In recent years, deep neural network models have shown to outperform many state of the art algorithms. The reason for this is, unsupervised pretraining with multi-layered deep neural networks have shown to learn better features, which further improves many supervised tasks. These models not only automate the feature extraction process but also provide with robust features for various machine learning tasks. But the unsupervised pretraining and feature extraction using multi-layered networks are restricted only to the input features and not to the output. The performance of many supervised learning algorithms (or models) depends on how well the output dependencies are handled by these algorithms [Dembczy´nski et al., 2012]. Adapting the standard neural networks to handle these output dependencies for any specific type of problem has been an active area of research [Zhang and Zhou, 2006, Ribeiro et al., 2012]. On the other hand, inference into multimodal data is considered as a difficult problem in machine learning and recently ‘deep multimodal neural networks’ have shown significant results [Ngiam et al., 2011, Srivastava and Salakhutdinov, 2012]. Several problems like classification with complete or missing modality data, generating the missing modality etc., are shown to perform very well with these models. In this work, we consider three nontrivial supervised learning tasks (i) multi-class classification (MCC), (ii) multi-label classification (MLC) and (iii) label ranking (LR), mentioned in the order of increasing complexity of the output. While multi-class classification deals with predicting one class for every instance, multi-label classification deals with predicting more than one classes for every instance and label ranking deals with assigning a rank to each label for every instance. All the work in this field is associated around formulating new error functions that can force network to identify the output dependencies. Aim of our work is to adapt neural network to implicitly handle the feature extraction (dependencies) for output in the network structure, removing the need of hand crafted error functions. We show that the multimodal deep architectures can be adapted for these type of problems (or data) by considering labels as one of the modalities. This also brings unsupervised pretraining to the output along with the input. We show that these models can not only outperform standard deep neural networks, but also outperform standard adaptations of neural networks for individual domains under various metrics over several data sets considered by us. We can observe that the performance of our models over other models improves even more as the complexity of the output/ problem increases.

During neocortical development, network activity undergoes a dramatic transition from largely synchronized, so-called cluster activity, to a relatively sparse pattern around the time of eye-opening in rodents. Biophysical mechanisms underlying this sparsification phenomenon remain poorly understood. Here, we present a dynamic systems modeling study of a developing neural network that provides the first mechanistic insights into sparsification. We find that the rest state of immature networks is strongly affected by the dynamics of a transient, unstable state hidden in their firing activities, allowing these networks to either be silent or generate large cluster activity. We address how, and which, specific developmental changes in neuronal and synaptic parameters drive sparsification. We also reveal how these changes refine the information processing capabilities of an in vivo developing network, mainly by showing a developmental reduction in the instability of network’s firing activity, an effective availability of inhibition-stabilized states, and an emergence of spontaneous attractors and state transition mechanisms. Furthermore, we demonstrate the key role of GABAergic transmission and depressing glutamatergic synapses in governing the spatiotemporal evolution of cluster activity. These results, by providing a strong link between experimental observations and model behavior, suggest how adult sparse coding networks may emerge developmentally.

The thesis focuses on banking regulation and on the nexus between financial sovereign crises. After illustrating the main mechanisms on the recent financial crisis, we construct several multi-agent network models of a financial system for testing its stability under different parameters. In the first part, we focus on the rationale for banking regulation and we describe its development including the recently introduced Basel III measures. The main conclusion of this part is that regulation is to a large extent influenced by the banks and it does not always secure financial system stability. In the second part, we build an agent-based model which enables us to simulate the impacts of various types of negative shocks given various settings of the banking system and the regulatory environment, including the capital and liquidity measures. Our simulations show firstly that sufficient capital buffers are crucial for systemic stability, secondly that the discretionary measures have little effect once a crisis breaks out and thirdly that liquidity measures are a relevant regulatory tool. In the third part, the model is extended so that it allows for testing effects of state support on systemic stability is tested with various parameter settings in Monte Carlo simulations and for testing of feedback loops in which...

