Making personalized recommendations has become a central part in many platforms, and is continuing to grow with more access to massive amounts of data online. Giving recommendations based on the interests of the individual, rather than recommending items that are popular, increases the user experience and can potentially attract more customers when done right. In order to make personalized recommendations, many platforms resort to machine learning algorithms. In the context of food recipes, these machine learning algorithms tend to consist of hybrid methods between collaborative filtering, content-based methods and matrix factorization. Most content-based approaches are ingredient based and can be very fruitful. However, fetching every single ingredient for recipes and processing them can be computationally expensive. Therefore, this paper investigates if clustering recipes according to what cuisine they belong to and what the main protein is can also improve rating predictions compared to when only collaborative filtering and matrix factorization methods are employed. This suggested content-based approach has a structure of a hierarchical classification, where recipes are first clustered into what cuisine group they belong to, then the specific cuisine and finally what the main protein is. The results suggest that the content-based approach can improve the predictions slightly but not significantly, and can help reduce the sparsity of the rating matrix to some extent. However, it suffers from heavily sparse data with respect to how many rating predictions it can give. / Att ge personliga rekommendationer har blivit en central del av många plattformar och fortsätter att bli det då tillgången till stora mängder data har ökat. Genom att ge personliga rekommendationer baserat på användares intressen, istället för att rekommendera det som är populärt, förbättrar användarupplevelsen och kan attrahera fler kunder. För att kunna producera personliga rekommendationer så vänder sig många plattformar till maskininlärningsalgoritmer. När det kommer till matrecept, så brukar dessa maskininlärningsalgoritmer bestå av hybrida metoder som sammanfogar collaborative filtering, innehållsbaserande metoder och matrisfaktorisering. De flesta innehållsbaserande metoderna baseras på ingredienser och har visats vara effektiva. Däremot, så kan det vara kostsamt för datorer att ta hänsyn till varenda ingrediens i varje matrecept. Därför undersöker denna artikel om att klassificera recept hierarkiskt efter matkultur och huvudprotein också kan förbättra rekommendationer när bara collaborative filtering och matrisfaktorisering används. Denna innehållsbaserande metod har en struktur av hierarkisk klassificering, där recept först indelas efter matkultur, specifik matkultur och till slut vad huvudproteinet är. Resultaten visar att innehållsbaserande metoden kan förbättra receptförslagen, men inte på en statistisk signifikant nivå, och kan reducera gleshet i en matris med tillsatta betyg från olika användare med olika recept något. Däremot så påverkas den ansenligt när det är glest med tillgänglighet av data. / Eatit

Traditional multi-class classification problems assume that each instance is associated with a single label from category set Y where |Y| > 2. Multi-label classification generalizes multi-class classification by allowing each instance to be associated with multiple labels from Y. In many real world data analysis problems, data objects can be assigned into multiple categories and hence produce multi-label classification problems. For example, an image for object categorization can be labeled as 'desk' and 'chair' simultaneously if it contains both objects. A news article talking about the effect of Olympic games on tourism industry might belong to multiple categories such as 'sports', 'economy', and 'travel', since it may cover multiple topics. Regardless of the approach used, multi-label learning in general requires a sufficient amount of labeled data to recover high quality classification models. However due to the label sparsity, i.e. each instance only carries a small number of labels among the label set Y, it is difficult to prepare sufficient well-labeled data for each class. Many approaches have been developed in the literature to overcome such challenge by exploiting label correlation or label dependency. In this dissertation, we propose a probabilistic model to capture the pairwise