Nesta tese, uma família de memórias associativas bidirecionais caóticas (família C-BAM) é proposta, implementada e testada com o objetivo de estender a relevância da presença e do estudo do fenômeno caótico a modelos de redes associativas. Na modelagem da família C-BAM, todos os neurônios da memória associativa bidirecional caótica (BAM), BAM com atraso e BAM exponencial (eBAM) foram substituídos por neurônios caóticos. Cada parâmetro do neurônio caótico na família C-BAM tem sua influência estimada através do planejamento de experimentos, em diferentes dinâmicas. Com base no planejamento de experimentos, valores de parâmetros são selecionados a fim de ilustrar a emergência de comportamentos dinâmicos como bifurcação, caos determinístico e crise. A existência de dinâmicas caóticas é confirmada pelo cálculo dos expoentes de Lyapunov. Experimentos empíricos mostraram que a dinâmica caótica modifica a acessibilidade à memória da família C-BAM. Ao invés de recuperar um único par, como a família BAM fazia, a versão caótica é capaz de gerar uma grande diversidade de padrões recuperados, envolvendo complexas transições entre os padrões armazenados, para algumas variações paramétricas. Tal comportamento permite à família C-BAM acessar padrões inacessíveis às redes BAMs originais. Além disso, a nova acessibilidade à memória, na qual seqüências de recuperação (com diferentes tamanhos) compostas de padrões treinados e não treinados têm emergido, pode ser usada para modelar a habilidade de um indivíduo saltar passos na solução de uma tarefa. Esta tese seleciona a rede C-BAM para ilustrar que a seqüência de recuperação da rede pode modelar a habilidade de um noviço ou a habilidade de um especialista executar uma tarefa. Embora a família C-BAM possa alcançar todos os padrões armazenados durante o comportamento caótico, ela não consegue convergir para um padrão específico. Duas estratégias de controle são propostas para permitir que as redes caóticas convirjam para a memória desejada: o método de controle por pinagem e um método de controle adaptativo. Conseqüentemente, os modelos C-BAM podem, de fato, realizar a hetero-associação de memórias antes inacessíveis, e a rede C-BAM pode estabilizar-se no estado final de uma tarefa, dado o primeiro estado / In this thesis, a family of bidirectional associative memories (C-BAM family) is proposed, implemented and tested to extend the study of chaotic phenomenon in associative models. In the C-BAM model, all the original neurons of bidirectional associative memory (BAM), BAM with delay and exponenetial BAM (eBAM) were substituted for chaotic neurons. Based on the experimental design, values of C-BAM family parameters are set to illustrate the emergence of a diversity of dynamic behavior, such as bifurcation, deterministic chaos and crisis. The existence of the chaotic dynamics is confirmed by calculation of Lyapunov exponents. Empiric experiments showed that the chaotic dynamics modifies the behavior of memory accessibility. Instead of recalling a single pair, as BAM did, its chaotic version yielded a wide diversity of recalled patterns, involving complex transitions via memorized patterns for some parametric variations. Hence, C-BAM family can access patterns that original BAM family cannot. Moreover, the new way of memory accessibility, in which several recall sequences (with distinct sizes) composed of trained and nontrained patterns have emerged, can be used to model the ability of skipping steps by an individual in a task solution. This thesis selected C-BAM network to illustrate that the retrieval sequence can model the ability of a novice or the ability of an expert to execute a task. There are also illustrated cases in which a novice recall can be transformed into an expert recall through parametric variation. Although C-BAM family can reach all stored patterns during the chaotic behavior, it can not converge towards a specific pattern, consequently a desired output is not produced. In this thesis, two control strategies are proposed in order to make the chaotic networks to converge towards the desired memory: the pinning control method and the adaptive control method. Consequently, the C-BAM models can effectively realize the correct heteroassociation to former non-accessible memories and the C-BAM network can quickly be stabilized in the final state of a task, given the first state