[pt] A produtividade de poços produtores de gás, que operam com pressões de fundo inferiores à pressão de orvalho, é afetada pelo aparecimento da saturação de líquido em seus entornos. Para entender esse fenômeno, conhecido como bloqueio por condensado, os simuladores em escala de poros são ferramentas úteis na investigação dos parâmetros que influenciam na quantidade e na distribuição da saturação de condensado, assim como seus efeitos na redução do fluxo de gás. Esse trabalho apresenta um modelo de rede de capilares composicional e isotérmico para o estudo do escoamento de gás retrógrado em meios porosos. Forças capilares e gravitacionais não foram consideradas. O escoamento monofásico é comutado para bifásico de padrão anular quando a pressão e a composição do fluido atingem um critério de estabilidade. O método de Newton é aplicado para resolver as equações de fluxo e consistência dos volumes e calcular o transporte de cada um dos componentes ao longo da rede. As propriedades do fluido e o comportamento do escoamento foram testadas contra os resultados de um simulador termodinâmico comercial e soluções analíticas, respectivamente. Após validação, o simulador foi utilizado para obter curvas de permeabilidade relativa gás-líquido através da despressurização de uma rede 2D e alguns resultados são discutidos. / [en] Gas well deliverability in retrograde gas reservoirs is affected by the appearance of liquid saturation around the wellbore when the bottom-hole pressure is below the dew point. Pore-scale simulators are used to model this phenomenon, known as condensate blockage, and to investigate parameters that ifluence the amount and the distribution of condensate saturation, as well as how it chokes the gas flow. Here, a fully-implicit isothermal compositional pore-scale network model is presented for retrograde gas flow in porous media. Capillary and gravitational forces are neglected. The model shifts from single-phase flow to annular flow regime when the pressure and the fluid composition reach a stability criteria. Newton s method is applied on flow and volume consistency equations to calculate the transport of each component through the network. Fluid properties and flow behavior were tested against a commercial thermodynamic simulator and analytical solutions respectively. After validation, the simulator was used to predict gas-liquid relative permeability from a depletion process in a 2D network and some results are discussed.

This thesis proposes innovative methods to extract information embedded in the frequency and time domain response of the transformer winding, and utilizes them to suggest solutions to a few tasks that have until now been thought difficult, if not impossible, to resolve. Pursuing this philosophy originated from the basic under- standing that the response of any physical system (behaving largely as a linear time invariant system) has embedded information that characterizes it completely. So, the prerequisite is to evolve ways to extract this information from measured responses. Once that is done, a variety of interesting applications can be envisaged. The two applications considered in this thesis are- •Investigate indirect measurement of the series capacitance of a transformer winding using the measured frequency or time domain response •Explore the possibility of increasing the physical resolution of the ladder network used to model a fully interleaved-disk winding In the former application, since direct measurement of series capacitance is impossible, alternatives based on indirect measurement were also not attempted. Similarly, in the latter application, the upper limit is known to be fixed by the number of distinctly observable peaks in the magnitude frequency response, so the question of increasing this limit was also never explored. Solutions to these tasks are proposed after a systematic analysis of frequency/time domain responses of a winding, initially modeled as a lumped parameter ladder network, to extract correlations that exist between them and winding parameters, and finally examine how these relations can be exploited together with the measured responses. Each of the five chapters is dedicated to describe the solution to one task. In each chapter, analytical formulation is presented first, followed by experimental results. Good agreement with the predicted results demonstrates its practicability. In final summary, indirect measurement of the series capacitance of a winding and en- hancing physical resolution of a ladder network model to represent a fully interleaved- disk winding was successfully demonstrated and they are the main contributions of this thesis.

Real world systems like biological, social, technological, infrastructural and many others can be modeled as networks. The field of network science aims to study these complex networks and understand their structure and dynamics. A common feature of networks across domains is the distribution of the degree of the nodes according to a power-law (scale invariance). As a consequence of this skewness, the high degree nodes dominate the properties of these networks. The rich-club phenomenon is observed when the high degree or the rich nodes of the network prefer to connect amongst themselves. In the first part, the thesis investigates the rich-club phenomenon in higher order neighborhoods of the network by providing an elegant quantification using a geodesic distance based approach. This quantification helped in identifying networks where the trend and intensity of the rich-club phenomenon is significantly different in higher order neighborhoods compared to the immediate neighbors. The thesis also proposes a quantification of the importance of the non-rich nodes in the communication structure of the rich nodes, and broadly classify networks into core-periphery or cellular. Further a lack of universality is noticed in the structure of the networks belonging to a particular domain. It has been observed in the previous literature that the rich club connectivity dominates assortativity, a measure quantifying the mixing patterns in complex networks. Thus, assortativity is biased. To overcome such drawbacks, in the second part of the thesis proposes a novel measure called regularity. The analytical bounds on regularity and formulation of regularity for different network models are provided. Along with this a measure to quantify the mixing patterns of the neighborhood of a node called local regularity is also defined. The analysis on real-world network based on local regularity and degree distribution shows presence of both type of network, uniformly and non-uniformly mixed across different regions. Further normalized regularity is proposed to quantify the extent of preferential mixing in networks discounting the effect of degree distribution.

