The proliferation of modern wireless networks increases demand for high capacity and throughput in order to provide faster, more robust, efficient and broadband services to end users. Mobile WiMAX and LTE are examples of such networks in which for some cases they have exposed limited connectivity due to harsh environment. Relay stations are preferred to overcome problems of weak or no access for such network devices, that are placed in specific positions to maintain high quality of data transfer at low cost and provide the required connectivity anywhere anytime. These stations should be equipped with an antenna system capable of establishing communication between base station (backhaul link) and end users (access link). This thesis focuses on the design and development of a new antenna system that is suitable for a relay-based wireless network. Planar geometries of microstrip patch antennas are utilized. The antenna system comprises two antenna modules: a new design of a single antenna for access link and a new design of an antenna array for backhaul link realization. Both antenna specifications are compatible with the IEEE802.16j protocol standard. Hence, relay station should be capable of pointing its radiation pattern to the base station antenna, thus to achieve the desired radiation pattern of the relay station, a new beam-forming module is proposed, designed and developed to generate the proper radiation pattern. The beam-forming module incorporating digital phase shifters and attenuator chips is fabricated and tested. The optimization process using the Least Mean Square (LMS) algorithm is considered in this study to assign the proper phase and amplitude that is necessary to each radiation element excitation current, to produce the desired steered radiation pattern. A comprehensive study on the coupling effects for several relative positions between two new backhaul and access link antenna elements is performed. Two new antenna configurations for coupling reduction are tested and the simulated and measured results in terms of antenna radiation performances were compared and commented.

Cette thèse présente trois chapitres qui mobilisent un cadre d’analyse évolutionniste et qui étudient empiriquement la relation « innovation-performance » à partir de données CIS. Nous souhaitons montrer que les sources de l’innovation et les méthodes d’appropriation varient en fonction des secteurs d’activité et des stratégies d’innovation des firmes. Dans un premier temps, nous décrivons les tendances de l’innovation à partir de quatre vagues d’enquêtes CIS (1994-2006) et nous analysons la persistance de l’innovation sur un échantillon de 431 firmes avec une régression logistique binaire. Nous montrons que la persistance de l’innovation est plus élevée pour les firmes qui innovent en produits car ces firmes sont contraintes d’investir de manière continue dans des projets d’innovation pour rester compétitives. Les firmes qui innovent en procédés sont moins persistantes car leur stratégie est plus orientée vers des ajustements de la qualité des produits ainsi que vers l’amélioration des processus de production. Les deux derniers essais explorent avec la méthode des doubles moindres carrés le lien innovation-performance économique sur un échantillon de 7 742 firmes portant sur la période 2002-2005. Nous expliquons que la source principale de l’innovation des firmes à « forte intensité scientifique » est la R&D, d’une part, et que les méthodes d’appropriation des rentes de l’innovation passent par l’acquisition d’actifs complémentaires (tels que l’utilisation combinée de titres de propriété intellectuelle et de secrets de fabrication), d’autre part. En revanche, les firmes dans les autres catégories (notamment celles à fortes économies d’échelle) fondent leurs avancées technologiques sur des sources externes de l’innovation telles que les concurrents, les fournisseurs et les utilisateurs avancés. De plus, ces firmes utilisent de manière plus importante des méthodes d’appropriation commerciale telles que les marques ou les stratégies marketing, car leurs produits sont moins exposés au risque d’imitation certes, mais aussi parce qu’elles sont sensibles aux changements de coûts. / This thesis is structured in three essays based on evolutionary theoretical grounds and provides empirical evidence from CIS. It aims at showing that the sources of innovation and the appropriation of innovation rents vary in function of firms’ activities and innovation strategies.In essay 1, we describe four waves of CIS, covering the period 1994-2006 and we study persistent innovation behavior with a discrete choice model on a data set of 431 firms. We find that innovation persistence is more important for product innovators because they need novel products to be more competitive and therefore enrich their base of knowledge continuously. By contrast, process innovators are less persistent because innovation strategy is less “market” oriented and intends to meet quality or production adjustments. The two last essays explore with the two stage least squares method how firms benefit economically from their innovations on a sample of 7 742 firms, on the period 2002-2005. We show that science-based firms rely more on R&D investments to develop their products and maintain their leads by acquiring complementary assets, i.e. they use mixed methods to appropriate the rents of innovation (the combined use IPRs and strategic methods for instance secrecy). By contrast, firms in other categories (for instance firms using cost-cutting strategies) draw more on external sources of knowledge coming either from suppliers or advanced users. Additionally, these firms use more extensively trademarks or non technological methods of appropriation (as marketing devices), because they are less exposed to potential imitation and because they are price sensitive.