Estudamos nesta dissertação o tempo de recorrência em sistemas dinâmicos, concentrando-nos na estatística do tempo de retorno. Calculamos numericamente a distribuição de tempo de retorno a uma região específica do espaço de fases de sistemas caóticos e comparamos com a distribuição binomial, deduzida para um processo aleatório. Os principais resultados obtidos foram: surgimento do efeito que denominamos memória de curto alcance, típico de sistemas determinísticos e associado à distribuição das órbitas periódicas instáveis; a distribuição de tempo de retorno caracteriza as principais propriedades temporais no caso de sistemas intermitentes. As conexões do tempo de retorno com regimes de transporte anômalo foram apresentadas, ressaltando suas limitações. O tempo de retorno foi utilizado ainda para analisar séries temporais, obtidas tanto de um modelo de mistura de um contaminante escalar passivo, como experimentalmente no plasma confinado magnéticamente. No primeiro caso constatamos que os retornos da série temporal assemelham-se às recorrências no espaço de fases do sistema dinâmico responsável pela mistura do contaminante: o mapa padrão com fase aleatória. Constatamos o surgimento de caudas de lei de potência na distribuição de tempo de retorno e calculamos sua dependência com o aumento da não linearidade e da aleatoriedade do sistema. Destacamos o efeito de múltiplas caudas de lei de potência, ausente no caso das distribuições obtidas no espaço de fases. Às séries obtidas em Tokamaks aplicamos o modelo de cascata log-normal para explicar sua função densidade de probabilidade. A distribuição de tempo de retorno destas séries mostrou estar diretamente relacionada com a correlação de curto e longo alcance presente na série. / We study the recurrence time in dynamical systems. The statistics of the recurrence time to a specific region of the phase space of chaotic dynamical systems were obtained numerically and compared with the binomial-like distribution, deduced for a random process. The main results are: the presence of the so called short time memory effect, typical for deterministic systems and related to the distribution of the unstable periodic orbits; the return time distribution captures the main temporal properties of intermittent systems. The possible connections of the recurrence time statistics to the anomalous transport were presented, with special attention to their limitations. The return time statistics was applied to analyze time series obtained from an Hamiltonian model and from magnetically confined plasma. In the first case we noticed that the recurrences of the series were similar to the recurrences obtained in the phase space of the Hamiltonian dynamical system: the standard map with a random phase. We analyze the dependence of the power-law tails of the distributions with the non-linearity and with the randomness of the system. One effect that appears only in the time series case is the multiple power law tails. We apply the log-normal cascade model to explain the probability density function of the series obtained in Tokamaks. The recurrence time statistics of the series is closely related to the short and long time correlation present on the series.

Submitted by Cássia Santos (cassia.bcufg@gmail.com) on 2017-01-17T12:39:40Z No. of bitstreams: 3 Dissertação - Joaquim Orlando Parada (parte1) - 2013.pdf: 11591347 bytes, checksum: e970b2f0fffd5ccc2222bce05ea90d41 (MD5) Dissertação - Joaquim Orlando Parada (parte 2) - 2013.pdf: 18027973 bytes, checksum: 6bdbe04565ae04f1d810137fc59f37e2 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2017-01-18T10:31:58Z (GMT) No. of bitstreams: 3 Dissertação - Joaquim Orlando Parada (parte1) - 2013.pdf: 11591347 bytes, checksum: e970b2f0fffd5ccc2222bce05ea90d41 (MD5) Dissertação - Joaquim Orlando Parada (parte 2) - 2013.pdf: 18027973 bytes, checksum: 6bdbe04565ae04f1d810137fc59f37e2 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2017-01-18T10:31:58Z (GMT). No. of bitstreams: 3 Dissertação - Joaquim Orlando Parada (parte1) - 2013.pdf: 11591347 bytes, checksum: e970b2f0fffd5ccc2222bce05ea90d41 (MD5) Dissertação - Joaquim Orlando Parada (parte 2) - 2013.pdf: 18027973 bytes, checksum: 6bdbe04565ae04f1d810137fc59f37e2 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2013-06-21 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / In this work, the linear and nonlinear instability of pipes conveying static and pulsating fluid flow is analyzed. The dynamic equation of motion was derived for cantilevered and clamped-clamped pipes. For this purpose, the Euler Bernoulli beam theory and Hamilton’s principle were applied, resulting in a partial differential equation of second order in time. Thus, a model with four degrees of freedom, which satisfies the boundary condition, is used and, the Galekin method is applied to derive the set of coupled non linear ordinary equations of motion which are, in turn, solved by the fourth order Runge-Kutta method, and then some numerical results were obtained as Argand diagram, stability boudaries, time response, phase plane and, Poincaré section, through computational algorithms modeled in C++. These results revealed the importance of the nonlinear terms in the stability of the system, especially in the post-critical analysis, also revealed the existence of quasi-periodic motions, for the system subjected to a static flow and, chaotic motions for pulsating fluid flow / Nesta dissertação analisa-se a instabilidade linear e não linear de tubos com fluido interno em escoamento estático e pulsante. A equação de movimento dinâmico foi deduzida para tubos em balanço e biengastados. Para tanto, utilizou-se a teoria de vigas de Euler Bernoulli e o princípio variacional de Hamilton, resultado em uma equação diferencial parcial de segunda ordem no tempo. Tal equação foi discretizada, pelo método de Galerkin, em quatro equações diferenciais ordinárias, uma para cada grau de liberdade, em seguida transformadas em um conjunto de equações diferenciais de primeira ordem. Tais equações foram integradas pelo método de Runge-Kutta de quarta ordem e, posteriormente, foram obtidos alguns resultados numéricos como: diagrama de Argand, curvas de escape, diagrama de bifurcação, resposta no tempo, plano fase e, seção de Poincaré, através de algoritmos implementados computacionalmente na linguagem C++. Tais resultados revelaram a importância dos termos não lineares na estabilidade do sistema, especialmente na análise pós-crítica, revelaram também a existência de movimentos quase periódicos, para o sistema submetido a um fluxo estático e, caóticos para fluxo pulsante.