interaction between labels so as to alleviate the label sparsity. Besides of the traditional setting that assumes training data is fully labeled, we also study multi-label learning under other scenarios. For instance, training data can be unreliable due to missing values. A conditional Restricted Boltzmann Machine (CRBM) is proposed to take care of such challenge. Furthermore, labeled training data can be very scarce due to the cost of labeling but unlabeled data are redundant. We proposed two novel multi-label learning algorithms under active setting to relieve the pain, one for standard single level problem and one for hierarchical problem. Our empirical results on multiple multi-label data sets demonstrate the efficacy of the proposed methods. / Computer and Information Science

Submitted by JOSIANE SANTOS DE OLIVEIRA (josianeso) on 2019-03-07T14:41:05Z No. of bitstreams: 1 Roger Alan Stein_.pdf: 476239 bytes, checksum: a87a32ffe84d0e5d7a882e0db7b03847 (MD5) / Made available in DSpace on 2019-03-07T14:41:05Z (GMT). No. of bitstreams: 1 Roger Alan Stein_.pdf: 476239 bytes, checksum: a87a32ffe84d0e5d7a882e0db7b03847 (MD5) Previous issue date: 2018-03-28 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Efficient distributed numerical word representation models (word embeddings) combined with modern machine learning algorithms have recently yielded considerable improvement on automatic document classification tasks. However, the effectiveness of such techniques has not been assessed for the hierarchical text classification (HTC) yet. This study investigates application of those models and algorithms on this specific problem by means of experimentation and analysis. Classification models were trained with prominent machine learning algorithm implementations—fastText, XGBoost, and Keras’ CNN—and noticeable word embeddings generation methods—GloVe, word2vec, and fastText—with publicly available data and evaluated them with measures specifically appropriate for the hierarchical context. FastText achieved an LCAF1 of 0.871 on a single-labeled version of the RCV1 dataset. The results analysis indicates that using word embeddings is a very promising approach for HTC. / Modelos eficientes de representação numérica textual (word embeddings) combinados com algoritmos modernos de aprendizado de máquina têm recentemente produzido uma melhoria considerável em tarefas de classificação automática de documentos. Contudo, a efetividade de tais técnicas ainda não foi avaliada com relação à classificação hierárquica de texto. Este estudo investiga a aplicação daqueles modelos e algoritmos neste problema em específico através de experimentação e análise. Modelos de classificação foram treinados usando implementações proeminentes de algoritmos de aprendizado de máquina—fastText, XGBoost e CNN (Keras)— e notórios métodos de geração de word embeddings—GloVe, word2vec e fastText—com dados disponíveis publicamente e avaliados usando métricas especificamente adequadas ao contexto hierárquico. Nesses experimentos, fastText alcançou um LCAF1 de 0,871 usando uma versão da base de dados RCV1 com apenas uma categoria por tupla. A análise dos resultados indica que a utilização de word embeddings é uma abordagem muito promissora para classificação hierárquica de texto.

Orientadora: Profa. Dra. Paula Homem de Mello / Tese (doutorado) - Universidade Federal do ABC. Programa de Pós-Graduação em Ciência e Tecnologia/Química, 2017. / A preocupação com a formação de professores remonta à antiguidade, e seus estudos vêm sendo sistematizados desde então. Ela é importante posto que impacta diretamente na relação ensino/aprendizado dos alunos, e assim no processo emancipatório proporcionado por uma educação significativa. As universidades são, contemporaneamente, as responsáveis pela formação dos professores e cada uma à sua maneira tenta desempenhar esse papel com excelência. A UFABC se destaca nesse cenário ao propor um currículo interdisciplinar que atende às demandas atuais por generalização e qualidade. Nesse contexto, se torna necessário entender como se dá o processo educacional nessa instituição e também se esse currículo traz resultados distintos de outras instituições tradicionais. Para isso foi proposto uma pesquisa mista comparando os licenciandos de química da UFABC e da Universidade de São Paulo, quanto ao seu Conhecimento Pedagógico do Conteúdo (PCK), dado que esse é o conhecimento que distingue o professor