Dans un marché en permanente progression et pour répondre aux besoins des brasseurs en malt de qualité, la maîtrise du procédé de maltage est indispensable. La qualité du malt est fortement dépendante des conditions opératoires, en particulier des conditions de trempe, mais également de la qualité de la matière première : l'orge. Dans cette étude, nous avons établi des modèles polynomiaux qui mettent en relation les conditions opératoires et la qualité du malt. Ces modèles ont été couplés à nos algorithmes génétiques et nous ont permis de déterminer les conditions optimales de maltage, soit pour atteindre une qualité ciblée de malt (friabilité), soit pour permettre un maltage à faible teneur en eau (pour réduire la consommation en eau et maîtriser les coûts environnementaux de production) tout en conservant une qualité acceptable de malt. Cependant, la variabilité de la matière première est un facteur limitant de notre approche. Les modèles établis sont en effet très sensibles à l'espèce d'orge (printemps, hiver) ou encore à la variété d'orge utilisée. Les modèles sont surtout très dépendants de l'année de récolte. Les variations observées sur les propriétés d'une année de récolte à une autre sont mal caractérisées et ne sont donc pas intégrées dans nos modèles. Elles empêchent ainsi de capitaliser l'information expérimentale au cours du temps. Certaines propriétés structurelles de l'orge (porosité, dureté) ont été envisagées comme nouveaux facteurs pour mieux caractériser la matière première mais ils n'ont pas permis d'expliquer les variations observés en malterie.Afin de caractériser la matière première, 394 échantillons d'orge issus de 3 années de récolte différentes 2009-2010-2011 ont été analysés par spectroscopie MIR. Les analyses ACP ont confirmé l'effet notable des années de récolte, des espèces, des variétés voire des lieux de culture sur les propriétés de l'orge. Une régression PLS a permis, pour certaines années et pour certaines espèces, de prédire les teneurs en protéines et en béta-glucanes de l'orge à partir des spectres MIR. Cependant, ces résultats, pourtant prometteurs, se heurtent toujours à la variabilité. Ces nouveaux modèles PLS peuvent toutefois être exploités pour mettre en place des stratégies de pilotage du procédé de maltage à partir de mesures spectroscopiques MIR / In a continuously growing market and in order to meet the needs of Brewers in high quality malt, control of the malting process is a great challenge. Malt quality is highly dependent on the malting process operating conditions, especially on the steeping conditions, but also the quality of the raw material: barley. In this study, we established polynomial models that relate the operating conditions and the malt quality. These models have been coupled with our genetic algorithms to determine the optimal steeping conditions, either to obtain a targeted quality of malt (friability), or to allow a malting at low water content while maintaining acceptable quality of malt (to reduce water consumption and control the environmental costs of malt production). However, the variability of the raw material is a limiting factor for our approach. Established models are very sensitive to the species (spring and winter barley) or to the barley variety. The models are especially highly dependent on the crop year. Variations on the properties of a crop from one to another year are poorly characterized and are not incorporated in our models. They thus prevent us to capitalize experimental information over time. Some structural properties of barley (porosity, hardness) were considered as new factors to better characterize barley but they did not explain the observed variations.To characterize barley, 394 samples from 3 years of different crops 2009-2010-2011 were analysed by MIR spectroscopy. ACP analyses have confirmed the significant effect of the crop-years, species, varieties and sometimes of places of harvest on the properties of barley. A PLS regression allowed, for some years and for some species, to predict content of protein and beta-glucans of barley using MIR spectra. These results thus still face product variability, however, these new PLS models are very promising and could be exploited to implement control strategies in malting process using MIR spectroscopic measurements

