No description available.

L'apprentissage statistique propose un vaste ensemble de techniques capables de construire des modèles prédictifs à partir d'observations passées. Ces techniques ont montré leurs capacités à traiter des volumétries importantes de données sur des problèmes réels. Cependant, de nouvelles applications génèrent de plus en plus de données qui sont seulement visibles sous la forme d'un flux et doivent être traitées séquentiellement. Parmi ces applications on citera : la gestion de réseaux de télécommunications, la modélisation des utilisateurs au sein d'un réseau social, le web mining. L'un des défis techniques est de concevoir des algorithmes permettant l'apprentissage avec les nouvelles contraintes imposées par les flux de données. Nous proposons d'abord ce problème en proposant de nouvelles techniques de résumé de flux de données dans le cadre de l'apprentissage supervisé. Notre méthode est constituée de deux niveaux. Le premier niveau utilise des techniques incrémentales de résumé en-ligne pour les flux qui prennent en compte les ressources mémoire et processeur et possèdent des garanties en termes d'erreur. Le second niveau utilise les résumés de faible taille, issus du premier niveau, pour construire le résumé final à l'aide d'une méthode supervisée performante hors-ligne. Ces résumés constituent un prétraitement qui nous permet de proposer de nouvelles versions du classifieur bayésien naïf et des arbres de décision fonctionnant en-ligne sur flux de données. Les flux de données peuvent ne pas être stationnaires mais comporter des changements de concept. Nous proposons aussi une nouvelle technique pour détecter ces changements et mettre à jour nos classifieurs. / Statistical learning provides numerous algorithms to build predictive models on past observations. These techniques proved their ability to deal with large scale realistic problems. However, new domains generate more and more data which are only visible once and need to be processes sequentially. These volatile data, known as data streams, come from telecommunication network management, social network, web mining. The challenge is to build new algorithms able to learn under these constraints. We proposed to build new summaries for supervised classification. Our summaries are based on two levels. The first level is an online incremental summary which uses low processing and address the precision/memory tradeoff. The second level uses the first layer summary to build the final sumamry with an effcient offline method. Building these sumamries is a pre-processing stage to develop new classifiers for data streams. We propose new versions for the naive-Bayes and decision trees classifiers using our summaries. As data streams might contain concept drifts, we also propose a new technique to detect these drifts and update classifiers accordingly.

[pt] Esse trabalho usa dados de empregador e empregado para investigar o canal de crédito como um canal de transmissão relevante da crise financeira em 2008 sobre o mercado de trabalho. Eu estudo o impacto entre setores da indústria das decisões das firmas em relação à força de trabalho. Eu uso medidas como taxas de contratação e demissão. Eu encontro que crédito foi um mecanismo de transmissão importante da crise sobre a economia real. Mais especificamente, eu encontro que as taxas de demissão são maiores para firmas mais dependentes de financiamento durante a crise. Trabalhadores mais jovens e menos qualificados foram mais afetados negativamente, através do canal de crédito. Eu também encontro evidências de realocação de trabalhadores entre setores; setores menos dependentes de financiamento roubaram mais trabalhadores de outros setores, especialmente de firmas pequenas. Eu também encontro evidências de realocação de trabalhadores dentro dos setores e entre intervalos de tamanho de firma. / [en] This paper uses matched employer-employee data to investigate the credit channel as a relevant transmission channel of the 2008 financial crisis on the labor market in Brazil. I study the cross-sector impact on employment decisions of firms of the manufacturing industry. I use measures such as hiring and firing rates as outcomes. I find that the credit was an important transmission mechanism of the crises to the real economy. More specifically, I find that the firing rate is higher for more financially dependent industries during the crisis. Younger and less skilled workers were more adversely affected through the credit channel. I also find evience of reallocation of workers across sectors; less financially dependent sectors poached more workers from other sectors, and particularly so from smaller firms. I also find some evidence of reallocation within sector and across firm size intervals.