do especialista. A pesquisa proposta visou inicialmente os modelos de PCK existentes por uma Classificação Hierárquica Decimal (DHC) e um Análise Fatorial de Correspondência (FAC), para então discutir duas tipologias de um conhecimento que desde 2012 deixou de ser considerado parte desse conhecimento, as Orientações para o Ensino, e sua relação com a forma com que o professor interage com seus alunos (Interação Discursiva). Com isso em mente, a forma pela qual se dá o desenvolvimento do PCK, na UFABC, foi o foco de um estudo de caso que observou a modificação nos Mapas de PCK. Finalmente um estudo comparativo entre as duas instituições foi desenvolvido por um Modelo Hierárquico Linear (HLM), utilizando um questionário validado pelo Modelo de Rasch. Os resultados mostraram que os discursos de PCK têm se aproximado desde 2012 e compartilham um vocabulário comum, e ainda que os modelos de PCK consideram esse conhecimento como sendo tópico específico. As duas tipologias das Orientações se mostraram relacionadas e assim fornecendo-as de subsídios empíricos. Também foi possível demonstrar que elas se relacionam com as Interações Discursivas quando a aula foi dividida em segmentos (começo, meio e fim). O estudo de caso mostrou que o professor universitário guiou os licenciandos para o desenvolvimento de seu PCK por meio de atividades, mas que quando fez apontamentos orais individualizados esse impacto foi maior. Por fim, o questionário proposto se mostrou confiável e capaz de distinguir a habilidade dos respondentes, no entanto, não foi possível distinguir significativamente dois grupos de respondentes dado a alta variabilidade dos respondentes da UFABC. Esse resultado é compatível com o currículo da referida instituição e dificulta a comparação com outros respondentes. / The concern with teacher formation dates back to antiquity, and its studies has been systematized ever since. Its importance is due to its direct impact on student¿s learning, and thus in the emancipatory process provided by a meaningful education. The universities are, nowadays, the responsible for teachers formation and each, on its way, tries to play this role with excellence. UFABC stands out in this scenario as it proposes an interdisciplinary curriculum that meets the current demands of generalization and quality. In this context, it is important to understand the educational process and also if this curriculum leads to distinct results from other traditional institutions. For this reason, a mixed research was proposed comparing the chemistry preservice teachers from the UFABC and from the University of São Paulo, regarding their Pedagogical Content Knowledge (PCK), since this is the knowledge that distinguishes the teacher from the specialist. The proposed research started by looking at the existing PCK models by a Hierarchical Decimal Classification (DHC) and a Factorial Analysis of Correspondence (FAC), to then, discuss two typologies of a knowledge that, since 2012 is not considered any more as a component of the PCK, the Orientations for Teaching, and its relation with the way the teacher interacts with his students (Discursive Interaction). Then, the way by which the development of the PCK takes place in the UFABC has been the focus of a case study that observes the modifications in PCK Maps. Finally a comparative study between the two institutions was developed by a Hierarchical Linear Model (HLM), using a test validated by the Rasch Model. The results showed that PCK discourses are becoming more similar since 2012 and that they share a common vocabulary, also that the PCK models consider this knowledge as being topic specific. The two typologies of the Orientations showed to be related providing empirical subsidies. It was also possible to demonstrate that they relate to the Discourse Interactions when the class was divided into segments (beginning, middle and end). The case study showed that the university teacher guided the development of his pupils¿s PCK through activities, but this impact was greater when individualized oral instructions were made. Finally, the proposed questionnaire proved to be reliable and capable of distinguishing the ability of respondents, however, it was not possible to distinguish two groups of respondents given the high variability of UFABC respondents. This result is compatible with the institution¿s curriculum and it makes difficult to compare with other respondents.

