Cette thèse s'intègre dans le cadre du traitement automatique du langage naturel. La problématique du résumé automatique de documents arabes qui a été abordée, dans cette thèse, s'est cristallisée autour de deux points. Le premier point concerne les critères utilisés pour décider du contenu essentiel à extraire. Le deuxième point se focalise sur les moyens qui permettent d'exprimer le contenu essentiel extrait sous la forme d'un texte ciblant les besoins potentiels d'un utilisateur. Afin de montrer la faisabilité de notre approche, nous avons développé le système "L.A.E", basé sur une approche hybride qui combine une analyse symbolique avec un traitement numérique. Les résultats d'évaluation de ce système sont encourageants et prouvent la performance de l'approche hybride proposée. Ces résultats, ont montré, en premier lieu, l'applicabilité de l'approche dans le contexte de documents sans restriction quant à leur thème (Éducation, Sport, Science, Politique, Reportage, etc.), leur contenu et leur volume. Ils ont aussi montré l'importance de l'apprentissage dans la phase de classement et sélection des phrases forment l'extrait final. / This thesis falls within the framework of Natural Language Processing. The problems of automatic summarization of Arabic documents which was approached, in this thesis, are based on two points. The first point relates to the criteria used to determine the essential content to extract. The second point focuses on the means to express the essential content extracted in the form of a text targeting the user potential needs.In order to show the feasibility of our approach, we developed the "L.A.E" system, based on a hybrid approach which combines a symbolic analysis with a numerical processing.The evaluation results are encouraging and prove the performance of the proposed hybrid approach.These results showed, initially, the applicability of the approach in the context of mono documents without restriction as for their topics (Education, Sport, Science, Politics, Interaction, etc), their content and their volume. They also showed the importance of the machine learning in the phase of classification and selection of the sentences forming the final extract.

Submitted by William Justo Figueiro (williamjf) on 2015-07-09T22:47:34Z No. of bitstreams: 1 28b.pdf: 928828 bytes, checksum: 03bf0090fd6d0473043b76c6366d0c61 (MD5) / Made available in DSpace on 2015-07-09T22:47:34Z (GMT). No. of bitstreams: 1 28b.pdf: 928828 bytes, checksum: 03bf0090fd6d0473043b76c6366d0c61 (MD5) Previous issue date: 2014-12-22 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / O presente estudo busca investigar as relações retóricas que emergem do encaixe de segmentos narrativos na organização macroestrutural de notícias de divulgação científica (DC) dirigidas aos públicos infantil e adulto. A organização narrativa é analisada segundo os postulados de Adam (2011) e Charaudeau (2008b); a organização retórica macroestrutural dos textos, por sua vez, é estudada de acordo com a Rhetorical Structure Theory – RST. (MANN; THOMPSON, 1988). O corpus do estudo é composto de 15 notícias de DC voltadas ao público infantil, publicadas na revista Ciência Hoje das Crianças entre dezembro de 2004 e setembro de 2010, e de 15 notícias de DC direcionadas ao público adulto, veiculadas na revista Ciência Hoje, entre agosto de 2005 e julho de 2012. A metodologia empregada consiste na análise quantitativa da emergência de relações retóricas entre os segmentos narrativos encontrados nos corpora (infantil e adulto) e as demais porções textuais, bem como na análise qualitativa de alguns textos selecionados para exemplificar cada uma das relações retóricas encontradas. Em seguida, é realizada uma análise comparativa entre os dados quantitativos provenientes das análises dos textos para crianças e dos textos para adultos. Assume-se que a esquematização de um texto é um processo de coconstrução, no qual o produtor, ao organizar seu plano textual, leva em consideração as características e conhecimentos de seu possível leitor e, a partir disso, lança mão de estratégias variadas para alcançar o fim discursivo pretendido e causar os efeitos desejados sobre o leitor. Dessa forma, conclui-se que a emergência recorrente de determinadas relações retóricas entre as narrativas encaixadas e as outras partes do texto revela estratégias do produtor textual para orientar a leitura e compreensão da notícia, tanto nos textos escritos para crianças quanto nos escritos para adultos. / This research aims to investigate the rhetorical relations which emerge from the insertion of narrative segments in the macrostructural organization of scientific popularization news directed to children and adults. The narrative organization is analyzed according to Adam’s (2011) and Charaudeau’s (2008b) postulates; the rhetorical organization of the texts, in turn, is studied according to Rhetorical Structure Theory – RST. (MANN; THOMPSON, 1988). The research corpus is composed of 15 scientific popularization news intended for children, published in the magazine Ciência Hoje das Crianças, and of 15 scientific popularization news directed to adults, issued in the magazine Ciência Hoje. The methodology consists of a quantitative analysis of the emergency of rhetorical relations between the narrative segments found in the corpora (texts for children and texts for adults) and other textual portions, as well as of a qualitative analysis of some selected texts in order to exemplify each one of the rhetorical relations found. Then, a comparative analysis is done between the quantitative data resultant from the analysis of the texts for children and of the texts for adults. We assume that the schematization of a text is a process of co-construction, in which the producer, when organizes the text plan, takes into consideration the characteristics and knowledge of his/her possible reader and, from this, resorts to different strategies in order to achieve the discursive aim intended and cause the desired effect over the reader. Thus, we conclude that the recurrent emergency of certain rhetorical relations between the inserted narratives and the other portions of the text reveals strategies of the textual producer in order to guide the reading and comprehension of the new, both in texts written forchildren and in texts written for adults.