O sensoriamento remoto é uma estratégia que pode ajudar no monitoramento da qualidade das pastagens. Objetivou-se com esse estudo analisar a resposta espectral das folhas de Brachiaria brizantha cv. Marandú, adubada com doses crescentes de ureia, para diferenciar e predizer teores foliares de nitrogênio (TFN). Os tratamentos foram distribuídos em blocos ao acaso (DBC), composto por quatro blocos e quatro tratamentos, totalizando 16 parcelas. Foram utilizadas doses crescentes de adubação com ureia: 0, 25, 50, 75 kg de N/ha/corte. Ao longo do experimento foram realizadas 7 coletas, sendo coletadas 8 folhas por parcela. Essas folhas foram submetidas à análise hiperespectral e posterior análise química do teor de nitrogênio. Ao analisar a resposta espectral das folhas, observou-se diferenças estatísticas entre os tratamentos na região do visível em todas as coletas, com ênfase na região de 550 nm (verde). Por meio de análise discriminante linear (LDA) realizada para cada coleta, os centróides gerados por todos os tratamentos apresentaram diferenças significativas, com exceção do LD1 nas coletas 6 e 7 que não apresentou distinção entre os tratamentos de 50 e 75 kg de N/ha/corte, e LD2 na coleta 5 que não apresentou distinção entre os tratamentos de 0 e 50 kg de N/ha/corte. As equações de regressão multivariada obtidas pelo método de quadrados mínimos parciais (PLSR), geraram valores razoáveis a bons de R2 (0,53 a 0,83) na predição dos TFN, onde os comprimentos de onda com maior peso nessas regressões estão na região do red edge (715 a 720 nm). Por fim, ao testar a performance de alguns Índices de Vegetação da literatura, as coletas 4, 6 e 7 apresentaram bons coeficientes de determinação (R2) que variaram de 0,65 a 0,73; uma característica em comum nos índices que melhor estimaram os TFN é a presença de comprimentos de ondas que fazem parte da região do red edge. / Remote sensing is a set of techniques that can help to monitor pasture quality. The object of this study is to analyze the spectral response from Brachiaria brizantha cv. Marandú leaves, under contrasting nitrogen levels, to differentiate and predict leaf nitrogen content. The treatments were set in a Randomized Block Design, composed of four blocks and four treatments, totaling 16 plots. Increasing doses of urea fertilization were used: 0, 25, 50, 75 kg N/ha/mowing. During the experiment, 7 data collections were performed, and 8 leaves per plot were extracted for each data collection. These leaves were submitted to hyperspectral data extraction and subsequent chemical analysis to quantify the nitrogen content. When analyzing the spectral pattern of the leaves, statistical differences among samples with different nitrogen levels were noticeable in the visible range of the spectrum in all the collections, with emphasis on the 550 nm region (green). Through linear discriminant analysis (LDA), performed for each collection, the generated centroids by the samples of each nitrogen level presented significant differences, except for LD1 in collections 6 and 7, which did not present a distinction between treatments of 50 and 75 kg of N/ha/mowing, and LD2 in collection 5 that did not distinguish between treatments of 0 and 50 kg of N/ha/mowing. The partial least square regression (PLSR) method generated reasonable to good values of R2 (0.53 to 0.83) for the prediction of leaf nitrogen content, where the wavelengths with the highest coefficient in these models are in the red edge region of the spectrum (715 to 720 nm). Finally, when testing the performance of some Vegetation Indexes from literature, collections 4, 6 and 7 presented good determination coefficients (R2) ranging from 0.65 to 0.73; a common feature in the indexes that best estimate the nitrogen content is the presence of wavelengths from the red edge region of the spectrum.