La prospection géoélectrique est largement utilisée au Maroc pour des reconnaissances hydrogéologique. Le but de ce travail et de proposer de nouvelles techniques d’interprétation des sondages électriques verticaux en un temps réduit, et aussi de bien exploiter une base de données de sondages électriques, par l’établissement entre autre des images 2D horizontales et verticales de l’estimation de la distribution des résistivités électriques apparentes (modélisation géostatistique, inversion, etc.). Dans le but de caractériser électriquement le secteur d’étude (nord-est du Bassin de Saïss), une analyse statistique des résistivités apparentes de sondages électriques verticaux a été réalisée. Cette simple analyse descriptive est suivie par une étude statistique multidirectionnelle : analyse en composantes principales (ACP) et par une classification hiérarchique ascendante (CHA). (...) Les résultats des analyses statistiques et géostatistiques complétés par les inversions des sondages moyens pas classe, ont mis en évidence la fiabilité de ces techniques pour l’interprétation d’un nombre important de sondages électriques au lieu de la méthode ordinaire qui se base sur l’inversion des sondages un par un et les corréler ultérieurement pour construire la structure globale du domaine étudié. Avec les techniques utilisées, dans le cadre de ce travail, des résultats très satisfaisants en un temps plus réduit sont obtenus. Les profils étudiés et inversés à l’aide du logiciel RES2Dinv montrent tous les trois grandes structures définies auparavant (Résistant-Conductrice-Résistant), par contre on note des variations intra formations. De plus, l’organisation spatiale des formations permet de confirmer l’existence de failles cohérentes avec la structure en horst et graben du bassin. / The Geoelectric prospection is usually used in Morocco for hydrogeological recognition. The purpose of this work is to propose new techniques for interpreting vertical electric soundings in a reduced time, and also to fully exploit a database of stored electrical soundings by the establishment, amongst other things, of the horizontal and vertical 2D images, estimating the distribution of apparent electrical resistivity (geostatistic modeling, inversion, etc.). In order to characterize electrically the study area (north-east of the Saïss Basin), a statistical analysis of apparent resistivity of vertical electric soundings was performed. This simple descriptive analysis is followed by a statistical analysis (principal component analysis PCA and ascending hierarchical classification HAC.) (...)The results of statistical analysis and geostatistical supplemented by inversion of the average electric sounding per class, highlighted the reliability of these techniques to the interpretation of a large number of electrical soundings instead of the usual method which is based on the inversion of the electrical sounding one by one and correlate them later, to build the global structure of the area studied. With the techniques used in this work, very satisfactory results in a more reduced time, for interpreting vertical electric soundings, are obtained. VIThe studied profiles and inverted using the software RES2Dinv show all three structures defined previously (Resistant – Conductive - resistant), on the other hand, there are variations within the same formation. In addition, the spatial organization of the formation makes it possible to confirm the existence of faults coherent with the structure in horst and graben basin.

We consider the broad framework of supervised learning, where one gets examples of objects together with some labels (such as tissue samples labeled as cancerous or non-cancerous, or images of handwritten digits labeled with the correct digit in 0-9), and the goal is to learn a prediction model which given a new object, makes an accurate prediction. The notion of accuracy depends on the learning problem under study and is measured by a performance measure of interest. A supervised learning algorithm is said to be 'statistically consistent' if it returns an `optimal' prediction model with respect to the desired performance measure in the limit of infinite data. Statistical consistency is a fundamental notion in supervised machine learning, and therefore the design of consistent algorithms for various learning problems is an important question. While this has been well studied for simple binary classification problems and some other specific learning problems, the question of consistent algorithms for general multiclass learning problems remains open. We investigate several aspects of this question as detailed below. First, we develop an understanding of consistency for multiclass performance measures defined by a general loss matrix, for which convex surrogate risk minimization algorithms are widely used. Consistency of such algorithms hinges on the notion of 'calibration' of the surrogate loss with respect to target loss matrix; we start by developing a general understanding of this notion, and give both necessary conditions and sufficient conditions for a surrogate loss to be calibrated with respect to a target loss matrix. We then define a fundamental quantity associated with any loss matrix, which we term the `convex calibration dimension' of the loss matrix; this gives one measure of the intrinsic difficulty of designing convex calibrated surrogates for a given loss matrix. We derive lower bounds on the convex calibration dimension which leads to several new results on non-existence of convex calibrated surrogates for various losses. For example, our results improve on recent results on the non-existence of low dimensional convex calibrated surrogates for various subset ranking losses like the pairwise disagreement (PD) and mean average precision (MAP) losses. We also upper bound the convex calibration dimension of a loss matrix by its rank, by constructing an explicit, generic, least squares type convex calibrated surrogate, such that the dimension of the surrogate is at most the (linear algebraic) rank of the loss matrix. This yields low-dimensional convex calibrated surrogates - and therefore consistent learning algorithms - for a variety of structured prediction problems for which the associated loss is of low rank, including for example the precision @ k and expected rank utility (ERU) losses used in subset ranking problems. For settings where achieving exact consistency is computationally difficult, as is the case with the PD and MAP losses in subset ranking, we also show how to extend these surrogates to give algorithms satisfying weaker notions of consistency, including both consistency over restricted sets of probability distributions, and an approximate form of consistency over the full probability space. Second, we consider the