<p>Aquaplaning accidents are relatively rare, but could have fatal effects. The task of this master’s thesis is to use data from the Laser Road Surface Tester to detect road sections with risk of aquaplaning. </p><p>A three-dimensional model based on data from road surface measurements is created using MATLAB (version 6.1). From this general geometrical model of the road, a pond model is produced from which the theoretical risk ponds are detected. A risk pond indication table is fur-ther created. </p><p>The pond model seems to work well assuming that the data from the road model is correct. Determining limits for depth and length of risk ponds can be made directly by the user. MATLAB code is reasonably easy to understand and this leaves great opportunities for changing different parameters in a simple way. </p><p>Supplementary research is needed to further improve the risk pond detection model. Collecting data at smaller intervals and with more measurement points would be desirable for achieving better correlation with reality. In a future perspective, it would be wise to port the code to another programming language and this could make the computations faster.</p>

Aquaplaning accidents are relatively rare, but could have fatal effects. The task of this master’s thesis is to use data from the Laser Road Surface Tester to detect road sections with risk of aquaplaning. A three-dimensional model based on data from road surface measurements is created using MATLAB (version 6.1). From this general geometrical model of the road, a pond model is produced from which the theoretical risk ponds are detected. A risk pond indication table is fur-ther created. The pond model seems to work well assuming that the data from the road model is correct. Determining limits for depth and length of risk ponds can be made directly by the user. MATLAB code is reasonably easy to understand and this leaves great opportunities for changing different parameters in a simple way. Supplementary research is needed to further improve the risk pond detection model. Collecting data at smaller intervals and with more measurement points would be desirable for achieving better correlation with reality. In a future perspective, it would be wise to port the code to another programming language and this could make the computations faster.

Purpose: This study assessed effectiveness of five tools recommended by the US Preventive Services Task Force (USPSTF), designed to help primary care clinicians determine which unaffected patients to refer to genetics specialists for breast cancer risk assessment based on concerning family history. Design: This descriptive secondary analysis included 85 women aged 40-74. All participants had a first-degree female relative previously diagnosed with breast cancer who also had uninformative negative BRCA1/2 tests. Methods: Each pedigree was evaluated using the five tools including the Family History Screen-7 (FHS-7), Pedigree Assessment Tool (PAT), Manchester Scoring System, Referral Screening Tool (RST), and Ontario-Family History Assessment Tool (Ontario-FHAT). All five tools were applied to each study participant. Sensitivity, specificity, positive predictive value, and negative predictive value were calculated to describe each tool’s ability to identify women with elevated risk as calculated by the Claus model. Receiver operating curves (ROC) were also plotted. Differences between areas under the curve (AUCs) for all possible pairs of tools were estimated through logistic regression to assess for differences in tool performance. Results: Claus calculations identified 14 women out of 85 whose lifetime risk of breast cancer was elevated at > 15%. Only two tools, the Ontario-FHAT and FHS-7, identified all 14 women with elevated risk, a sensitivity of 100%. The FHS-7 tool flagged all 85 participants, meaning its specificity was zero. The Ontario-FHAT flagged 59 participants as needing referral (specificity 36.2%) and had a negative predictive value (NPV) of 100%, indicating that if a woman was not found to need a referral to a genetics professional, it is likely she did not have an elevated lifetime risk of developing breast cancer. AUC values were not significantly different between tools (all p values > .05), and thus were not helpful in discriminating between the tools. Conclusion: In this population, the Ontario-FHAT out-performed other tools in terms of sensitivity and negative predictive value; however, low specificity and positive predictive value must be balanced against these findings. Thus, the Ontario-FHAT can help determine which women would benefit from referral to a genetics specialist.