Return on Capital Employed (ROCE) is a popular financial instrument and communication tool for the appraisal of companies. Often, companies management and other practitioners use untested rules and behavioural approach when investigating the key determinants of ROCE, instead of the scientific statistical paradigm. The aim of this dissertation was to identify and quantify key determinants of ROCE of individual companies listed on the Johannesburg Stock Exchange (JSE), by comparing classical multiple linear regression, principal components regression, generalized least squares regression, and robust maximum likelihood regression approaches in order to improve companies decision making. Performance indicators used to arrive at the best approach were coefficient of determination ( ), adjusted ( , and Mean Square Residual (MSE). Since the ROCE variable had positive and negative values two separate analyses were done. The classical multiple linear regression models were constructed using stepwise directed search for dependent variable log ROCE for the two data sets. Assumptions were satisfied and problem of multicollinearity was addressed. For the positive ROCE data set, the classical multiple linear regression model had a of 0.928, an of 0.927, a MSE of 0.013, and the lead key determinant was Return on Equity (ROE),with positive elasticity, followed by Debt to Equity (D/E) and Capital Employed (CE), both with negative elasticities. The model showed good validation performance. For the negative ROCE data set, the classical multiple linear regression model had a of 0.666, an of 0.652, a MSE of 0.149, and the lead key determinant was Assets per Capital Employed (APCE) with positive effect, followed by Return on Assets (ROA) and Market Capitalization (MC), both with negative effects. The model showed poor validation performance. The results indicated more and less precision than those found by previous studies. This suggested that the key determinants are also important sources of variability in ROCE of individual companies that management need to work with. To handle the problem of multicollinearity in the data, principal components were selected using Kaiser-Guttman criterion. The principal components regression model was constructed using dependent variable log ROCE for the two data sets. Assumptions were satisfied. For the positive ROCE data set, the principal components regression model had a of 0.929, an of 0.929, a MSE of 0.069, and the lead key determinant was PC4 (log ROA, log ROE, log Operating Profit Margin (OPM)) and followed by PC2 (log Earnings Yield (EY), log Price to Earnings (P/E)), both with positive effects. The model resulted in a satisfactory validation performance. For the negative ROCE data set, the principal components regression model had a of 0.544, an of 0.532, a MSE of 0.167, and the lead key determinant was PC3 (ROA, EY, APCE) and followed by PC1 (MC, CE), both with negative effects. The model indicated an accurate validation performance. The results showed that the use of principal components as independent variables did not improve classical multiple linear regression model prediction in our data. This implied that the key determinants are less important sources of variability in ROCE of individual companies that management need to work with. Generalized least square regression was used to assess heteroscedasticity and dependences in the data. It was constructed using stepwise directed search for dependent variable ROCE for the two data sets. For the positive ROCE data set, the weighted generalized least squares regression model had a of 0.920, an of 0.919, a MSE of 0.044, and the lead key determinant was ROE with positive effect, followed by D/E with negative effect, Dividend Yield (DY) with positive effect and lastly CE with negative effect. The model indicated an accurate validation performance. For the negative ROCE data set, the weighted generalized least squares regression model had a of 0.559, an of 0.548, a MSE of 57.125, and the lead key determinant was APCE and followed by ROA, both with positive effects.The model showed a weak validation performance. The results suggested that the key determinants are less important sources of variability in ROCE of individual companies that management need to work with. Robust maximum likelihood regression was employed to handle the problem of contamination in the data. It was constructed using stepwise directed search for dependent variable ROCE for the two data sets. For the positive ROCE data set, the robust maximum likelihood regression model had a of 0.998, an of 0.997, a MSE of 