practically important problem of hierarchical classification, where the labels to be predicted are organized in a tree hierarchy. We design a new family of convex calibrated surrogate losses for the associated tree-distance loss; these surrogates are better than the generic least squares surrogate in terms of easier optimization and representation of the solution, and some surrogates in the family also operate on a significantly lower dimensional space than the rank of the tree-distance loss matrix. These surrogates, which we term the `cascade' family of surrogates, rely crucially on a new understanding we develop for the problem of multiclass classification with an abstain option, for which we construct new convex calibrated surrogates that are of independent interest by themselves. The resulting hierarchical classification algorithms outperform the current state-of-the-art in terms of both accuracy and running time. Finally, we go beyond loss-based multiclass performance measures, and consider multiclass learning problems with more complex performance measures that are nonlinear functions of the confusion matrix and that cannot be expressed using loss matrices; these include for example the multiclass G-mean measure used in class imbalance settings and the micro F1 measure used often in information retrieval applications. We take an optimization viewpoint for such settings, and give a Frank-Wolfe type algorithm that is provably consistent for any complex performance measure that is a convex function of the entries of the confusion matrix (this includes the G-mean, but not the micro F1). The resulting algorithms outperform the state-of-the-art SVMPerf algorithm in terms of both accuracy and running time. In conclusion, in this thesis, we have developed a deep understanding and fundamental results in the theory of supervised multiclass learning. These insights have allowed us to develop computationally efficient and statistically consistent algorithms for a variety of multiclass learning problems of practical interest, in many cases significantly outperforming the state-of-the-art algorithms for these problems.

Dans cet article, nous présentons une nouvelle approche de détection multi-classes basée sur un parcours hiérarchique de classifieurs appris simultanément. Pour plus de robustesse et de rapidité, nous proposons d’utiliser un arbre de classes d’objets. Notre modèle de détection est appris en combinant les contraintes de tri et de classification dans un seul problème d’optimisation. Notre formulation convexe permet d’utiliser un algorithme de recherche pour accélérer le temps d’exécution. Nous avons mené des évaluations de notre algorithme sur les benchmarks PASCAL VOC (2007 et 2010). Comparé à l’approche un-contre-tous, notre méthode améliore les performances pour 20 classes et gagne 10x en vitesse. / Recent years have witnessed a competition in autonomous navigation for vehicles boosted by the advances in computer vision. The on-board cameras are capable of understanding the semantic content of the environment. A core component of this system is to localize and classify objects in urban scenes. There is a need to have multi-class object detection systems. Designing such an efficient system is a challenging and active research area. The algorithms can be found for applications in autonomous driving, object searches in images or video surveillance. The scale of object classes varies depending on the tasks. The datasets for object detection started with containing one class only e.g. the popular INRIA Person dataset. Nowadays, we witness an expansion of the datasets consisting of more training data or number of object classes. This thesis proposes a solution to efficiently learn a multi-class object detector. The task of such a system is to localize all instances of target object classes in an input image. We distinguish between three major efficiency criteria. First, the detection performance measures the accuracy of detection. Second, we strive low execution times during run-time. Third, we address the scalability of our novel detection framework. The two previous criteria should scale suitably with the number of input classes and the training algorithm has to take a reasonable amount of time when learning with these larger datasets. Although single-class object detection has seen a considerable improvement over the years, it still remains a challenge to create algorithms that work well with any number of classes. Most works on this subject extent these single-class detectors to work accordingly with multiple classes but remain hardly flexible to new object descriptors. Moreover, they do not consider all these three criteria at the same time. Others use a more traditional approach by iteratively executing a single-class detector for each target class which scales linearly in training time and run-time. To tackle the challenges, we present a novel framework where for an input patch during detection the closest class is ranked highest. Background labels are rejected as negative samples. The detection goal is to find the highest scoring class. To this end, we derive a convex problem formulation that combines ranking and classification constraints. The accuracy of the system is improved by hierarchically arranging the classes into a tree of classifiers. The leaf nodes represent the individual classes and the intermediate nodes called super-classes group recursively these classes together. The super-classes benefit from the shared knowledge of their descending classes. All these classifiers are learned in a joint optimization problem along with the previouslymentioned constraints. The increased number of classifiers are prohibitive to rapid execution times. The formulation of the detection goal naturally allows to use an adapted tree traversal algorithm to progressively search for the best class but reject early in the detection process the background samples and consequently reduce the system’s run-time. Our system balances between detection performance and speed-up. We further experimented with feature reduction to decrease the overhead of applying the high-level classifiers in the tree. The framework is transparent to the used object descriptor where we implemented the histogram of orientated gradients and deformable part model both introduced in [Felzenszwalb et al., 2010a]. The capabilities of our system are demonstrated on two challenging datasets containing different object categories not necessarily semantically related. We evaluate both the detection performance with different number of classes and the scalability with respect to run-time. Our experiments show that this framework fulfills the requirements of a multi-class object detector and highlights the advantages of structuring class-level knowledge.