Selon Asher (2013), la cohérence discursive force l’inférence de (1c) dans les exemples (1a)-(1b), expliquant ainsi l’(in)acceptabilité des exemples. (1) a. ‘John a un nombre pair d’enfants. Il en a 4.’ b. ‘ ? ?John a un nombre pair d’enfants. Il en a 3.’ c. +> John a n enfants et pas plus Nous avons tenté de déterminer si les implicatures nécessaires au maintien de la cohérence discursive sont systématiquement inférées en nous appuyant sur les théories de la RST et de la SDRT. Des tests linguistiques et la vérification du respect des contraintes sémantiques associées aux relations de discours ont mis en évidence deux catégories d’exemples contenant le quantificateur certains : ceux de type- RenfNA, dont les implicatures ne sont pas nécessaires à la cohérence et ceux de typeRenfA dans lesquels elles le sont. Nos tests ayant révélé que le renforcement est nécessaire dans les exemples de typeRenfA, nous avons conclu que les implicatures ne sont pas systématiquement inférées. Nous avons tenté d’apporter une explication à ce phénomène en effectuant des analyses de la structure discursive de nos exemples et avons démontré que dans les exemples de typeRenfNA, les relations de discours visent le constituant π∃ (certains), tandis que dans ceux de typeRenfA, le constituant π¬∀ (mais pas tous) est visé. Nos travaux ont démontré que les implicatures scalaires ne sont pas systématiquement inférées rendant parfois leur renforcement obligatoire. Nous avons également proposé un modèle à granularité fine prenant en compte la structure discursive et la pragmatique afin d’expliquer ce phénomène. / According to Asher (2013), discourse coherence forces the inference of (2c) in examples (2a)-(2b), thus explaining the (in)acceptability of these examples. (2) a. ‘John has an even number of children. He has 4.’ b. ‘??John has an even number of children. He has 3.’ c. +> John has n children and not more We attempted to determine whether the implicatures that are necessary to maintain discourse coherence are systematically inferred by drawing on the theories of RST and SDRT. Through linguistic tests and checking the respect of semantic constraints associated with discourse relations, we identified two categories of examples containing the quantifier some: typeRenfNA examples, in which implicatures are not necessary for discourse coherence, and typeRenfA examples in which they are. As our tests revealed that reinforcement is necessary in typeRenfA examples, we concluded that implicatures are not systematically inferred. We then attempted to explain this phenomenon. We performed analyses of the discourse structure of our examples and showed that in typeRenfNA examples, the discourse relations target the π∃ (some) constituent, while in typeRenfA examples, the π¬∀ (but not all) constituent is targeted. Thus, our work has shown that scalar implicatures are not systematically inferred, making implicature reinforcement sometimes mandatory. We also proposed a fine-grained model taking discourse structure and pragmatics into account to explain this phenomenon.

In the scope of our BACE1 inhibitor project we used an originally designed microtiter plate-based screening to discover 4 triazole-linked reduced amide isosteres that showed modest (single digit micromolar) BACE1 inhibition. Our ligands were designed based on a very potent (single digit nanomolar) isopththalamide ligand from Merck. We supplanted one of the amide linkages in order to incorporate our triazole and saw a 1000-fold decrease in potency. We then enlisted Molsoft, L.L.C. to compare our ligand to Merck's in silico to account for this discrepancy. They found that the triazole linkage gives rise to a significantly different docking pose in the active site of the BACE1 enzyme, therefore diminishing its potency relative to the Merck ligand. The ability to control the regio- and stereochemical outcome of organic reactions is an ongoing interest and challenge to synthetic chemists. The pre-association of reacting partners through hydrogen bonding (H-bonding) can often to yield products with extremely high stereoselectivity. We were able to show that anilines, due to their enhanced acidity relative to amines, can serve as substrate directing moieties in the opening of cyclohexene oxides. We observed that by judicious choice of conditions we could control the regiochemical outcome of the reaction. These studies demonstrate that an intramolecular anilino-NH hydrogen bond donor can direct Fürst-Plattner epoxide opening. A unified mechanism for this phenomenon has been proposed in this work which consists of a novel mechanistic route we call "NH-directed Fürst-Plattner." We further studied the opening of cyclohexene oxides by incorporating amide and amide derivative substituents in both the allylic and homoallylic position relative to the epoxide moiety. Our attempts to control regioselectivity in the allylic systems were unsuccessful; however when the directing substituent was in the homoallylic position, we could demonstrate some degree of regioselectivity. An additional project that the author worked on for approximately one year during his graduate student tenure is not described within this work. In February of 2009 AstraZeneca, Mayo Clinic, and Virginia Tech Intellectual Properties Inc. concomitantly announced that AstraZeneca licensed a portfolio of preclinical Triple Reuptake Inhibitor (TRI) compounds for depression. The lead compound, PRC200, was discovered by a collaborative effort between the Carlier and Richelson (Mayo Clinic Jacksonville) research groups in 1998. The author was tasked to develop backup candidates of PRC200 in order to improve the pharmacokinetics of the lead compound. Due to confidentiality agreements, this work is not reported herein. / Ph. D.