6.739, and the lead key determinant was ROE with positive effect, followed by DY and lastly D/E, both with negative effects. The model showed a strong validation performance. For the negative ROCE data set, the robust maximum likelihood regression model had a of 0.990, an of 0.984, a MSE of 98.883, and the lead key determinant was APCE with positive effect and followed by ROA with negative effect. The model also showed a strong validation performance. The results reflected that the key determinants are major sources of variability in ROCE of individual companies that management need to work with. Overall, the findings showed that the use of robust maximum likelihood regression provided more precise results compared to those obtained using the three competing approaches, because it is more consistent, sufficient and efficient; has a higher breakdown point and no conditions. Companies management can establish and control proper marketing strategies using the key determinants, and results of these strategies can see an improvement in ROCE. / Mathematical Sciences / M. Sc. (Statistics)

Les gyroscopes laser à gaz constituent une solution technique de haute performances dans les problématiques de navigation inertielle. Néanmoins, pour de très faibles vitesses de rotation, les petites imperfections des miroirs de la cavité optique font que les deux faisceaux contra-propageant sont verrouillés en phase. En conséquence, les mesures en quadrature de leur différence de phase ne permettent plus de remonter directement aux vitesses de rotation à l'intérieur d'une zone autour de zéro, dite zone aveugle statique, ou, si l'on utilise une procédure d'activation mécanique, dite zone aveugle dynamique. Ce travail montre qu'il est néanmoins possible, en utilisant des méthodes issues du filtrage et de l'estimation, de remonter aux vitesses de rotation mêmes si ces dernières sont en zone aveugle. Pour cela, on part d'une modélisation physique de la dynamique que l'on simplifie par des techniques de perturbations singulières pour en déduire une généralisation des équations de Lamb. Il s'agit de quatre équations différentielles non-linéaires qui décrivent la dynamique des intensités et des phases des deux faisceaux contra-propageant. Une étude qualitative par perturbations régulières, stabilité exponentielle des points d'équilibre et applications de Poincaré permet de caractériser les zones aveugles statiques et dynamiques en fonction des imperfections dues aux miroirs. Il est alors possible d'estimer en ligne avec un observateur asymptotique fondé sur les moindre carrés récursifs ces imperfections en rajoutant aux deux mesures en quadrature celles des deux intensités. La connaissance précise de ces imperfections permet alors de les compenser dans la dynamique de la phase relative, et ainsi d'estimer les rotations en zone aveugle. Des simulations numériques détaillées illustrent l'intérêt de ces observateurs pour augmenter la précision des gyroscopes à gaz. / Gaz ring laser gyroscopes provide a high performance technical solution for inertial navigation. However, for very low rotational speeds, the mirrors imperfections of the optical cavity induce a locking phenomena between the phases of the two counter-propagating Laser beams. Hence, the measurements of the phase difference can no longer be used when the speed is within an area around zero, called lock-in zone, or,if a procedure of mechanical dithering is implemented, dithering lock-in zone. Nevertheless, this work shows that it is possible using filtering and estimation methods to measure the speed even within the lock-in zones. To achieve this result, we exploit a physical modeling of the dynamics that we simplify, using singular perturbation techniques, to obtain a generalization of Lamb's equations. There are four non-linear differential equations describing the dynamics of the intensities and phases of the two counter-propagating beams. A qualitative study by regular perturbation theory, exponential stability of the equilibrium points and Poincaré maps allows a characterisation of the lock-in zones according to the mirrors imperfections. It is then possible to estimate online, with an asymptotic observer based on recursive least squares, these imperfections by considering the additional measurements of the beam intensities. Accurate knowledge of these imperfections enables us to compensate them in the dynamic of the relative phase, and thus to estimate rotational speeds within the lock-in zones. Detailed numerical simulations illustrate the interest of those observers to increase the accuracy of gas ring laser gyroscopes.