Water distribution systems convey drinking water from treatment plant and make available to consumers’ taps. It consists of essential components like pipes, valves, pumps, tanks and reservoirs etc. The main concern in the working of a water distribution system is to assure customer demands under a choice of quantity and quality throughout the complete life span for the probable loading situations. However, in some cases, the existing infrastructure may not be adequate to meet the customer’s requirements. In such cases, system modeling plays an important role in proper management of water supply systems. In present scenario, modeling plays a significant task in appropriate execution of water distribution system. From the angle of taking management decisions valve throttling control and pumps speed control are very important. These operational problems can be addressed by manual control or by automatic control. The problem is the use of manual controls that slow down the effectiveness of the system. It reduces the efficiency of operation of valve or pump. To improve the efficiency of such water distribution systems, an automatic control based technology has been developed that links the operation of the variable speed pump control or valve throttling control. By employing an automatic control, the pump can adjust its speed at all times to meet the actual flow requirements of each load served. In case of real system design Simulink is the most widely used tool. Commercial software package Matlab/Simulink used for creation of WDS model. The goal was to produce a model that could numerically analyze the dynamic performance of a water distribution system. A Comparison of single platform methodology (Simulink based control) and double platform methodology (Matlab and EPANET based control) has been done. Nonlinear Dynamic Inversion (DI) Control system model is developed for WDS model in Matlab/Simulink environment. Controller gain parameters are the very important value in control prospective. If the controller gain parameters are chosen incorrectly, the controlled process input can be unstable, i.e. its output diverges, with or without oscillation Tuning is the adjustment of control parameters (gains) to the optimum values for the desired control response. There are several methods for tuning controller like manual tuning (Trial and error procedure), Ziegler-Nichols method, Output Constraint Tuning (OCT) etc. Establishment of a pump operational policy by which all the reservoirs can be fed simultaneously to meet their requirements without creating undue transients. Tune the gain of DI controllers by different tuning methods and evaluate the best tuning method on the basis of controller performance. Development of meaningful additional objective is search of lower bound pump speed on the basis of control time or settling time. To bring the pump speeds in feasible range, application of constraint in pumps speed is introduced. The magnitude of constraints can be found using Monte Carlo methods. Monte Carlo methods are frequently used in simulating physical and mathematical systems. This method may be the most commonly applied statistical method in engineering and science disciplines. Another benefit is providing increased confidence that a model is robust using Monte Carlo testing. Model development for generalized control system for water distribution network provides the simplification needed for the simulation of large systems. Model development is based on the study of symmetric and non symmetric small, irregular networks, as well as large, regular and open bifurcating water distribution system. The problem considered in this section is that of flow dynamics in simple to complex, regular network which bifurcates in the form of a branching tree. In addition the control application of the flow network is investigated using valves as the manipulated variables to control branch flow rates. Communication between the network hydraulics coming from EPANET and control algorithm develop on Matlab (Programming Language) can be generalized with the help of development of general purpose control algorithm model.

