This thesis focuses on the problem of "autonomous agents". It is assumed that such agents want to be in a desired state which can be assessed by the agent itself when it observes the consequences of its own actions. Therefore the feedback from the motor output via the environment to the sensor input is an essential component of such a system. As a consequence an agent is defined in this thesis as a self-referential system which operates within a closed sensor- mot or-sensor feedback loop. The generic situation is that the agent is always prone to unpredictable disturbances which arrive from the outside, i.e. from its environment. These disturbances cause a deviation from the desired state (for example the organism is attacked unexpectedly or the temperature in the environment changes, ...). The simplest mechanism for managing such disturbances in an organism is to employ a reflex loop which essentially establishes reactive behaviour. Reflex loops are directly related to closed loop feedback controllers. Thus, they are robust and they do not need a built-in model of the control situation. However, reflexes have one main disadvantage, namely that they always occur 'too late'; i.e., only after a (for example, unpleasant) reflex eliciting sensor event has occurred. This defines an objective problem for the organism. This thesis provides a solution to this problem which is called Isotropic Sequence Order (ISO-) learning. The problem is solved by correlating the primary reflex and a predictive sensor input: the result is that the system learns the temporal relation between the primary reflex and the earlier sensor input and creates a new predictive reflex. This (new) predictive reflex does not have the disadvantage of the primary reflex, namely of always being too late. As a consequence the agent is able to maintain its desired input-state all the time. In terms of engineering this means that ISO learning solves the inverse controller problem for the reflex, which is mathematically proven in this thesis. Summarising, this means that the organism starts as a reactive system and learning turns the system into a pro-active system. It will be demonstrated by a real robot experiment that ISO learning can successfully learn to solve the classical obstacle avoidance task without external intervention (like rewards). In this experiment the robot has to correlate a reflex (retraction after collision) with signals of range finders (turn before the collision). After successful learning the robot generates a turning reaction before it bumps into an obstacle. Additionally it will be shown that the learning goal of 'reflex avoidance' can also, paradoxically, be used to solve an attraction task.

The interaction between coding and learning rules in neural nets (NNs), and between coding and genetic operators in genetic algorithms (GAs) is discussed. The underlying principle advocated is that similar things in "the world" should have similar codes. Similarity metrics are suggested for the coding of images and numerical quantities in neural nets, and for the coding of neural network structures in genetic algorithms. A principal component analysis of natural images yields receptive fields resembling horizontal and vertical edge and bar detectors. The orientation sensitivity of the "bar detector" components is found to match a psychophysical model, suggesting that the brain may make some use of principal components in its visual processing. Experiments are reported on the effects of different input and output codings on the accuracy of neural nets handling numeric data. It is found that simple analogue and interpolation codes are most successful. Experiments on the coding of image data demonstrate the sensitivity of final performance to the internal structure of the net. The interaction between the coding of the target problem and reproduction operators of mutation and recombination in GAs are discussed and illustrated. The possibilities for using GAs to adapt aspects of NNs are considered. The permutation problem, which affects attempts to use GAs both to train net weights and adapt net structures, is illustrated and methods to reduce it suggested. Empirical tests using a simulated net design problem to reduce evaluation times indicate that the permutation problem may not be as severe as has been thought, but suggest the utility of a sorting recombination operator, that matches hidden units according to the number of connections they have in common. A number of experiments using GAs to design network structures are reported, both to specify a net to be trained from random weights, and to prune a pre-trained net. Three different coding methods are tried, and various sorting recombination operators evaluated. The results indicate that appropriate sorting can be beneficial, but the effects are problem-dependent. It is shown that the GA tends to overfit the net to the particular set of test criteria, to the possible detriment of wider generalisation ability. A method of testing the ability of a GA to make progress in the presence of noise, by adding a penalty flag, is described.