This study provides an empirical analysis of Shadow banking as a factor influencing cross- border financial flows. It builds upon emerging literature on Shadow banking and empirical literature on global imbalances and global financial flows. The aim of the thesis is to test three hypotheses, which relate global financial flows to lending, change in cross-border bank liabilities, and shadow banking, respectively. The second and third hypotheses are tested on gross capital flows, which, in contrast to net flows, better reflect financing activities. The results suggest that Shadow banking activities are related to higher gross capital flows in periods, when this sector is growing. These flows, however, tend to dry up when Shadow banking activities level off or decline. Among other important factors is the output growth differential and global risk aversion. JEL Classification E44, G20, G23, F32, F34, F21, F65 Keywords Shadow Banking, Global Imbalances, Capital Flows, Financial Stability Author's e-mail tomas.vaclavicek@gmail.com Supervisor's e-mail petr.teply@fsv.cuni.cz

The ability of Next-Generation Sequencing (NGS) to produce massive quantities of genomic data inexpensively has allowed to study the structure of viral populations from an infected host at an unprecedented resolution. As a result of a high rate of mutation and recombination events, an RNA virus exists as a heterogeneous "swarm". Virologists and computational epidemiologists are widely using NGS data to study viral populations. However, discerning rare variants is muddled by the presence of errors introduced by the sequencing technology. We develop and implement time- and cost-efficient strategy for NGS of multiple viral samples, and computational methods to analyze large quantities of NGS data and to handle sequencing errors. In particular, we present: (i) combinatorial pooling strategy for massive NGS of viral samples; (ii) kGEM and 2SNV — methods for viral population haplotyping; (iii) ShotMCF — a Multicommodity Flow (MCF) based method for frequency estimation of viral haplotypes; (iv) QUASIM — an agent-based simulator of viral evolution taking in account viral variants and immune response.

Cette thèse est consacrée à l’étude du drainage dans des micromodèles de milieu poreux. Les techniques classiques de microfluidique (verre, PDMS) sont utilisées pour la réalisation de poreux modèles 2D de type grille. Des techniques de traitement de surface permettent de faire varier les conditions de mouillage. Le mouillage total, pseudo-partiel et partiel est étudié. Des méthodes d’analyse d’images sont développées afin de quantifier les vitesses locales lors de l’invasion du milieu poreux, vitesses qui sont ensuite comparées aux vitesses imposées. Ces données mettent en évidence avant percolation des comportements très différents selon la nature du mouillage. Un modèle permet de rendre compte des phénomènes observés. Nous expliquons pourquoi l’évolution après la percolation diffère selon la nature du mouillage. Enfin, l’influence de la rhéologie du fluide pousseur est abordée, celle d’un balayage initial à l’eau à très faible nombre capillaire également. / Drainage experiments in model porous media are performed. Our 2D micromodels consist of a regular network; they are made in glass or in PDMS. The wetting properties of the chip vary from total, pseudo-partial to partial wetting. Experiments are performed under flow rate control. Taking advantages of microfluidic devices, local velocities of the injected fluid are measured. The average of all these local velocities is compared to the velocity imposed by the syringe pump. Before percolation, the invasion percolation process for all wetting cases is studied. Depending on the wetting properties, several behaviours are observed. We develop a model to explain our experimental data. After percolation, the effects of wetting are huge; we explain why oil could remain trapped or not. Finally, we consider the influence of rheology by injecting non Newtonian liquids as pushing fluids, and also the effects of preferential paths.

In this work the predictive capability of a number of Reynolds stress transport(RST) models was first tested in a range of non-equilibrium homogeneous flows, comparisons being drawn with existing direct numerical simulation (DNS) results and physical measurements. The cases considered include both shear and normally strained flows, in some cases with a constant applied strain rate, and in others where this varied with time. Models were generally found to perform well in homogeneous shear at low shear rates, but their performance increasingly deteriorated at higher shear rates. This was attributed mainly to weaknesses in the pressure-strain rate models, leading to over-prediction of the shear stress component of the stress anisotropy tensor at high shear rates. Performance in irrotational homogeneous strains was generally good, and was more consistent over a much wider range of strain rates. In the experimental plane strain and axisymmetric contraction cases, with time-varying strain rates, there was evidence of an accelerated dissipation rate generation. Significant improvement was achieved through the use of an alternative dissipation rate generation term, Pε , in these cases, suggesting a possible route for future modelling investigation. Subsequently, the models were also tested in the inhomogeneous case of pulsating channel flow over a wide range of frequencies, the reference for these cases being the LES of Scotti and Piomelli (2001). A particularly challenging feature in this problem set was the partial laminarisation and re-transition that occurred cyclically at low and, to a lesser extent, intermediate frequencies. None of the models tested were able to reproduce correctly all of the observed flow features, and none returned consistently superior results in all the cases examined. Finally, models were tested in the case of a plane jet interacting with a rectangular dead-end enclosure. Two geometric configurations are examined, corresponding a steady regime, and an intrinsically unsteady regime in which periodic flow oscillations are experimentally observed (Mataoui et al., 2003). In the steady case generally similar flow patterns were returned by the models tested, with some differences arising in the degree of downward deflection of the impinging jet, which in turn affected the level of turbulence energy developing in the lower part of the cavity. In the unsteady case, only two of the models tested, a two-equation k-ε model and an advanced RST model, correctly returned purely periodic solutions. The other two RST models, based on linear pressure-strain rate terms, returned unsteady flow patterns that exhibited complex oscillations with significant cycle-to-cycle variations. Unfortunately, the limited availability of reliable experimental data did not allow a detailed quantitative examination of model performance.