Les composés électrochomes changent de couleur en fonction d'une tension qui leur est appliquée. La tenue en cyclage de couches minces à base de NiO, électrochrome cathodique, dépend fortement de la température et de la pression de dépôt. D’autre part la proportion de phase électrochimiquement active dépend fortement de l’épaisseur des couches. Selon le modèle proposé NiO en milieu KOH se transforme en hydroxyde de nickel Ni(OH)2 puis dans la phase de coloration en oxyhydroxyde NiOOH avant de revenir à sa forme réduite durant la phase décoloration. À la fin de ce cycle, des traces de phase colorée persistent. Nous avons développé des nouvelles méthodes numériques d’analyse des spectres d'absorption de rayons X, caractérisant la structure locale autour du nickel dans ces phases non cristallisées, afin de déterminer les concentrations des diverses espèces au cours du cycle. Aucune des méthodes habituellement pratiquées, comme la combinaison linéaire d’espèces modèles par la méthode des moindres carrés linéaire, et la méthode d’Analyse en Composantes Principales, ne sont adaptées aux cas de spectres fortement corrélés comme les oxydes-hydroxydes de nickel. Nous montrons qu’il est possible d’améliorer la méthode des moindres carrés en utilisant un algorithme original : « la méthode des Moindres Carrés Linéaire Progressive ». La base de cette méthode est fondée sur l’étude statistique des erreurs et corrélations. Ce travail a permis de valider le modèle électrochimique et d’évaluer pour la première fois la concentration en NiOOH dans la phase réduite, signature de l'irréversibilité. Seule, la présence avant cyclage de Ni(OH)2 dans les films de NiO n'a pu être expliquée / Electrochromic materials change their color versus an applied electric voltage. The electrochemical study of NiO thin films, which are cathodic electromic materials, shows an important oxidationreduction cycles lifetime dependance on thin films deposition temperature and pressure. The proportion of electrochemically active material depends on the film thickness. The proposed model suggests that in KOH NiO is transformed first to nickel hydroxide Ni(OH)2, and to nickel oxihydroxide in the coloration process, before returning to Ni(OH)2 in the decoloration step. At the end of the cycle, traces of the colored phase are still observed. We have developed a new X-Ray absorption spectroscopy numerical analysis method in order to characterize the concentration of each species present in these non crystalline materials. Mixed Ni oxides and hydroxides XAS spectra are highly correlated and Linear Least Squares and Principal Component Analysis methods proved to be totally inefficient. In order to improve the linear least squares method, we have developed a home-made algorithm named « Progressive Linear Least Squares method ». This method is based on the use of statistical errors and correlations evaluation of the spectra. It was then possible to valid the electrochemical model and to evaluate for the first time the concentration of residual NiOOH in the reduced thin film phases, related to the irreversibility of the electrochromic process. However, we were unable to explain the presence of a small amount of de Ni(OH)2 in the NiO films prepared at room temperature, prior to any electrochemical treatment

La littérature sur le stress au travail considère souvent que les personnes dépourvues de soutien social tendent à avoir un taux de stress plus élevé. Si cette vision est confirmée empiriquement, elle a toutefois une portée limitée : elle ne tient pas toujours compte de l’inégalité d’accès au soutien, inégalité qui affecte la perception de ce soutien. Pourquoi certains salariés ont plus de facilité à accéder au soutien social ? Qu’est-ce qui fait que l’aide est plus disponible et plus variée pour une personne plutôt que pour une autre ? Ces interrogations nous amènent à situer le soutien social perçu, et plus précisément le soutien des collègues perçu, dans un modèle théorique plus large nourri par la théorie des réseaux sociaux. A l’aide d’un modèle explicatif, l’objectif de notre recherche est d’étudier l’impact du positionnement de l’individu dans le réseau social sur le stress au travail perçu. Les hypothèses de recherche ont été testées en utilisant les techniques de régression en moindres carrés partiels pour estimer les équations structurelles. A partir de données de type « réseau complet » collectées auprès d’une entreprise de services de taille moyenne (N=343), nous avons montré que la force des liens favorise l’accès au soutien des collègues et, par conséquent, réduit le stress professionnel. Les résultats indiquent que le soutien des collègues est médiateur total dans cette relation, et que le lien direct entre la force des liens et le stress perçu n’est pas établi. De plus, nous avons confirmé l’ambivalence des bridging ties (liens vers des personnes de départements différents) : ils influencent négativement la perception du soutien