Ao passar por uma fina camada magnética uma corrente spin polarizada pode produzir um efeito de torque clássico atuando na camada, sendo capaz de gerar precessão e reversão da magnetização. Esse efeito tem sido alvo de inúmeras pesquisas, em especial pela possibilidade de aplicações em memórias magnéticas não voláteis e em nano-osciladores de alta frequência, entretanto outras características podem ser exploradas. Em particular, devido ao seu caráter não-linear, torques de spin aplicados em camadas magnéticas podem fornecer condições para aparecimento de caos determinístico e ressonância estocástica. Caos determinístico pode ocorrer em sistemas dinâmicos contínuos que tenham ao menos três graus de liberdade. Nesse caso, mesmo que apenas termos determinísticos sejam considerados, a combinação de termos não-lineares e alta sensibilidade em relação a condições iniciais ou pequenas perturbações pode gerar irregularidade e imprevisibilidade no sistema. Ressonância estocástica é o nome que se dá para fenômenos em que a adição de ruído a um sistema pode melhorar a resposta do mesmo, existindo um nível ótimo de ruído. Esse fenômeno pode ser usado para detecção e amplificação de sinais de baixa intensidade, por exemplo. Aqui analisamos a dinâmica da magnetização da camada livre de junções magnéticas em geometrias do tipo nanopilar, com o estudo dividido em dinâmicas determinísticas e estocásticas. Dentro da análise apenas com termos determinísticos, buscamos verificar comportamentos regulares, irregulares e caóticos, caracterizando o sistema através da geração de diagramas com as fases dinâmicas para diferentes valores de parâmetros. Foram vistas duas geometrias diferentes, sendo que em uma delas foi possível fazer a caracterização completa das fases dinâmicas do sistema. No caso de dinâmicas estocásticas, buscamos explorar efeitos não-lineares e flutuações térmicas, analisando ressonância estocástica e sincronização facilitada por ruído em uma junção túnel magnética, além de estudar as respostas dinâmicas quando há apenas o torque de Slonczewski e quando também está presente o torque tipo campo. Foi possível observar a influência de diversos parâmetros, como a amplitude da corrente aplicada e a frequência de entrada, na resposta magnética e na sincronização de dispositivos estocásticos. Além disso, vimos que com a inclusão do torque tipo campo aparece um possível novo comportamento, similar à ressonância, em alta frequência, ainda não detectado experimentalmente. Esses resultados são importantes pela possibilidade de uso desses dispositivos spintrônicos em transmissão segura de dados, comunicação em alta frequência e em uma nova geração de dispositivos bio-inspirados e eficientes energeticamente. / When passing through a fine magnetic layer a spin polarized electric current may result in a classical torque acting on the layer, being capable of causing magnetization precession and reversal. This effect has been object of numerous researches, specially because of possible applications in non-volatile magnetic memories and high frequency nanooscillators. However, other characteristics can be exploited. In particular, because of its non-linear features, spin torques acting on magnetic layers can generate the conditions for deterministic chaos and stochastic resonance to arise. Deterministic chaos may happen in continuous nonlinear dynamical systems with at least three degrees of freedom. In this case, even if only deterministic terms are considered, the combination of nonlinearities with high sensitivity on initial conditions or small perturbations can produce irregularity and unpredictability in the dynamical behaviour. Stochastic resonance is the phenomenon in which the addition of noise in a system can produce a better output, or system response, existing an optimal noise level. This effect can be used as a way to detect and amplify low intensity signal, for example. In this PhD Thesis we study the magnetization dynamics on the free layer of magnetic junctions in nanopillar geometries. The work is divided into two parts: deterministic and stochastic dynamics. When analysing the deterministic case we tried to characterize regular, irregular and chaotic behaviours, producing dynamical phases diagrams for different system parameters. Two different geometries were analysed, being possible to generate a complete characterization of the dynamical phases in one of them. For the stochastic case we tried to explore nonlinear effects and thermal fluctuations, analysing stochastic resonance and noise-enhanced synchronization in a magnetic tunnel junction and studying the dynamical response when only one spin torque is considered, the Slonczewski torque, and also when a perpendicular torque, the field-like torque, is present. We were able to see the influence of several system parameters, such as the amplitude of the applied electric current and the input frequency, on the system response and on the synchronization of stochastic systems. Also, we noticed that with the inclusion of the field-like torque a possibly new high frequency resonance-like behaviour appears. These results are important because of the possibility of using new spintronic devices for secure data transmission, high frequency communications and on a new generation of bio-inspired devices.