Orientador: Fernando Antônio Campos Gomide / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-22T00:50:20Z (GMT). No. of bitstreams: 1 Bordignon_FernandoLuis_M.pdf: 1666872 bytes, checksum: 4d838dfb4ec418698d9ecd3b74e7c981 (MD5) Previous issue date: 2013 / Resumo: Sistemas evolutivos são sistemas com alto nível de adaptação capazes de modificar simultaneamente suas estruturas e parâmetros a partir de um fluxo de dados, recursivamente. Aprendizagem a partir de fluxos de dados é um problema contemporâneo e difícil devido à taxa de aumento da dimensão, tamanho e disponibilidade temporal de dados, criando dificuldades para métodos tradicionais de aprendizado. Esta dissertação, além de apresentar uma revisão da literatura de sistemas evolutivos e redes neurais fuzzy, aborda uma estrutura e introduz um método de aprendizagem evolutivo para treinar redes neurais híbridas baseadas em uninormas, usando conceitos de aprendizado extremo. Neurônios baseados em uninormas fundamentados nas normas e conormas triangulares generalizam neurônios fuzzy. Uninormas trazem flexibilidade e generalidade a modelos neurais fuzzy, pois elas podem se comportar como normas triangulares, conormas triangulares, ou de forma intermediária por meio do ajuste de elementos identidade. Este recurso adiciona uma forma de plasticidade em modelos de redes neurais. Um método de agrupamento recursivo para granularizar o espaço de entrada e um esquema baseado no aprendizado extremo compõem um algoritmo para treinar a rede neural. _E provado que uma versão estática da rede neural fuzzy baseada em uninormas aproxima funções contínuas em domínios compactos, ou seja, _e um aproximador universal. Postula-se, e experimentos computacionais endossam, que a rede neural fuzzy evolutiva compartilha capacidade de aproximação equivalente, ou melhor, em ambientes dinâmicos, do que as suas equivalentes estáticas / Abstract: Evolving systems are highly adaptive systems able to simultaneously modify their structures and parameters from a stream of data, online. Learning from data streams is a contemporary and challenging issue due to the increasing rate of the size and temporal availability of data, turning the application of traditional learning methods limited. This dissertation, in addition to reviewing the literature of evolving systems and neuro fuzzy networks, addresses a structure and introduces an evolving learning approach to train uninorm-based hybrid neural networks using extreme learning concepts. Uninorm-based neurons, rooted in triangular norms and conorms, generalize fuzzy neurons. Uninorms bring flexibility and generality to fuzzy neuron models as they can behave like triangular norms, triangular conorms, or in between by adjusting identity elements. This feature adds a form of plasticity in neural network modeling. An incremental clustering method is used to granulate the input space, and a scheme based on extreme learning is developed to train the neural network. It is proved that a static version of the uninorm-based neuro fuzzy network approximate continuous functions in compact domains, i.e. it is a universal approximator. It is postulated and computational experiments endorse, that the evolving neuro fuzzy network share equivalent or better approximation capability in dynamic environments than their static counterparts / Mestrado / Engenharia de Computação / Mestre em Engenharia Elétrica

Submitted by Maria Suzana Diniz (msuzanad@hotmail.com) on 2015-11-05T12:39:09Z No. of bitstreams: 1 arquivototal.pdf: 6708796 bytes, checksum: 6e76285c41e0c7cce86179c206fac0db (MD5) / Made available in DSpace on 2015-11-05T12:39:09Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 6708796 bytes, checksum: 6e76285c41e0c7cce86179c206fac0db (MD5) Previous issue date: 2015-02-26 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Cerebrovascular Accident (CVA) is a disease characterized by an interruption of blood flow to the encephalon, which represents the leading cause of long-term disability and functional impairment in adult population. Therefore, the individual who had suffered CVA needs to access health services that offer rehabilitation assistance as they promote a better physical, functional, and mental capacity, helping the reinsertion and reintegration of this individual into society. Thus, this study aims to