des navires de pêche et l'application à la détection et à l'évaluation des activités de pêche illégale, b) l'évaluation d'un effort de pêche spatialisé à partir des données VMS. Nous obtenons de très bons taux de reconnaissance (environ 97%) pour la première tâche et nos expériences soutiennent le potentiel d'une exploration opérationnelle de l'approche proposée. En raison du nombre limité de données d'observateurs embarqués, seules des analyses préliminaires on pu être effectuées pour l'estimation de l'effort de pêche à partir des données VMS. Au-delà des développements méthodologiques potentiels, cette thèse met l'accent sur l'importance de la qualité de données d'observation en mer représentatives pour développer davantage l'exploitation des données VMS tant pour la recherche que pour les questions opérationnelles. / Monitoring, control and surveillance (MCS) of marine fisheries are critical issues for the sustainable management of marine fisheries. In this thesis we investigate the space-based monitoring of fishing vessel activities using Vessel Monitoring System (VMS) trajectory data in the context of INDESO project (2013-2017). Our general objective is to develop a processing chain of VMS data in order to: i) perform a follow-up of the fishing effort of the Indonesian longline fleets, ii) detect illegal fishing activities and assess their importance. The proposed approach relies on classical latent class models, namely Gaussian Mixture Models (GMM) and Hidden Markov Models (HMM), with a view to identifying elementary fishing vessel behaviors, such as travelling, searching and fishing activities, in a unsupervised framework. Following state-of-the-art approaches, we consider different parameterizations of these models with a specific focus on Indonesian longliners, for which we can benefit from at-sea observers¿ data to proceed to a quantitative evaluation. We then exploit these statistical models for two different objectives: a) the discrimination of different fishing fleets from fishing vessel trajectories and the application to the detection and assessment of illegal fishing activities, b) the assessment of a spatialized fishing effort from VMS data. We report good recognition rate (about 97%) for the former task and our experiments support the potential for an operational exploration of the proposed approach. Due to limited at-sea observers¿ data, only preliminary analyses could be carried out for the proposed VMS-derived fishing effort. Beyond potential methodological developments, this thesis emphasizes the importance of high-quality and representative at-sea observer data for further developing the exploitation of VMS data both for research and operational issues.

Supervised learning of piecewise linear models is a well studied problem in machine learning community. The key idea in piecewise linear modeling is to properly partition the input space and learn a linear model for every partition. Decision trees and regression trees are classic examples of piecewise linear models for classification and regression problems. The existing approaches for learning decision/regression trees can be broadly classified in to two classes, namely, fixed structure approaches and greedy approaches. In the fixed structure approaches, tree structure is fixed before hand by fixing the number of non leaf nodes, height of the tree and paths from root node to every leaf node of the tree. Mixture of experts and hierarchical mixture of experts are examples of fixed structure approaches for learning piecewise linear models. Parameters of the models are found using, e.g., maximum likelihood estimation, for which expectation maximization(EM) algorithm can be used. Fixed structure piecewise linear models can also be learnt using risk minimization under an appropriate loss function. Learning an optimal decision tree using fixed structure approach is a hard problem. Constructing an optimal binary decision tree is known to be NP Complete. On the other hand, greedy approaches do not assume any parametric form or any fixed structure for the decision tree classifier. Most of the greedy approaches learn tree structured piecewise linear models in a top down fashion. These are built by binary or multi-way recursive partitioning of the input space. The main issues in top down decision tree induction is to choose an appropriate objective function to rate the split rules. The objective function should be easy to optimize. Top-down decision trees are easy to implement and understand, but there are no optimality guarantees due to their greedy nature. Regression trees are built in the similar way as decision trees. In regression trees, every leaf node is associated with a linear regression function. All piece wise linear modeling techniques deal with two main tasks, namely, partitioning of the input space and learning a linear model for every partition. However, Partitioning of the input space and learning linear models for different partitions are not independent problems. Simultaneous optimal estimation of partitions and learning linear models for every partition, is a combinatorial problem and hence computationally hard. However, piecewise linear models provide better insights in to the classification or regression problem by giving explicit representation of the structure in the data. The information captured by piecewise linear models can be summarized in terms of simple rules, so that, they can be used to analyze the properties of the domain from which the data originates. These