This thesis investigates how pre-trained contextualized language models can be adapted for multi-label text classification of Swedish news articles. Various classifiers are built on pre-trained BERT and ELECTRA models, exploring global and local classifier approaches. Furthermore, the effects of domain specialization, using additional metadata features and model compression are investigated. Several hundred thousand news articles are gathered to create unlabeled and labeled datasets for pre-training and fine-tuning, respectively. The findings show that a local classifier approach is superior to a global classifier approach and that BERT outperforms ELECTRA significantly. Notably, a baseline classifier built on SVMs yields competitive performance. The effect of further in-domain pre-training varies; ELECTRA’s performance improves while BERT’s is largely unaffected. It is found that utilizing metadata features in combination with text representations improves performance. Both BERT and ELECTRA exhibit robustness to quantization and pruning, allowing model sizes to be cut in half without any performance loss.

La poursuite du rythme d’augmentation de l’espérance de vie des générations issue du baby-boom dans les pays développés serait souvent accompagnée de limitations fonctionnelles, d’incapacité, de plus en plus observées dans la population gériatrique. L'objectif général de cette thèse était de contribuer à la connaissance de l’évolution de l’autonomie fonctionnelle des personnes âgées dans une population hétérogène. Il s’agissait dans un premier temps d'identifier des groupes homogènes dans une population hétérogène de personnes âgées suivant la même trajectoire d'autonomie fonctionnelle sur une période de deux ans, ainsi que des facteurs prédictifs potentiels. Dans un second temps, d’analyser les conséquences cliniques des trajectoires et la survie des patients sur la même période d’observation. Le SMAF (Système de Mesure de l’Autonomie Fonctionnelle) et les échelles ADL (Activities of Daily Living) ont été employés comme indicateurs d’évaluation de l’autonomie. Dans ce contexte, des données de 221 patients issues de la cohorte UPSAV (Unité de Prévention, de Suivi et d’Analyse du Vieillissement) ont été exploitées. Nous avons employé trois méthodes d’analyse de trajectoires dont le GBTM (Group-Based Trajectory Modeling), k-means et classification ascendante hiérarchique. Les résultats ont révélé trois trajectoires distinctes d’autonomie fonctionnelle : stable, stable pendant un temps puis détériorée, continuellement altérée. Les facteurs prédictifs des trajectoires obtenus à l’aide de la régression logistique sont des critères socio-démographiques, médicaux et biologiques. Les personnes âgées affectées à la trajectoire de perte d’autonomie (trajectoire continuellement altérée) ont montré de fortes proportions de chutes dommageables. A partir d’un modèle de Cox, les troubles neurocognitifs, l’insuffisance cardiaque, la perte de poids involontaire et l’alcool ont été révélés comme facteurs prédictifs de la survenue du décès. On conclut de ces travaux que l’analyse longitudinale sur deux ans de suivi a permis de trouver des sous-groupes homogènes de personnes âgées en termes d’évolution de l’indépendance fonctionnelle. Quel que soit le niveau d’autonomie, la prévention de l’UPSAV devient utile même si le niveau d’utilité n’est pas le même. La prévention et le dépistage de la perte d’autonomie de la personne âgée suivie sur son lieu de vie doivent être anticipés afin de retarder la dégradation et maintenir l’autonomie à domicile. Des analyses ultérieures devraient s’intéresser à l’exploration de plus larges cohortes de personnes âgées pour confirmer et généraliser notre travail. / The increase in life expectancy of baby boom generations in developed countries would often be accompanied by functional limitations, disability, increasingly observed in the geriatric population. The general objective of this thesis was to contribute to the knowledge of the evolution of the functional independence of older people in a heterogeneous population. First, it was to identify homogeneous groups in a heterogeneous population of elderly people following the same functional independence trajectory over a two-year period, and potential predictive factors. Second, it was to analyze the clinical consequences of trajectories and patient survival over the same observation period. The SMAF (Système de Mesure de l'Autonomie Fonctionnelle) and ADL (Activities of Daily Living) scales were used as indicators for measuring independence. Analysis were performed from a sample of 221 patients of UPSAV (Unit for Prevention, Monitoring and Analysis of Aging) cohort. We used three methods including trajectory analysis including GBTM (Group-Based Trajectory Modeling), k-means and ascending hierarchical classification. The results suggest three distinct trajectories of functional independence: stable, stable then decline, continuously decline. The predictors of trajectories obtained using logistic regression are socio-demographic, medical and biological criteria. Patients assigned to the loss of independence trajectory (continuously altered trajectory) reported high proportions of injurious falls. Based on a Cox model, neurocognitive disorders, heart failure, involuntary weight loss and alcohol were revealed as predictors of death. We conclude from this work that the two-year longitudinal analysis identified homogeneous subgroups of elderly people in terms of changes in functional independence. The prevention of UPSAV becomes a useful even if the utility level is not the same. Prevention and screening of the loss of independence of the elderly person followed at home must be anticipated in order to delay the deterioration and to maintaining the autonomy. Future analyses should focus on exploring large cohorts of older people to confirm and generalize our research.

Cette thèse porte dans une perspective sociohistorique et ethnographique sur le mouvement piquetero en Argentine et ses transformations successives durant les années 2000 en un vaste réseau d’organisation prestataires de services. Ce mouvement est né des actions collectives des chômeurs et travailleurs précaires à la fin des années 1990 contre les effets des réformes « néolibérales » et tient son nom de l’un de ses modes de protestation privilégié, le barrage de route ou piquete. Dès ses débuts, les piqueteros ont fait l’objet d’un double traitement de la part de l’État, entre répression et récupération dans le cadre de la mise en place de programmes de transferts conditionnés des ressources (Conditional Cash Transfer Programs). Dans cette nouvelle configuration de l’action publique ciblée, il s’agit désormais pour l’État de déléguer la distribution des aides sociales aux organisations, au regard de leur proximité territoriale avec les populations précarisées. Or, cette thèse montre que ce rôle flou de guichet, qui tend à introduire d’une manière ou d’une autre de la concurrence entre les organisations, a ainsi rapidement contribué à fragmenter l’espace piquetero, et produit des effets ambivalents sur les pratiques et les trajectoires des participants. La thèse s’appuie sur des méthodes mixtes, qualitatives et quantitatives, à partir d’une enquête de terrain menée pendant 40 mois, entre 2000 et 2015, dans deux provinces argentines. D’une part, à travers une ethnographie et des entretiens biographiques approfondis auprès des leaders, des délégués et des militants de la base (N=104), nous avons observé les interactions entre ces différentes catégories. Une prosopographie des leaders (N=76) nous a, d’autre part, permis à partir des méthodes statistiques de l’analyse factorielle (ACM) et de la classification (CAH) de rendre compte de la structuration de cet espace des organisations. Dans une première partie, la thèse s’attache – à l’appui d’archives et d’entretiens – à mettre en lumière les conditions de possibilité de la cristallisation progressive d’un mouvement social en un espace d’organisations. Nous avons cherché ici à appréhender le contexte, les enjeux et les moyens d’action de ce mouvement contestataire, en rapportant son inscription à l’évolution depuis le début du XXè siècle des rapports entre État, partis politiques et syndicats. La deuxième partie de la thèse est, elle, consacrée à l’analyse des pratiques militantes et des formes d’encadrement au sein des organisations. L’ouverture de la boîte noire des organisations révèle ainsi à quel point leur fonctionnement interne résulte de la capacité d’un ensemble d’intermédiaires à mener un travail de représentation, de mobilisation et de gestion des ressources vis-à-vis de certaines fractions des classes populaires particulièrement disposées à s’engager dans la durée. L’examen statistique des trajectoires de leaders nous renseigne par ailleurs sur les ressources nécessaires à l’occupation d’un tel poste et aussi sur ce que l’engagement fait aux parcours