The fast growth of science parks around the world has inspired many researchers to investigate the function and performance of science parks (SPs). Interestingly, mixed findings are reported on science park performance in the literature: some researchers found that SPs have benefits for the firms located on site; whereas other researchers doubt the benefits that SPs are claimed to have. This thesis aims to explain these mixed findings and proposes a relational approach to study the general view of interorganisational knowledge flows. The main research question to be answered is: How can the mixed findings of previous research studies regarding innovative performances of science park firms be explained? A sample of 52 new technology-based firms (NTBFs) situated in Gauteng was interviewed by means of structured questionnaires. Despite the fact that this research did not empirically confirm the theoretical explanation of the mixed findings found in science park literature, some issues raised in the recommendation section of this thesis could account for the mixed findings, namely differences in the scanning processes of new entrant firms, the nature of networking activities, services provided by SP management teams, academic-industry links and configurations of knowledge flows. The findings and recommendations of this study may help policy makers to further improve the design and functioning of science parks in emerging economies. / Thesis (PhD(Technology Management))--University of Pretoria, 2010. / Graduate School of Technology Management (GSTM) / unrestricted

The remarkable efficacy with which normal surface mass transpiration (blowing and suction) alters a given base flow to achieve a desired predefined objective has motivated several investigations on drag reduction, self-propulsion and suppression of separation and wake unsteadiness in bluff body flows. However, the energetic efficiency, a critical parameter that determines the true efficacy and in particular practical feasibility of this control strategy, has received significantly less attention. In this work, we determine the optimal zero net mass transpiration blowing and suction profiles that minimize net power consumption while reducing drag or enabling self-propulsion in typical bluff body flows. We establish the influence of prescribed blowing and suction profiles on the hydrodynamic loads and net power consumption for a representative bluff body flow involving flow past a stationary two-dimensional circular cylinder. Using analysis based on Oseen’s equations, we find that all the symmetric modes, except the first one, lead to an increase in the net power consumption without affecting hydrodynamic drag. The optimal blowing and suction profile that yields minimum power consumption is such that the normal stress acting on the cylinder surface vanishes identically. Furthermore, we show that a self-propelling state corresponding to zero net drag force is attained when the first mode of blowing and suction profile is such that the flow field be-comes irrigational. Based on these findings we employ direct numerical simulation tools to decipher the Reynolds number dependence of the optimal profiles and the associated power consumption for both drag reduction and self-propulsion. For a typical Reynolds number, the time-averaged drag coefficient first decreases due to vortex shedding suppression, then increases and eventually decreases again after attaining a local maximum as the strength of the first mode is increased. The net power consumption continues to decrease with an increase in the strength of the first mode before reaching a minima after which it rises continuously. For a Reynolds number of 1000 over fifteen fold reduction in drag is achieved for an optimal blowing and suction profile with a maximum radial surface velocity that is nearly 1.97 times the free stream velocity. Next, to establish whether or not higher modes play a role in decreasing net power consumption at finite Reynolds number, we perform theoretical analysis of a configuration similar to the one described above for a spherical body. At zero Reynolds number, as a result of mode independence, we show that surface blow-ing and suction of any form that involves second or higher order axisymmetric or non-axisymmetric modes does not contribute to drag and only leads to an increase in total power consumption. However, at finite Reynolds number, using analysis based on Oseen’s equations, we find that the second and higher modes contribute substantially to the optimal profiles. Finally to understand the effects of a change in shape we consider generalization of the above analysis to axisymmetric prolate and oblate spheroidal bodies. We find that for a general axisymmetric body with non-constant curvature, the optimal drag reducing and self-propelling blowing and suction profiles for minimum power consumption contain second and higher-order modes along with the first mode even when the Reynolds number is zero. The net decrease in power consumption with the use of second and higher order modes exceeds 33% for a disk-like low aspect ratio self-propelling oblate spheroid. Moreover, we perform comparisons between blowing and suction and tangential surface velocity based boundary deformation propulsion mechanisms. Below an aspect ratio of 0.56 we find blowing and suction mechanism to be more efficient for self-propulsion of an oblate spheroid. In contrast, for a self-propelling pro-late spherical micro-swimmer, we show that the tangential surface tread milling consumes less power irrespective of the aspect ratio.