properties make piecewise linear models, like decision trees and regression trees, extremely useful in many data mining applications and place them among top data mining algorithms. In this thesis, we address the problem of supervised learning of piecewise linear models for classification and regression. We propose novel algorithms for learning piecewise linear classifiers and regression functions. We also address the problem of noise tolerant learning of classifiers in presence of label noise. We propose a novel algorithm for learning polyhedral classifiers which are the simplest form of piecewise linear classifiers. Polyhedral classifiers are useful when points of positive class fall inside a convex region and all the negative class points are distributed outside the convex region. Then the region of positive class can be well approximated by a simple polyhedral set. The key challenge in optimally learning a fixed structure polyhedral classifier is to identify sub problems, where each sub problem is a linear classification problem. This is a hard problem and identifying polyhedral separability is known to be NP complete. The goal of any polyhedral learning algorithm is to efficiently handle underlying combinatorial problem while achieving good classification accuracy. Existing methods for learning a fixed structure polyhedral classifier are based on solving non convex constrained optimization problems. These approaches do not efficiently handle the combinatorial aspect of the problem and are computationally expensive. We propose a method of model based estimation of posterior class probability to learn polyhedral classifiers. We solve an unconstrained optimization problem using a simple two step algorithm (similar to EM algorithm) to find the model parameters. To the best of our knowledge, this is the first attempt to form an unconstrained optimization problem for learning polyhedral classifiers. We then modify our algorithm to find the number of required hyperplanes also automatically. We experimentally show that our approach is better than the existing polyhedral learning algorithms in terms of training time, performance and the complexity. Most often, class conditional densities are multimodal. In such cases, each class region may be represented as a union of polyhedral regions and hence a single polyhedral classifier is not sufficient. To handle such situation, a generic decision tree is required. Learning optimal fixed structure decision tree is a computationally hard problem. On the other hand, top-down decision trees have no optimality guarantees due to the greedy nature. However, top-down decision tree approaches are widely used as they are versatile and easy to implement. Most of the existing top-down decision tree algorithms (CART,OC1,C4.5, etc.) use impurity measures to assess the goodness of hyper planes at each node of the tree. These measures do not properly capture the geometric structures in the data. We propose a novel decision tree algorithm that ,at each node, selects hyperplanes based on an objective function which takes into consideration geometric structure of the class regions. The resulting optimization problem turns out to be a generalized eigen value problem and hence is efficiently solved. We show through empirical studies that our approach leads to smaller size trees and better performance compared to other top-down decision tree approaches. We also provide some theoretical justification for the proposed method of learning decision trees. Piecewise linear regression is similar to the corresponding classification problem. For example, in regression trees, each leaf node is associated with a linear regression model. Thus the problem is once again that of (simultaneous) estimation of optimal partitions and learning a linear model for each partition. Regression trees, hinge hyperplane method, mixture of experts are some of the approaches to learn continuous piecewise linear regression models. Many of these algorithms are computationally intensive. We present a method of learning piecewise linear regression model which is computationally simple and is capable of learning discontinuous functions as well. The method is based on the idea of K plane regression that can identify a set of linear models given the training data. K plane regression is a simple algorithm motivated by the philosophy of k means clustering. However this simple algorithm has several problems. It does not give a model function so that we can predict the target value for any given input. Also, it is very sensitive to noise. We propose a modified K plane regression algorithm which can learn continuous as well as discontinuous functions. The proposed algorithm still retains the spirit of k means algorithm and after every iteration it improves the objective function. The proposed method learns a proper Piece wise linear model that can be used for prediction. The algorithm is also more robust to additive noise than K plane regression. While learning classifiers, one normally assumes that the class labels in the training data set are noise free. However, in many applications like Spam filtering, text classification etc., the training data can be mislabeled due to subjective errors. In such cases, the standard learning algorithms (SVM, Adaboost, decision trees etc.) start over fitting on the noisy points and lead to poor test accuracy. Thus analyzing the vulnerabilities of classifiers to label noise has recently attracted growing interest from the machine learning community. The existing noise tolerant learning approaches first try to identify the noisy points and then learn classifier on remaining points. In this thesis, we address the issue of developing learning algorithms which are inherently noise tolerant. An algorithm is inherently noise tolerant if, the classifier it learns with noisy samples would have the same performance on test data as that learnt from noise free samples. Algorithms having such robustness (under suitable assumption on the noise) are attractive for learning with noisy samples. Here, we consider non uniform label noise which is a generic noise model. In non uniform label noise, the probability of the class label for an example being incorrect, is a function of the feature vector of the example.