Redes neurais biológicas contêm bilhões de células (neurônios) agrupadas em regiões espacial e funcionalmente distintas. Elas também apresentam comportamentos complexos, tais como dinâmicas periódicas e caóticas. Na área da Inteligência Artificial, pesquisas mostram que Redes Neurais Caóticas, isto é, modelos de Redes Neurais Artificiais que operam com dinâmicas complexas, são mais eficientes do que modelos tradicionais no que diz respeito a evitar memórias espúrias. Inspirado pelo fato de que o córtex cerebral contém agrupamentos de células e motivado pela eficiência no uso de dinâmicas complexas, este projeto de pesquisa investiga o comportamento dinâmico de um modelo de Rede Neural Artificial Recorrente, como o de Hopfield, porém com a topologia sináptica reorganizada a ponto de originar agrupamentos de neurônios, tal como acontece em uma Rede Complexa quando esta apresenta uma estrutura de comunidades. O modelo de treinamento tradicional de Hopfield também é alterado para uma regra de aprendizado que posta os padrões em ciclos, gerando uma matriz de pesos assimétrica. Resultados indicam que o modelo proposto oscila entre comportamentos periódicos e caóticos, dependendo do grau de fragmentação das sinapses. Com baixo grau de fragmentação, a rede opera com dinâmica periódica, como consequência da regra de treinamento utilizada. Dinâmicas caóticas parecem surgir quando existe um alto grau de fragmentação. Mostra-se, também, que é possível obter caoticidade em uma topologia adequadamente modular, ou seja, como uma estrutura de comunidades válida. Desta forma, este projeto de pesquisa provê uma metodologia alternativa para se construir um modelo de Rede Neural Artificial que realiza tarefas de reconhecimento de padrões, explorando dinâmicas complexas por meio de uma estrutura de conexões que se mostra mais similar à topologia existente no cérebro / Biological neural networks contain billions of neurons divided in spatial and functional clusters to perform dierent tasks. It also operates with complex dynamics such as periodic and chaotic ones. It has been shown that Chaotic Neural Networks are more efficient than conventional recurrent neural networks in avoiding spurious memory. Inspired by the fact that the cerebral cortex has speficic groups of cells and motivated by the efficiency of complex behaviors, in this document we investigate the dynamics of a recurrent neural network, as the Hopfield one, but with neurons coupled in such a way to form a complex network community structure. Also, we generate an asymmetric weight matrix placing pattern cycles during learning. Our study shows that the network can operate with periodic and chaotic dynamics, depending on the degree of the connection\'s fragmentation. For low fragmentation degree, the network operates with periodic dynamic duo to the employed learning rule. Chaotic behavior seems to rise for a high fragmentation degree. We also show that the neural network can hold both chaotic dynamic and a high value of modularity measure at the same time, indicating an acceptable community structure. These findings provide an alternative way to design dynamical neural networks to perform pattern recognition tasks exploiting periodic and chaotic dynamics by using a more similar topology to the topology of the brain

在許多的研究和應用之中都需要預測的技巧。本論文中, 我們建構了一個 新的神經網路模式--動態徑向基底函數 (dynamical radial basis function) 網路 (DRBF網路) , 並且用這種模式的神經網路作為「函數近 似子」(function approximator) 去處理預測上的問題。另外我們也設計 幾種不同的學習演算法以測試DRBF網路的功能。 / The forecasting technique is important for many researches and applications. In this paper, we shall construct a new model of neural networks -- the dynamical radial basis function (DRBF) networks and use the DRBF networks as "function approximators" to solve some forecasting problems. Different learning algorithms are used to test the capability of DRBF networks.