Ce sujet s'inscrit dans la continuité des travaux commencés au sein du L2ES dans le cadre du programme ECCE en partenariat avec CREEBEL qui assure le nancement. La plateforme ECCE est un véhicule hybride électrique série à 4 roues indépendantes d'une puissance nominale en propulsion de 120kW électrique. La source d'énergie principale est constituée de deux moteurs diesels entraînant deux alternateurs. L'énergie électrique produite alimente les 4 moteurs électriques de traction et le reste est stocké dans le pack des batteries. Cette thèse développe les stratégies de couplage énergétique entre ce pack des batteries et les supercondensateurs an d'assurer au véhicule une dynamique de fourniture et du stockage de l'énergie électrique. Une étude bibliographique a permis de passer en revue le bilan technologique et les applications potentielles des supercondensateurs, puis d'élaborer le modèle simplié des supercondensateurs qui traduit dèlement le comportement des cellules durant les phases de charge et de décharge. Diérentes topologies des convertisseurs DC/DC avec des stratégies originales de gestion d'énergie électrique embarquée sont traitées. Les topologies proposées sont basées sur les convertisseurs Buck-Boost et les convertisseurs DC/AC-AC/DC à étage intermédiaire haute fréquence. Pour une raison de coût, les maquettes expérimentales des topologies ont été réalisées à l'échelle réduite ( 1/10 ). Les résultats expérimentaux obtenus ont permis de comparer les performances des topologies pour deux types de commande. La stratégie de gestion d'énergie à base des correcteurs polynomiaux (RST) est comparée à celle utilisant des correcteurs PI classiques. Ces études comparatives ont permis de choisir la meilleure topologie destinée au couplage des supercondensateurs sur le bus continu du banc ECCE.

Le développement des véhicules hybrides et électriques s'est intensifié ces dernières années, face aux problématiques environnementales et économiques. Afin que les performances de ces derniers soient comparables à celle d'un véhicule à moteur thermique, de nombreuses avancées technologiques sont indispensables. Le déploiement de véhicule entièrement à traction électrique, ne serait être réalisable, sans des infrastructures de recharge adaptées. Cette thèse constitue une contribution à l'étude d'un chargeur de batteries intégré au véhicule électrique, dans le cadre du projet SOFRACI.L'architecture de ce chargeur entièrement réversible sert également pour la traction du véhicule. L'utilisation de tous les éléments y compris le moteur de traction pour les deux fonctions, réduit l'encombrement et le coût de la structure. L'objectif principal consiste à modéliser le système de conversion alternatif-continu du chargeur et à établir les lois de commandes.Lors de la première phase de l'étude, une attention particulière a été portée à l'utilisation des enroulements de la machine en tant qu'inductance de filtrage du convertisseur. Un modèle correspondant à ce fonctionnement a été obtenu et a permis de mettre en évidence un couplage magnétique existant entre les diverses phases.En s'appuyant sur les modèles obtenues, la seconde étape a consisté à définir les stratégies de commande. Deux méthodes ont été employées. La première, par le biais d'une transformation a permis de se ramener dans un repère, où la majorité des termes de couplage ont été éliminés. Pour l'autre méthode, tous les éléments du couplage ont été considérés. Pour chacune de ces stratégies, des correcteurs PI et RST fondés sur le placement de pôles robustes ont été dimensionnés pour garantir la stabilité du système.Une part importante du travail a été consacrée à la réalisation d'un banc d'essai expérimental reconstituant la conversion alternative-continue. Compte tenu des différences de dynamiques des grandeurs asservies, une carte FPGA et un processeur ont été utilisés. L'exploitation de ce moyen d'essai a permis de comparer et de valider les lois de commande développées.

Cette thèse s'intègre dans le cadre du traitement automatique du langage naturel. La problématique du résumé automatique de documents arabes qui a été abordée, dans cette thèse, s'est cristallisée autour de deux points. Le premier point concerne les critères utilisés pour décider du contenu essentiel à extraire. Le deuxième point se focalise sur les moyens qui permettent d'exprimer le contenu essentiel extrait sous la forme d'un texte ciblant les besoins potentiels d'un utilisateur. Afin de montrer la faisabilité de notre approche, nous avons développé le système "L.A.E", basé sur une approche hybride qui combine une analyse symbolique avec un traitement numérique. Les résultats d'évaluation de ce système sont encourageants et prouvent la performance de l'approche hybride proposée. Ces résultats, ont montré, en premier lieu, l'applicabilité de l'approche dans le contexte de documents sans restriction quant à leur thème (Éducation, Sport, Science, Politique, Reportage, etc.), leur contenu et leur volume. Ils ont aussi montré l'importance de l'apprentissage dans la phase de classement et sélection des phrases forment l'extrait final.