(We assume that this probability is less than 0.5 for all feature vectors.) This can account for most cases of noisy data sets. There is no provably optimal algorithm for learning noise tolerant classifiers in presence of non uniform label noise. We propose a novel characterization of noise tolerance of an algorithm. We analyze noise tolerance properties of risk minimization frame work as risk minimization is a common strategy for classifier learning. We show that risk minimization under 01 loss has the best noise tolerance properties. None of the other convex loss functions have such noise tolerance properties. Empirical risk minimization under 01 loss is a hard problem as 01 loss function is not differentiable. We propose a gradient free stochastic optimization technique to minimize risk under 01 loss function for noise tolerant learning of linear classifiers. We show (under some conditions) that the algorithm converges asymptotically to the global minima of the risk under 01 loss function. We illustrate the noise tolerance of our algorithm through simulations experiments. We demonstrate the noise tolerance of the algorithm through simulations.

[pt] Predição de porosidade em reservatórios de óleo e gás representa em uma tarefa crucial e desafiadora na indústria de petróleo. Neste trabalho é proposto um novo modelo não-linear para predição de porosidade que trata fácies sedimentares como variáveis ocultas ou latentes. Esse modelo, denominado Transductive Conditional Random Field Regression (TCRFR), combina com sucesso os conceitos de Markov random fields, ridge regression e aprendizado transdutivo. O modelo utiliza volumes de impedância sísmica como informação de entrada condicionada aos valores de porosidade disponíveis nos poços existentes no reservatório e realiza de forma simultânea e automática a classificação das fácies e a estimativa de porosidade em todo o volume. O método é capaz de inferir as fácies latentes através da combinação de amostras precisas de porosidade local presentes nos poços com dados de impedância sísmica ruidosos, porém disponíveis em todo o volume do reservatório. A informação precisa de porosidade é propagada no volume através de modelos probabilísticos baseados em grafos, utilizando conditional random fields. Adicionalmente, duas novas técnicas são introduzidas como etapas de pré-processamento para aplicação do método TCRFR nos casos extremos em que somente um número bastante reduzido de amostras rotuladas de porosidade encontra-se disponível em um pequeno conjunto de poços exploratórios, uma situação típica para geólogos durante a fase exploratória de uma nova área. São realizados experimentos utilizando dados de um reservatório sintético e de um reservatório real. Os resultados comprovam que o método apresenta um desempenho consideravelmente superior a outros métodos automáticos de predição em relação aos dados sintéticos e, em relação aos dados reais, um desempenho comparável ao gerado por técnicas tradicionais de geo estatística que demandam grande esforço manual por parte de especialistas. / [en] Estimating porosity in oil and gas reservoirs is a crucial and challenging task in the oil industry. A novel nonlinear model for porosity estimation is proposed, which handles sedimentary facies as latent variables. It successfully combines the concepts of conditional random fields (CRFs), transductive learning and ridge regression. The proposed Transductive Conditional Random Field Regression (TCRFR) uses seismic impedance volumes as input information, conditioned on the porosity values from the available wells in the reservoir, and simultaneously and automatically provides as output the porosity estimation and facies classification in the whole volume. The method is able to infer the latent facies states by combining the local, labeled and accurate porosity information available at well locations with the plentiful but imprecise impedance information available everywhere in the reservoir volume. That accurate information is propagated in the reservoir based on conditional random field probabilistic graphical models, greatly reducing uncertainty. In addition, two new techniques are introduced as preprocessing steps for the application of TCRFR in the extreme but realistic cases where just a scarce amount of porosity labeled samples are available in a few exploratory wells, a typical situation for geologists during the evaluation of a reservoir in the exploration phase. Both synthetic and real-world data experiments are presented to prove the usefulness of the proposed methodology, which show that it outperforms previous automatic estimation methods on synthetic data and provides a comparable result to the traditional manual labored geostatistics approach on real-world data.