social (qui réduit le stress), mais ont aussi un effet négatif direct sur le stress au travail. En soulignant le rôle des relations informelles comme antécédent au soutien social, nous avons contribué à fournir un outil analytique susceptible d’être mis en œuvre dans la sphère managériale. / The literature on workplace stress often considers that people who lack social support tend to have higher levels of perceived stress. This is empirically confirmed, but it is not always taken into account and offers limited scope. Indeed, why do some employees have more access to social support? What renders support more available and more varied from one person to the other? These questions allow us to situate perceived social support, and more accurately the perceived support of colleagues, in a larger theoretical model, enhanced by social network theory. Through an explanatory model, the objective of this research is to explore the role of the positioning of the individual in the social network on perceived workplace stress. Based on a “complete network” data in a medium-sized IT services company, we used partial least squares to test our hypothesis (N = 343). The strength of ties affects stress through social support, such that people with stronger ties perceive more support and ultimately exhibit less stress. However, the direct link between strength and stress is not established. Bridging ties (supportive ties to other departments) negatively influence social support (a situation which increases stress) but also have a direct negative effect on stress. By stressing the role of social relationships as an antecedent of social support and stress, our results offer new potential managerial actions within organizations

A associação dos métodos cinéticos de análises e dos sistemas de análises em fluxo foi demonstrada em relação à determinação espectrofotométrica de ferro e vanádio em ligas Fe-V O método se baseia na influência de Fe2+ e VO2+ na taxa de oxidação de iodeto por dicromato sob condições ácidas; por esta razão o emprego do redutor de Jones foi necessário. Um sistema de análises por injeção em fluxo (FIA) e um sistema multi-impulsão foram dimensionados e avaliados. Em ambos os sistemas, a solução da amostra era inserida no fluxo transportador / reagente iodeto, e a solução de dicromato era adicionada por confluência. Sucessivas medidas eram realizadas durante a passagem da zona de amostra processada pelo detector, cada uma relacionada a uma diferente condição para o desenvolvimento da reação. O tratamento dos dados envolveu calibração multivariada, particularmente o algorítmo PLS. O sistema FIA se mostrou pouco adequado para as determinações multi-paramétricas, uma vez que os elementos de fluído resultantes da natureza de escoamento laminar não continham informações cinéticas suficientes para compor as etapas de modelagem. Por outro lado, MPFS mostrou que a natureza do fluxo pulsado resulta em melhorias nas figuras de mérito devido ao movimento caótico dos elementos de fluído. O sistema proposto é simples e robusto, capaz de analisar 50 amostras por hora, significando em um consumo de 48 mg KI por determinação. A duas primeiras variáveis latentes contém ca 94 % da informação analítica, mostrando que a dimensionalidade dupla intrínsica ao conjunto de dados. Os resultados se apresentaram concordantes com aqueles obtidos por espectrometria de emissão optica com plasma induzido em argônio. / Differential kinetic analysis can be implemented in a flow system analyser, and this was demonstrated in designing an improved spectrophotometric catalytic determination of iron and vanadium in Fe-V alloys. The method relied on the influence of Fe2+ and VO2+ on the rate of the iodide oxidation by Cr2O7 under acidic conditions; therefore the Jones reductor was needed. To this end, a flow injection system (FIA) and a multi-pumping flow system (MPFS) were dimensioned and evaluated. In both systems, the alloy solution was inserted into an acidic KI solution that acted also as carrier stream, and a dichromate solution was added by confluence. Successive measurements were performed during sample passage through the detector, each one related to a different yet reproducible condition for reaction development. Data treatment involved multivariate calibration by the PLS algorithm. The FIA system was less recommended for multi-parametric determination, as the laminar flow regimen could not provide suitable kinetic information. On the other hand, a MPFS demonstrated that pulsed flow led to enhance figures of merit due to chaotic movement of its fluid elements. The proposed MPFS system is very simple and rugged, allowing 50 samples to be run per hour, meaning 48 mg KI per determination. The first two latent variables carry ca 94 % of the analytical information, pointing out that the intrinsic dimensionality of the data set is two. Results are in agreement with inductively coupled argon plasma – optical emission spectrometry.