Submitted by Luanna Matias (lua_matias@yahoo.com.br) on 2015-02-04T20:56:59Z No. of bitstreams: 2 Dissertação - Augusta Finotti Brazão - 2014.pdf: 4325407 bytes, checksum: ed015d93a79ebdcbed577af5e0f9a797 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2015-02-05T09:48:34Z (GMT) No. of bitstreams: 2 Dissertação - Augusta Finotti Brazão - 2014.pdf: 4325407 bytes, checksum: ed015d93a79ebdcbed577af5e0f9a797 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-02-05T09:48:34Z (GMT). No. of bitstreams: 2 Dissertação - Augusta Finotti Brazão - 2014.pdf: 4325407 bytes, checksum: ed015d93a79ebdcbed577af5e0f9a797 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2014-04-04 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / The present study aims to investigate the influence of uncertainties in physical and geometric parameters to obtain the load parametric instability of cylindrical shell, using the Galerkin method with the stochastic polynomial Hermite-Caos. The nonlinear equations of motion of the cylindrical shell are deduced from their functional power considering the strain field proposed by Donnell´s nonlinear shallow shell theory. The uncertainties are considered as random parameters with probability density function known in the partial differential equation of motion of the cylindrical shell, which it becomes a stochastic partial differential equation due to the presence of randomness. First, the discretization of the stochastic problem is performed using the stochastic Galerkin method together with polynomial Hermite-Chaos, to transform the stochastic partial differential equation into a set of equivalent deterministic partial differential equations, which take into account the randomness of the system. Then, the discretization of the lateral field displacement is made by a perturbation procedure, indicating the nonlinear vibration modes which couple to the linear vibration mode. The set of partial differential equations is transformed into a deterministic system of equations deterministic ordinary second order in time. Uncertainty is considered in one of its parameters: the Young modulus, thickness and amplitude of initial geometric imperfection. Then we analyze the influence of randomness in two parameters simultaneously: the thickness and the Young modulus. Once obtained the system of ordinary differential equations deterministic containing the randomness of the parameters, the integration over discrete time system is made from the Runge- Kutta fourth order to obtain results as the time response, bifurcation diagrams and boundaries of instability which are compared with deterministic analysis, indicating that polynomial Hermite-Chaos is a good numerical tool for predicting the load parametric instability without the need to perform a process of sampling. / O presente trabalho tem como objetivo investigar a influência de incertezas nos parâmetros físicos e geométricos para a determinação da carga de instabilidade paramétrica da casca cilíndrica, utilizando o método de Galerkin Estocástico juntamente com o polinômio de Hermite-Caos. As equações não-lineares de movimento da casca cilíndrica são deduzidas a partir de seus funcionais de energia considerando o campo de deformações proposto pela teoria não linear de Donnell para cascas esbeltas. As incertezas são consideradas como parâmetros aleatórios com função de densidade de probabilidade conhecida na equação diferencial parcial de movimento da casca cilíndrica, que passa a ser uma equação diferencial parcial estocástica devido à presença da aleatoriedade. Primeiramente, faz-se a discretização do problema estocástico utilizando o método de Galerkin Estocástico juntamente com o polinômio de Hermite-Caos, para transformar a equação diferencial parcial estocástica em um conjunto de equações diferenciais parciais determinísticas equivalentes, que levem em consideração a aleatoriedade do sistema. Em seguida, apresenta-se a discretização do campo de deslocamentos laterais através do Método da Perturbação, indicando os modos não-lineares de vibração que se acoplam ao modo linear de vibração, para que o conjunto de equações diferenciais parciais determinísticas seja transformado em um sistema de equações ordinárias determinísticas de segunda ordem no tempo. A incerteza é considerada inicialmente em apenas um de seus parâmetros: no módulo de elasticidade, na espessura e na amplitude da imperfeição geométrica inicial. Em seguida, analisa-se a influência de aleatoriedades em dois parâmetros simultaneamente, sendo eles: a espessura e o módulo de elasticidade. Uma vez obtido o sistema de equações diferenciais ordinárias determinísticas que contêm as aleatoriedades dos parâmetros, a integração ao longo do tempo do sistema discretizado é feita a partir do método de Runge-Kutta de quarta ordem, obtendo-se resultados como resposta no tempo, diagramas de bifurcação e fronteiras de instabilidade, que são comparados com análises determinísticas, indicando que o polinômio de Hermite-Caos é uma boa ferramenta numérica para prever a carga de instabilidade paramétrica sem a necessidade de se realizar um processo de amostragens.