Dans notre communication quotidienne avec les autres, nous avons autant de considération pour l’interlocuteur lui-même que pour l’information transmise. En permanence coexistent en effet deux modes de transmission : le verbal et le non-verbal. Sur ce dernier thème intervient principalement l’expression faciale avec laquelle l’interlocuteur peut révéler d’autres émotions et intentions. Habituellement, un processus de reconnaissance d’émotions faciales repose sur 3 étapes : le suivi du visage, l’extraction de caractéristiques puis la classification de l’expression faciale. Pour obtenir un processus robuste apte à fournir des résultats fiables et exploitables, il est primordial d’extraire des caractéristiques avec de forts pouvoirs discriminants (selon les zones du visage concernées). Les avancées récentes de l’état de l’art ont conduit aujourd’hui à diverses approches souvent bridées par des temps de traitement trop couteux compte-tenu de l’extraction de descripteurs sur le visage complet ou sur des heuristiques mathématiques et/ou géométriques.En fait, aucune réponse bio-inspirée n’exploite la perception humaine dans cette tâche qu’elle opère pourtant régulièrement. Au cours de ces travaux de thèse, la base de notre approche fut ainsi de singer le modèle visuel pour focaliser le calcul de nos descripteurs sur les seules régions du visage essentielles pour la reconnaissance d’émotions. Cette approche nous a permis de concevoir un processus plus naturel basé sur ces seules régions émergentes au regard de la perception humaine. Ce manuscrit présente les différentes méthodologies bio-inspirées mises en place pour aboutir à des résultats qui améliorent généralement l’état de l’art sur les bases de référence. Ensuite, compte-tenu du fait qu’elles se focalisent sur les seules parties émergentes du visage, elles améliorent les temps de calcul et la complexité des algorithmes mis en jeu conduisant à une utilisation possible pour des applications temps réel. / Communication in any form i.e. verbal or non-verbal is vital to complete various daily routine tasks and plays a significant role inlife. Facial expression is the most effective form of non-verbal communication and it provides a clue about emotional state, mindset and intention. Generally automatic facial expression recognition framework consists of three step: face tracking, feature extraction and expression classification. In order to built robust facial expression recognition framework that is capable of producing reliable results, it is necessary to extract features (from the appropriate facial regions) that have strong discriminative abilities. Recently different methods for automatic facial expression recognition have been proposed, but invariably they all are computationally expensive and spend computational time on whole face image or divides the facial image based on some mathematical or geometrical heuristic for features extraction. None of them take inspiration from the human visual system in completing the same task. In this research thesis we took inspiration from the human visual system in order to find from where (facial region) to extract features. We argue that the task of expression analysis and recognition could be done in more conducive manner, if only some regions are selected for further processing (i.e.salient regions) as it happens in human visual system. In this research thesis we have proposed different frameworks for automatic recognition of expressions, all getting inspiration from the human vision. Every subsequently proposed addresses the shortcomings of the previously proposed framework. Our proposed frameworks in general, achieve results that exceeds state-of-the-artmethods for expression recognition. Secondly, they are computationally efficient and simple as they process only perceptually salient region(s) of face for feature extraction. By processing only perceptually salient region(s) of the face, reduction in feature vector dimensionality and reduction in computational time for feature extraction is achieved. Thus making them suitable for real-time applications.