Dans ce travail, les équations moyennes des mouvements planétaires sont calculées à un ordre élevé grâce à une programmation en calcul formel dédié. Le système résultant comprend plus de 150 000 termes polynomiaux et fournit une très bonne approximation de l'évolution à long terme du système solaire. Le système d'équations est développé à ordre 2 dans les masses et l'ordre 5 en excentricité et inclinaison. le système de degré 3 est intégré analytiquement au premier ordre ce qui fournit une solution de plus de 25 000 termes. Le problème des petits diviseurs séculaires est discuté et une liste de petits diviseurs de grande amplitude dans la solution est donnée, le terme principal étant lié à l'argument $ g_1-g_5 + (s_2-s_1) $ qui intervient dans l'excentricité de Mercure et de Jupiter, et dans l'inclinaison de Mercure et de Vénus. Il est démontré que la présence de ces diviseurs compromet grandement la construction d'une solution analytique, sans tenir compte du fait que la même solution de degré 5 comprend plus de 3 000 000 termes. Les équations séculaires sont ensuite intégrées numériquement d'une façon très efficace sur plus de 1 million d'années, avec un pas de 500 ans pour toutes les planètes, après avoir rajouté la contribution moyenne de la relativité générale et de la Lune. Par comparaison avec les éphémérides DE102 sur plus de 3000 ans, il est démontré que les équations séculaires représentent très bien l'évolution à long terme du système solaire.

With the rapid development of optical communications and the increasing amount of data exchanged, it has become utterly important to provide effective ar- chitectures to protect sensitive data. The use of chaotic optoelectronic devices has already demonstrated great potential in terms of additional computational security at the physical layer of the optical network. However, the determination of the security level and the lack of a multi-user framework are two hurdles which have prevented their deployment on a large scale. In this thesis, we propose to address these two issues. First, we investigate the security of a widely used chaotic generator, the external cavity semiconductor laser (ECSL). This is a time-delay system known for providing complex and high-dimensional chaos, but with a low level of security regarding the identification of its most critical parameter, the time delay. We perform a detailed analysis of the influence of the ECSL parameters to devise how higher levels of security can be achieved and provide a physical interpretation of their origin. Second, we devise new architectures to multiplex optical chaotic signals and realize multi-user communications at high bit rates. We propose two different approaches exploiting known chaotic optoelectronic devices. The first one uses mutually cou- pled ECSL and extends typical chaos-based encryption strategies, such as chaos-shift keying (CSK) and chaos modulation (CMo). The second one uses an electro-optical oscillator (EOO) with multiple delayed feedback loops and aims first at transpos- ing coded-division multiple access (CDMA) and then at developing novel strategies of encryption and decryption, when the time-delays of each feedback loop are time- dependent.

When statistical assumptions do not hold and coherent structures are present in spatially extended systems such as fluid flows, flame fronts and field theories, a dynamical description of turbulent phenomena becomes necessary. In the dynamical systems approach, theory of turbulence for a given system, with given boundary conditions, is given by (a) the geometry of its infinite-dimensional state space and (b) the associated measure, that is, the likelihood that asymptotic dynamics visits a given state space region. In this thesis this vision is pursued in the context of Kuramoto-Sivashinsky system, one of the simplest physically interesting spatially extended nonlinear systems. With periodic boundary conditions, continuous translational symmetry endows state space with additional structure that often dictates the type of observed solutions. At the same time, the notion of recurrence becomes relative: asymptotic dynamics visits the neighborhood of any equivalent, translated point, infinitely often. Identification of points related by the symmetry group action, termed symmetry reduction, although conceptually simple as the group action is linear, is hard to implement in practice, yet it leads to dramatic simplification of dynamics. Here we propose a scheme, based on the method of moving frames of Cartan, to efficiently project solutions of high-dimensional truncations of partial differential equations computed in the original space to a reduced state space. The procedure simplifies the visualization of high-dimensional flows and provides new insight into the role the unstable manifolds of equilibria and traveling waves play in organizing Kuramoto-Sivashinsky flow. This in turn elucidates the mechanism that creates unstable modulated traveling waves (periodic orbits in reduced space) that provide a skeleton of the dynamics. The compact description of dynamics thus achieved sets the stage for reduction of the dynamics to mappings between a set of Poincare sections.