develop a decision-making model to determine the access to physiotherapy services for rehabilitation of patients who had suffered acute CVA in the cities of João Pessoa and Cabedelo. This is an observational-longitudinal study among man and women who were admitted at a public hospital in João Pessoa and live in its metro area, who had presented CVA as primary cause of hospitalization. A questionnaire was used containing items related to socioeconomic, demographic, and clinical data from this person, such as general health conditions, risk factors, functionality evaluation, and access to physiotherapy services. Interviews were conducted between 15 and 21 days after hospital discharge (T0) and between 90 and 105 days after the first interview (T1). There was a homogeneous distribution of sexes, group age over 60 years (mean age= 61.6 years; standard deviation= 15.7 years). Most of the subjects (69.2%) have had a ischemic CVA, which the right side was more affected (46.2%) and 89.7% have had up to two CVA episodes. From interviewed patients, 69.2% have not had access to physiotherapy services after three months from the first interview. For utilization of decision model, 16 variables were selected helped by WEKA software, generating a feedfoward Artificial Neural Network model composed by 16 neurons in the input layer, followed by two hidden layers with two hidden neurons in each layer and an output layer with 2 neurons with backpropagation learning. This decision model allowed classifying correctly almost all subjects that accessed or not the physiotherapy services, achieving 97.4% of successes, representing a greater reliability. Therefore, this model is constituted as an important tool in the visibility of the problem, helping in the decision-making process, planning, and reorganization of public health system and its several attention levels. / O Acidente Vascular Encefálico (AVE) é uma doença causada pela interrupção no suprimento sanguíneo ao encéfalo, representando a primeira causa de incapacidade prolongada e o comprometimento funcional em adultos. Assim, o indivíduo com AVE necessita acessar os serviços de saúde que oferecem assistência de reabilitação, pois promovem uma melhora na capacidade física, funcional e/ou mental, proporcionando a reinserção e a reintegração à sociedade. Portanto, o objetivo deste estudo foi elaborar um modelo de tomada de decisão para averiguar o acesso aos serviços de fisioterapia para reabilitação de pacientes com AVE agudo dos municípios de João Pessoa e Cabedelo. Trata-se de um estudo longitudinal observacional com indivíduos de ambos os sexos, admitidos em um hospital público de João Pessoa/PB e residentes na região metropolitana de João Pessoa, que apresentaram como causa primária da internação o AVE. Para tanto, foi utilizado um questionário contendo itens referentes aos dados socioeconômicos, demográficos e clínicos do sujeito, condições gerais de saúde, fatores de risco, avaliação da funcionalidade e do acesso aos serviços de fisioterapia. As entrevistas foram realizadas entre 15 e 21 dias após a alta hospitalar (T0) e entre 90 e 105 dias após a realização da primeira entrevista (T1). Verificou-se uma distribuição homogênea dos sexos, com faixa etária acima de 60 anos (média de idade=61,6 anos, dp=15,7). A maioria dos sujeitos (69,2%) tiveram um AVE do tipo isquêmico, sendo o lado direito mais afetado (46,2%) e 89,7% tiveram até dois episódios de AVE. Dos pacientes entrevistados, 69,2% não tiveram acesso aos serviços de fisioterapia após três meses da primeira entrevista. Para a utilização do modelo de decisão, selecionou-se 16 variáveis com auxílio do software WEKA, gerando um modelo de Redes Neurais Artificiais do tipo feedforward composta por 16 neurônios na camada de entrada, seguido por duas camadas ocultas com dois neurônios ocultos em cada e uma camada de saída com 2 neurônios com aprendizagem por backpropagation. Este modelo de decisão permitiu classificar corretamente quase todos os sujeitos que acessaram ou não os serviços de fisioterapia, obtendo 97,4% de acertos, representando uma maior confiabilidade. Portanto, este modelo constitui-se como uma ferramenta importante na visibilidade do problema, auxiliando no processo de tomada de decisão, no planejamento e na reorganização da rede de saúde em seus diversos níveis de atenção.