individuels. Enfin, une troisième partie a servi à appréhender les pratiques associatives au sein des organisations. Restituer les logiques de recrutement et les profils des recrutés a donné à voir dans la durée aussi bien les conditions de l’engagement de ces chômeurs et travailleurs précaires que les effets sur leurs trajectoires. L’observation des pratiques notamment lors des assemblées permet de montrer les principes d’encadrement tendus entre militantisme et entreprenariat qui pèsent sur les participants. Si cette fraction de précaires témoigne au sein des classes populaires de formes de mobilisation et de résistance particulièrement exemplaires, ils tendent également à déployer des modalités d’accommodement aux organisations, différenciées suivant leur socialisation et le volume et la nature de leurs ressources. / This thesis discusses the Piquetero movement in Argentina and its successive transformations during the 2000s into an extensive network of service provider organizations throughout the territory from a sociohistorical and ethnographic perspective. This movement was born out of the collective actions of the unemployed and precarious workers in the late 1990s against the effects of "neoliberal" reforms, and takes its name from one of their preferred modes of protest, the roadblock or picket. Since its beginnings, the Piquetero movement has been the subject of a double treatment by the State, between repression and recovery in the context of the establishment of Conditional Cash Transfer Programs. In this new configuration of targeted public action, it is now up to the State to delegate the distribution of social assistance to a network of organizations, given their territorial proximity to the underprivileged populations. However, this thesis shows that this fuzzy wicket role, which tends to introduce in one way or another the competition amid the organizations, has thus quickly contributed to fragment the piquetero space, and produces ambivalent effects on the practices and the trajectories of the participants. The thesis is based on mixed methods, qualitative and quantitative, from a large 40-month field survey conducted between 2000 and 2015 in two Argentinian provinces. On the one hand, through an ethnography and in-depth biographical interviews with leaders, delegates and grassroots activists (N=104), we observed the interactions between these different categories. A prosopography of the leaders (N=76) allowed us, on the other hand, from the statistical methods of factor analysis (ACM) and hierarchical classification (CAH), to report on the structuring of this space of organizations. In the first part, the thesis focuses – with the support of archives and interviews – on the conditions of the gradual crystallization of a social movement into a space of organizations. We sought here to understand the context, the stakes and the means of action of this protest movement, relating its inscription to the evolution since the beginning of the XXth century of the relations between State, political parties and unions. The second part of our thesis is devoted to the analysis of activist practices and forms of supervision within organizations. The opening of the black box of the organizations thus reveals to what extent their internal functioning results from the capacity of a set of intermediaries to carry out a work of representation, mobilization and management of resources among working classes particularly willing to engage in the long term. The statistical examination of the trajectories of leaders also informed us about the resources that were necessary to occupy such a position and also about the effects of their engagement to their individual trajectories. Finally, a third part serves to apprehend associative practices within organizations. Restoring the recruiting logics and the profiles of the recruits has shown in the long term both the conditions of the commitment of these unemployed and precarious workers and the effects on their trajectories. The observation of practices, especially during assemblies, shows the principles of supervision stretched between activism and entrepreneurship which weighed on the participants. If this fraction of precarious people testifies within the working classes of forms of mobilization and resistance particularly exemplary, they also tend to deploy modes of accommodation to organizations, differentiated according to their socialization, and the volume and nature of their resources.