Cette thèse s'intéresse aux méthodes d'itération sur les politiques dans l'apprentissage par renforcement à grand espace d'états avec approximation linéaire de la fonction de valeur. Nous proposons d'abord une unification des principaux algorithmes du contrôle optimal stochastique. Nous montrons la convergence de cette version unifiée vers la fonction de valeur optimale dans le cas tabulaire, ainsi qu'une garantie de performances dans le cas où la fonction de valeur est estimée de façon approximative. Nous étendons ensuite l'état de l'art des algorithmes d'approximation linéaire du second ordre en proposant une généralisation de Least-Squares Policy Iteration (LSPI) (Lagoudakis et Parr, 2003). Notre nouvel algorithme, Least-Squares [lambda] Policy Iteration (LS[lambda]PI), ajoute à LSPI un concept venant de [lambda]-Policy Iteration (Bertsekas et Ioffe, 1996) : l'évaluation amortie (ou optimiste) de la fonction de valeur, qui permet de réduire la variance de l'estimation afin d'améliorer l'efficacité de l'échantillonnage. LS[lambda]PI propose ainsi un compromis biais-variance réglable qui peut permettre d'améliorer l'estimation de la fonction de valeur et la qualité de la politique obtenue. Dans un second temps, nous nous intéressons en détail au jeu de Tetris, une application sur laquelle se sont penchés plusieurs travaux de la littérature. Tetris est un problème difficile en raison de sa structure et de son grand espace d'états. Nous proposons pour la première fois une revue complète de la littérature qui regroupe des travaux d'apprentissage par renforcement, mais aussi des techniques de type évolutionnaire qui explorent directement l'espace des politiques et des algorithmes réglés à la main. Nous constatons que les approches d'apprentissage par renforcement sont à l'heure actuelle moins performantes sur ce problème que des techniques de recherche directe de la politique telles que la méthode d'entropie croisée (Szita et Lorincz, 2006). Nous expliquons enfin comment nous avons mis au point un joueur de Tetris qui dépasse les performances des meilleurs algorithmes connus jusqu'ici et avec lequel nous avons remporté l'épreuve de Tetris de la Reinforcement Learning Competition 2008 / This thesis studies policy iteration methods with linear approximation of the value function for large state space problems in the reinforcement learning context. We first introduce a unified algorithm that generalizes the main stochastic optimal control methods. We show the convergence of this unified algorithm to the optimal value function in the tabular case, and a performance bound in the approximate case when the value function is estimated. We then extend the literature of second-order linear approximation algorithms by proposing a generalization of Least-Squares Policy Iteration (LSPI) (Lagoudakis and Parr, 2003). Our new algorithm, Least-Squares [lambda] Policy Iteration (LS[lambda]PI), adds to LSPI an idea of [lambda]-Policy Iteration (Bertsekas and Ioffe, 1996): the damped (or optimistic) evaluation of the value function, which allows to reduce the variance of the estimation to improve the sampling efficiency. Thus, LS[lambda]PI offers a bias-variance trade-off that may improve the estimation of the value function and the performance of the policy obtained. In a second part, we study in depth the game of Tetris, a benchmark application that several works from the literature attempt to solve. Tetris is a difficult problem because of its structure and its large state space. We provide the first full review of the literature that includes reinforcement learning works, evolutionary methods that directly explore the policy space and handwritten controllers. We observe that reinforcement learning is less successful on this problem than direct policy search approaches such as the cross-entropy method (Szita et Lorincz, 2006). We finally show how we built a controller that outperforms the previously known best controllers, and shortly discuss how it allowed us to win the Tetris event of the 2008 Reinforcement Learning Competition