Dans ce travail, nous nous intéressons à la dynamique de croissance d'une poche de vapeur par vaporisation en milieu poreux, en relation avec l'analyse des transferts couplés de masse et de chaleur dans les mèches poreuses des boucles fluides diphasiques à pompage capillaire. Nous proposons un modèle pour les régimes transitoires, régimes encore très mal compris en dépit de leur grande importance pratique (phase de démarrage, variations de puissance, etc.). Une approche de type "réseau de pores" est adoptée et permet de prédire la distribution des phases à l'échelle de l'espace des pores. Dans une étape préliminaire, une étude spécifique de drainage (déplacement d'un fluide mouillant par un fluide non mouillant) par pressurisation du fluide envahisseur est abordée. Cette étape, nécessaire au développement et au test d'un algorithme de croissance de poche de gaz, permet de valider le modèle hydrodynamique quantitativement par une étude expérimentale dédiée. Il est mis en évidence le rôle des films liquides et de la compressibilité du gaz. Le modèle est ensuite complété par l'ajout des transferts thermiques et du changement de phase. Ici encore, une étude expérimentale dédiée est proposée, afin de valider l'outil numérique mis en place. Enfin, un ultime ajout au modèle permet de prendre en compte les phénomènes particuliers liés à l'imbibition (déplacement d'un fluide non mouillant par un fluide mouillant). Des résultats statistiques concernant la réponse dynamique d'une poche de vapeur à l'application d'une densité de puissance sont présentés, ainsi que certaines situations oscillantes dans la mèche poreuse. Nous finissons par discuter de l'influence du re-mouillage de la mèche poreuse, phénomène qui entraîne une hystérésis significative. / We study the dynamic of a vapour pocket growing by vaporisation in a porous medium, in relation with the analysis of coupled heat and mass transfers in the porous wick of loop heat pipes (LHP). We propose a model for transient modes, which are still poorly understood in spite of their importance (start-ups, power transitions, etc.). This work is based on a pore network approach enabling us to predict the phase distribution at the pore space scale. In a preliminary step, a study of drainage (displacement of a wetting fluid by a non wetting one) by pressurisation of the invading fluid is performed. This step is necessary for the development and the test of the vapour pocket growing algorithm. A quantitative validation of the hydro-dynamical model is obtained thanks to a dedicated experimental study. The influence of liquid films as well as gas compressibility is investigated. Our model is then improved to deal with heat transfer and phase change. Again, a dedicated experimental study is performed in order to validate the numerical tool. The model is finally improved a last time to deal with the effects due to imbibition mechanisms (displacement of a non wetting fluid by a wetting one). Statistical results concerning the dynamic response of a vapour pocket to the application of a power density are presented, andsome specific oscillating situations in the wick are identified. We finish discussing the influence of the re- etting of the porous wick, a phenomenon which induces a significant hysteresis effect.

During neocortical development, network activity undergoes a dramatic transition from largely synchronized, so-called cluster activity, to a relatively sparse pattern around the time of eye-opening in rodents. Biophysical mechanisms underlying this sparsification phenomenon remain poorly understood. Here, we present a dynamic systems modeling study of a developing neural network that provides the first mechanistic insights into sparsification. We find that the rest state of immature networks is strongly affected by the dynamics of a transient, unstable state hidden in their firing activities, allowing these networks to either be silent or generate large cluster activity. We address how, and which, specific developmental changes in neuronal and synaptic parameters drive sparsification. We also reveal how these changes refine the information processing capabilities of an in vivo developing network, mainly by showing a developmental reduction in the instability of network’s firing activity, an effective availability of inhibition-stabilized states, and an emergence of spontaneous attractors and state transition mechanisms. Furthermore, we demonstrate the key role of GABAergic transmission and depressing glutamatergic synapses in governing the spatiotemporal evolution of cluster activity. These results, by providing a strong link between experimental observations and model behavior, suggest how adult sparse coding networks may emerge developmentally.

[pt] A microtomografia computadorizada de uma amostra de rocha possibilita uma caracterização do meio poroso e pode ser utilizada para estimar propriedades da rocha em macroescala, isto é, em escala de reservatório. Métodos baseados em mapas de distâncias e em algoritmos de erosão são as principais abordagens utilizadas para extração de uma rede de poros e gargantas a partir de imagens microtomográficas de rocha. Este trabalho propõe um método híbrido para a construção da rede, de modo que, durante o processo de modelagem na escala de poros, obtemos um esqueleto do espaço poroso por meio de um algoritmo de erosão e utilizamos um mapa de distâncias para construir uma rede de poros e gargantas. A determinação dos poros e gargantas a partir do esqueleto adota uma abordagem não-determinística possibilitando a geração de múltiplas redes com configurações distintas a partir de um mesmo esqueleto. Avaliamos a variabilidade dos cenários gerados e comparamos as estimativas para as propriedades petrofísicas com as obtidas pelo método de Bolas Máximas por meio dos resultados de uma simulação de fluxo monofásica na rede. / [en] Computerized microtomography of a rock sample enables a characterization of the porous medium and can be used to estimate rock properties at the macro-scale, i.e., reservoir-scale. Methods based on distance maps and thinning algorithms are the main approaches used for extracting a pore and throats network from microtomographic rock images. This paper proposes a hybrid method for constructing the network. So that during the pore-scale modeling process, we obtain a skeleton of the pore space by using a thinning algorithm and a distance map to build a network of pores and throats. The determination of pores and throats from the skeleton assumes a non-deterministic approach enabling the generation of multiple networks with distinct configurations from the same skeleton. We evaluate the variability of the generated scenarios and compare the estimates for the petrophysical properties with those obtained by the Maximum Ball Method through the results of a single-phase flow simulation on the network.