Deep learning, a sub-topic of machine learning inspired by biology, have achieved wide attention in the industry and research community recently. State-of-the-art applications in the area of computer vision and speech recognition (among others) are built using deep learning algorithms. In contrast to traditional algorithms, where the developer fully instructs the application what to do, deep learning algorithms instead learn from experience when performing a task. However, for the algorithm to learn require training, which is a high computational challenge. High Performance Computing can help ease the burden through parallelization, thereby reducing the training time; this is essential to fully utilize the algorithms in practice. Numerous work targeting GPUs have investigated ways to speed up the training, less attention have been paid to the Intel Xeon Phi coprocessor. In this thesis we present a parallelized implementation of a Convolutional Neural Network (CNN), a deep learning architecture, and our proposed parallelization scheme, CHAOS. Additionally a theoretical analysis and a performance model discuss the algorithm in detail and allow for predictions if even more threads are available in the future. The algorithm is evaluated on an Intel Xeon Phi 7120p, Xeon E5-2695v2 2.4 GHz and Core i5 661 3.33 GHz using various architectures and thread counts on the MNIST dataset. Findings show a 103.5x, 99.9x, 100.4x speed up for the large, medium, and small architecture respectively for 244 threads compared to 1 thread on the coprocessor. Moreover, a 10.9x - 14.1x (large to small) speed up compared to the sequential version running on Xeon E5. We managed to decrease training time from 7 days on the Core i5 and 31 hours on the Xeon E5, to 3 hours on the Intel Xeon Phi when training our large network for 15 epochs

For the last twenty years, several assumptions have been expressed in the fields of information processing, neurophysiology and cognitive sciences. First, neural networks and their dynamical behaviors in terms of attractors is the natural way adopted by the brain to encode information. Any information item to be stored in the neural network should be coded in some way or another in one of the dynamical attractors of the brain, and retrieved by stimulating the network to trap its dynamics in the desired item’s basin of attraction. The second view shared by neural network researchers is to base the learning of the synaptic matrix on a local Hebbian mechanism. The third assumption is the presence of chaos and the benefit gained by its presence. Chaos, although very simply produced, inherently possesses an infinite amount of cyclic regimes that can be exploited for coding information. Moreover, the network randomly wanders around these unstable regimes in a spontaneous way, thus rapidly proposing alternative responses to external stimuli, and being easily able to switch from one of these potential attractors to another in response to any incoming stimulus. Finally, since their introduction sixty years ago, cell assemblies have proved to be a powerful paradigm for brain information processing. After their introduction in artificial intelligence, cell assemblies became commonly used in computational neuroscience as a neural substrate for content addressable memories. <p> <p>Based on these assumptions, this thesis provides a computer model of neural network simulation of a brain-like memory. It first shows experimentally that the more information is to be stored in robust cyclic attractors, the more chaos appears as a regime in the background, erratically itinerating among brief appearances of these attractors. Chaos does not appear to be the cause, but the consequence of the learning. However, it appears as an helpful consequence that widens the network’s encoding capacity. To learn the information to be stored, two supervised iterative Hebbian learning algorithm are proposed. One leaves the semantics of the attractors to be associated with the feeding data unprescribed, while the other defines it a priori. Both algorithms show good results, even though the first one is more robust and has a greater storing capacity. Using these promising results, a biologically plausible alternative to these algorithms is proposed using cell assemblies as substrate for information. Even though this is not new, the mechanisms underlying their formation are poorly understood and, so far, there are no biologically plausible algorithms that can explain how external stimuli can be online stored in cell assemblies. This thesis provide such a solution combining a fast Hebbian/anti-Hebbian learning of the network's recurrent connections for the creation of new cell assemblies, and a slower feedback signal which stabilizes the cell assemblies by learning the feed forward input connections. This last mechanism is inspired by the retroaxonal hypothesis. <p> / Doctorat en Sciences / info:eu-repo/semantics/nonPublished