Thesis (MPhil)--University of Stellenbosch, 2007. / ENGLISH ABSTRACT: Many variations of Systems Development Life Cycle models have evolved over the last fifty years of systems engineering and software science, yet not enough knowledge is available to better understand these as Complex Adaptive Systems by studying chaos and complexity theories. The primary application domain of the thesis is focused on the development of electronic hardware and software products. There is a great need for innovation to reach all corners of the development ecosystem; however a large cognitive distance exists between the concept of systematic product development and that of value creation. Instruments are needed to aid process agility, for defusing imminent problems as they mount, and for making effective decisions to sustain maximum productivity. Many of these objectives are neglected in systems development practices. As with so many management fads, it appears that no single one of these models lived up to all of the expectations and in many cases ended up being recipes for disaster. The statistics available on failed projects are concerning but has not stopped the scientific and engineering communities from trying over, and over again, to make progress. The goal of the thesis is therefore to identify the most viable model that supports the sustainability of systems development team performance. The research draws insights from extant literature, by applying a knowledge management theory based analysis on the various models with specific attention given to complexity theory. The dominant metric discovered is to measure the Value Velocity of a Systems Development Team. This metric is determined by two independent variables, being Value Created and Delivery Delay. Complex Adaptive Systems simply requires a guiding vision and a carefully selected set of generative rules for increasing and sustaining the Value Velocity. / AFRIKAANSE OPSOMMING: Menige variasies van stelselsontwikkelingsmodelle het ontwikkel oor die afgelope vyftig jaar in stelselsingenieurswese en sagtewarewetenskap, en steeds is daar nie genoegsame kennis beskikbaar om beter begrip te kry oor hoe hierdie stelsels as Komplekse Aanpassende Sisteme bestudeer kan word nie, ten einde die bestuur daarvan te verbeter. Die primêre toepassingsgebied in die tesis is gespits op die ontwikkeling van rekenaarhardeware en - sagteware. Die behoefte vir innovasie moet al die fasette van die ontwikkelingsekosisteem bereik. Die bewusheidsgaping tussen sistemiese produkontwikkeling en waardeskepping, is te wyd. Instumentasie word benodig om te help met ratsheid in prosesuitvoering, om dreigende probleme te ontlont, en effektief besluitneming toe te pas, en sodoende produktiwiteit op ‘n maksimum vlak te hou. Hierdie doelwitte word tot ’n meerdere mate in die huidige praktyk verontagsaam. Net soos somige bestuursadvies oneffektief is, blyk dit dat daar nog steeds geen stelselsmodelle is wat alle verwagtinge bevredig nie. In baie gevalle eindig die toepassing daarvan in waan en mislukking. Die statistiek beskikbaar op mislukte projekte is onrusbarend, tog het dit nie vooruitgang gekelder nie, en die behoefte na verbetering bestaan steeds. Die doelwit van die tesis is dus om die mees lewensvatbare model wat die voortbestaan van stelselsontwikkelingsgroepe sal kan handhaaf, uit te sonder. Die navorsing neem insigte uit hedendagse literatuur en is gebasseer op ’n analiese van verskeide kennisbestuursteorieё teenoor die bestaande stelselsontwikkelingsmodelle. Die fokus is meer spesifiek toegespits op kompleksiteitsteorie. Die hoofmaatstaaf is om die Waardesnelheid van ’n stelselsontwikkelingspan te bepaal. Hierdie maatstaaf word gepyl deur twee onafhanklike veranderlikes, naamlik die Waarde Geskep en die Afleweringsvertraging. Ten slotte, vereis Kompleks Aanpassende Sisteme slegs die aanwesigheid van 'n leidende visie tesame met 'n goeddeurdagte stel ontwikkelingsreëls, wat aanleiding sal gee tot die verhoging en behoud van die Waardesnelheid.

Thesis (MSc (Computer Science))-- University of Stellenbosch, 2001. / ENGLISH ABSTRACT: This thesis explores the use of neural networks for predicting difficult, real-world time series. We first establish and demonstrate methods for characterising, modelling and predicting well-known systems. The real-world system we explore is seismic event data obtained from a South African gold mine. We show that this data is chaotic. After preprocessing the raw data, we show that neural networks are able to predict seismic activity reasonably well. / AFRIKAANSE OPSOMMING: Hierdie tesis ondersoek die gebruik van neurale netwerke om komplekse, werklik bestaande tydreekse te voorspel. Ter aanvang noem en demonstreer ons metodes vir die karakterisering, modelering en voorspelling van bekende stelsels. Ons gaan dan voort en ondersoek seismiese gebeurlikheidsdata afkomstig van ’n Suid-Afrikaanse goudmyn. Ons wys dat die data chaoties van aard is. Nadat ons die rou data verwerk, wys ons dat neurale netwerke die tydreekse redelik goed kan voorspel. / Integrated Seismic Systems International