Les écoulements pyroclastiques sont des mélanges à haute température de gaz et de particules volcaniques qui peuvent se propager sur de très grandes distances. Cette forte « mobilité » est souvent attribuée à leur capacité à se fluidiser, c’est-à-dire à générer et conserver une forte pression interstitielle de gaz qui réduit les forces de friction interne. L’objectif principal de cette thèse est de comprendre comment les irrégularités des terrains sur lesquelles se propagent les écoulements pyroclastiques peuvent favoriser leur fluidisation. Une première série d’expériences de laboratoire a consisté à générer des écoulements de particules fines (diamètre de 45-90 μm) sur des substrats de différentes rugosités. Les résultats montrent que la distance de parcours des écoulements augmente avec la rugosité, allant jusqu’à doubler par rapport à la distance de parcours sur fond lisse. Des analyses de vidéos haute vitesse et des mesures de pression interstitielle d’air à la base des écoulements montrent que la tête (partie antérieure) des écoulements qui se propagent sur un substrat rugueux s’auto-fluidisent en conséquence de la sédimentation des particules dans les interstices du substrat, chassant l’air qui remonte et percole dans l’écoulement. Ce mécanisme d’auto-fluidisation est efficace pour toutes les inclinaisons étudiées (0-30°), suggérant qu’il est susceptible de se produire tout au long de la mise en place d’un écoulement pyroclastique. Une seconde étude a consisté à faire chuter des lits de particules dans une colonne statique. Les résultats montrent que même pour une hauteur de relâchement relativement faible (20 cm), le mélange peut entièrement s’auto-fluidiser durant sa chute. Quand les particules sont suffisamment fines (<100 μm) la pression interstitielle dans le dépôt diffuse pendant plusieurs secondes, la durée de cette diffusion augmentant avec l’augmentation de l’épaisseur du lit et la diminution de taille des particules. Les temps de diffusions les plus longs sont observés avec un matériau provenant d’un dépôt d’écoulement pyroclastique (~30 s pour des lits de 28.5 cm d’épaisseur). Ces résultats suggèrent que les écoulements pyroclastiques qui se propagent sur des terrains accidentés peuvent s’auto-fluidiser et conserver une faible friction au cours de leur mise en place. / Pyroclastic flows are hot mixtures of gas and particles that can propagate over large distances. This high “mobility” is often attributed to their ability to be fluidized, that is, to generate and retain high gas pore pressure that reduces internal friction forces. The main objective of this thesis is to understand how irregularities of substrates on which pyroclastic flows propagate can enhance their fluidization. A first set of laboratory experiments consisted of the generation of fine-grained flows (diameter of 45-90 μm) on substrate of various roughness. Results show that the flow runout distance increases with the substrate roughness, and is up to twice the runout on a smooth substrate. High speed video analyses and air pore pressure measurements at the flow base show that the flow head propagating over a rough substrate can auto-fluidize because of particles sedimentation into the substrate interstices, which forces the air to escape upward and percolate through the flow. This auto-fluidization mechanism is efficient at all inclinations investigated (0-30°), suggesting that it could occur during the whole emplacement of a pyroclastic flow. A second study consisted of the vertical release of beds of particles in a static column. Results show that the granular mixture can be fully fluidized, even when collapsing from a relatively low height (20 cm). When particles are fine enough (<100 μm), pore pressure in the deposit diffuses for several seconds, the diffusion duration increasing with increasing bed thickness and decreasing particle size. The longest diffusion durations are observed for pyroclastic flow deposit materials (~30 s for 28.5 cm thick beds). These results suggest that pyroclastic flows propagating on irregular terrains can auto-fluidize and preserve low internal friction during their emplacement.