A numerical code for solving the time-dependent incompressible 3D Navier-Stokes equations with finite volumes on overlapping staggered grids in cylindrical and rectangular geometry is developed. In the code, written in FORTRAN, the momentum equation for the velocity is solved by projection method and Poisson equation for the pressure is solved by ADI implicit method in two directions combined with discrete fast Fourier transform in the third direction. A special technique for overcoming the singularity on the cylinder's axis is developed. This code, taking into account dependence upon temperature of the viscosity, density and surface tension of the liquid, is used to study the fluid motion in a cylinder with free cylindrical surface (under normal and zero-gravity conditions); and in a rectangular closed cell with a source of thermocapillary convection (bubble inside attached to one of the cell's faces). They are significant problems in crystal growth and in general experiments in fluid dynamics respectively. Nevertheless, the main study is dedicated to the liquid bridge problem.<p><p>The development of thermocapillary convection inside a cylindrical liquid bridge is investigated by using a direct numerical simulation of the 3D, time-dependent problem for a wide range of Prandtl numbers, Pr = 0.01 - 108. For Pr > 0.08 (e.g. silicon oils), above the critical value of temperature difference between the supporting disks, two counter propagating hydrothermal waves bifurcate from the 2D steady state. The existence of standing and traveling waves is discussed. The dependence of viscosity upon temperature is taken into account. For Pr = 4, 0-g conditions, and for Pr = 18.8, 1-g case with unit aspect ratio an investigation of the onset of chaos was numerically carried out. <p><p>For a Pr = 108 liquid bridge under terrestrial conditions ,the appearance and the development of thermoconvective oscillatory flows were investigated for different ambient conditions around the free surface.<p><p>Transition from 2D thermoconvective steady flow to a 3D flow is considered for low-Prandtl fluids (Pr = 0.01) in a liquid bridge with a non-cylindrical free surface. For Pr < 0.08 (e.g. liquid metals), in supercritical region of parameters 3D but non-oscillatory convective flow is observed. The computer program developed for this simulation transforms the original non-rectangular physical domain into a rectangular computational domain.<p><p>A study of how presence of a bubble in experimental rectangular cell influences the convective flow when carrying out microgravity experiments. As a model, a real experiment called TRAMP is numerically simulated. The obtained results were very different from what was expected. First, because of residual gravity taking place on board any spacecraft; second, due to presence of a bubble having appeared on the experimental cell's wall. Real data obtained from experimental observations were taken for the calculations.<p> / Doctorat en sciences appliquées / info:eu-repo/semantics/nonPublished

Cílem této práce je se seznámení s možnostmi programu Matlab z hlediska detailní analýzy deterministických dynamických systémů. Jedná se především o analýzu časové posloupnosti a o nalezení Lyapunových exponentů. Dalším cílem je navrhnout algoritmus umožňující specifikovat chování systému na základě znalosti příslušných diferenciálních rovnic. To znamená, nalezení chaotických systémů.

The aim of this thesis is to design a universal analog oscillator using transconductance amplifiers. For studying behaviour of chaotic dynamical systems can be used systems Class C. Suitable way for the purpose modeling dynamic phenomena arising in these systems is an electronic circuit that exhibits the same behavior as modeled system. After familiarization with the basic principles of synthesis of integrators systems, and studying the involvement of frequently used functional blocks were designed the concept of universal chaotic oscillator using transconductance amplifiers. The functionality of this circuit has been verified by PSpice simulation program. A typical feature of chaotic oscillator is extremely sensitivity to initial conditions. Each small change on the initial parameters can lead to major change in the shape of the attractor. The result of this thesis is a functional sample of a universal chaotic oscillator, which was verified by the dynamic behavior of the given differential equations.

The thesis deals with the design and optimization of artificial neural networks (specifically nonlinear autoregressive networks) and their subsequent usage in predictive application of stock market time series.

This thesis deals with a complex analysis and prediction of foreign exchange markets. It uses advanced artificial intelligence methods, namely neural networks and chaos theory. It introduces unconventional approaches and methods of each of these areas, compares them and uses on a real problem. The core of this thesis is a comparison of several prediction models based on completely different principles and underlying theories. The outcome is then a selection of the most appropriate prediction model called NAR + H. The model is evaluated according to several criteria, the pros and cons are discussed and approximate expected profitability and risk are calculated. All analytical, prediction and partial algorithms are implemented in Matlab development environment and form a unified library of all used functions and scripts. It also may be considered as a secondary main outcome of the thesis.

Classical partial transport barriers govern both classical and quantum dynamics of generic Hamiltonian systems. Chaotic eigenstates of quantum systems are known to localize on either side of a partial barrier if the flux connecting the two sides is not resolved by means of Heisenberg's uncertainty. Surprisingly, in open systems, in which orbits can escape, chaotic resonance states exhibit such a localization even if the flux across the partial barrier is quantum mechanically resolved. We explain this using the concept of conditionally invariant measures by introducing a new quantum mechanically relevant class of such fractal measures. We numerically find quantum-to-classical correspondence for localization transitions depending on the openness of the system and on the decay rate of resonance states. Moreover, we show that the number of long-lived chaotic resonance states that localize on one particular side of the partial barrier is described by an individual fractal Weyl law. For a generic phase space, this implies a hierarchy of fractal Weyl laws, one for each region of the hierarchical decomposition of phase space.