O aprendizado de máquina consiste de conceitos e técnicas que permitem aos computadores melhorar seu desempenho com a experiência, ou, em outras palavras, aprender com dados. Um dos principais tópicos do aprendizado de máquina é o agrupamento de dados que, como o nome sugere, procura agrupar os dados de acordo com sua similaridade. Apesar de sua definição relativamente simples, o agrupamento é uma tarefa computacionalmente complexa, tornando proibitivo o emprego de algoritmos exaustivos, na busca pela solução ótima do problema. A importância do agrupamento de dados, aliada aos seus desafios, faz desse campo um ambiente de intensa pesquisa. Também a classe de fenômenos naturais conhecida como comportamento coletivo tem despertado muito interesse. Isso decorre da observação de um estado organizado e global que surge espontaneamente das interações locais presentes em grandes grupos de indivíduos, caracterizando, pois, o que se chama auto-organização ou emergência, para ser mais preciso. Os desafios intrínsecos e a relevância do tema vêm motivando sua pesquisa em diversos ramos da ciência e da engenharia. Ao mesmo tempo, técnicas baseadas em comportamento coletivo vêm sendo empregadas em tarefas de aprendizado de máquina, mostrando-se promissoras e ganhando bastante atenção. No presente trabalho, objetivou-se o desenvolvimento de técnicas de agrupamento baseadas em comportamento coletivo. Faz-se cada item do conjunto de dados corresponder a um indivíduo, definem-se as leis de interação local, e então os indivíduos são colocados a interagir entre si, de modo que os padrões que surgem reflitam os padrões originalmente presentes no conjunto de dados. Abordagens baseadas em dinâmica de troca de energia foram propostas. Os dados permanecem fixos em seu espaço de atributos, mas carregam certa informação a energia , a qual é progressivamente trocada entre eles. Os grupos são estabelecidos entre dados que tomam estados de energia semelhantes. Este trabalho abordou também o aprendizado semissupervisionado, cuja tarefa é rotular dados em bases parcialmente rotuladas. Nesse caso, foi adotada uma abordagem baseada na movimentação dos próprios dados pelo espaço de atributos. Procurou-se, durante todo este trabalho, não apenas propor novas técnicas de aprendizado, mas principalmente, por meio de muitas simulações e ilustrações, mostrar como elas se comportam em diferentes cenários, num esforço em mostrar onde reside a vantagem de se utilizar a dinâmica coletiva na concepção dessas técnicas / Machine learning consists of concepts and techniques that enable computers to improve their performance with experience, i.e., enable computers to learn from data. Data clustering (or just clustering) is one of its main topics, which aims to group data according to their similarities. Regardless of its simple definition, clustering is a complex computational task. Its relevance and challenges make this field an environment of intense research. The class of natural phenomena known as collective behavior has also attracted much interest. This is due to the observation that global patterns may spontaneously arise from local interactions among large groups of individuals, what is know as self-organization (or emergence). The challenges and relevance of the subject are encouraging its research in many branches of science and engineering. At the same time, techniques based on collective behavior are being employed in machine learning tasks, showing to be promising. The objective of the present work was to develop clustering techniques based on collective behavior. Each dataset item corresponds to an individual. Once the local interactions are defined, the individuals begin to interact with each other. It is expected that the patterns arising from these interactions match the patterns originally present in the dataset. Approaches based on dynamics of energy exchange have been proposed. The data are kept fixed in their feature space, but they carry some sort of information (the energy), which is progressively exchanged among them. The groups are established among data that take similar energy states. This work has also addressed the semi-supervised learning task, which aims to label data in partially labeled datasets. In this case, it has been proposed an approach based on the motion of the data themselves around the feature space. More than just providing new machine learning techniques, this research has tried to show how the techniques behave in different scenarios, in an effort to show where lies the advantage of using collective dynamics in the design of such techniques