Made available in DSpace on 2014-06-11T19:22:31Z (GMT). No. of bitstreams: 0 Previous issue date: 2011-12-02Bitstream added on 2014-06-13T19:48:52Z : No. of bitstreams: 1 freitas_lps_me_ilha.pdf: 463364 bytes, checksum: 35c3a3450e5ec638595c65e3a7508c09 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Neste trabalho descreve-se o desenvolvimento de duas redes neurais que identificam e classificam dados da distribuição do peso corporal na região plantar de pessoas normais e hemiplégicas. Esses dados são experimentais e foram obtidos através da utilização de uma plataforma de força contendo 48 sensores. As arquiteturas utilizadas para esta aplicação foram as redes neurais MLP (Multilayer Perceptron) com o algoritmo retropropagação (backpropagation), e ARTMAP Nebulosa. A escolha de tais arquiteturas se deve ao treinamento (supervisionado) o qual associa de forma direta a distribuição de força plantar com os respectivos pacientes (normais e hemiplégicos). Ambas as arquiteturas, MLP e ARTMAP Nebulosa, conseguiram fazer a discriminação entre quase todas as pessoas normais e hemiplégicos. A rede neural ARTMAP Nebulosa possui a vantagem de efetuar a classificação de forma rápida e eficiente. Esta aplicação é importante nas áreas de Podologia, Posturologia e Podoposturologia, pois propicia ao profissional de saúde uma nova metodologia de diagnóstico / This work describes the development of two neural networks that identify and classify data distribution of plantar body weight of normal or hemiplegic individuals. The architectures used for this application were, respectively, MLP neural networks (Multilayer Perceptron) with backpropagation algorithm, and Fuzzy ARTMAP. The choice of such architectures was due to the training (supervised training) which directly associates the distribution of plantar force with the patients (normal or hemiplegic). The input data used for training and diagnosis of the neural networks were obtained from a force plate, with 48 sensors, containing measurements of the weight distribution on the plantar region (right and left) of normal or hemiplegic patients. Both architectures, MLP and Fuzzy ARTMAP, were able to discriminate almost all normal and hemiplegic patients. The Fuzzy ARTMAP neural network was more efficient than MLP neural network in the classification of the patients. This application is important in areas of Podiatry, Posturology and Podoposturology because it can help the health care professionals

Orientador: Carlos Roberto Minussi / Banca: Francisco Villarreal Alvarado / Banca: Nobuo Oki / Banca: Geraldo Roberto Martins da Costa / Banca: Mário Oleskovicz / Resumo: Nos dias atuais, principalmente pelo fato de alguns sistemas serem desregulamentados, o estudo dos problemas de análise, planejamento e operação de sistemas de energia elétrica é de extrema importância para o funcionamento do sistema. Para isso é necessário que se obtenha, com antecedência, o comportamento da carga elétrica com o propósito de garantir o fornecimento de energia aos consumidores de forma econômica, segura e contínua. Este trabalho propõe o desenvolvimento de redes neurais artificiais utilizadas para resolver o problema de previsão de cargas elétricas. Para tanto, inicialmente, propôs-se a introdução de melhorias na rede neural feedforward com treinamento realizado utilizando o algoritmo retropropagação. Neste caso, foi desenvolvida/implementada a adaptação dos parâmetros de inclinação e translação da função sigmóide (função de ativação da rede neural). A inclusão desta nova estrutura de redes neurais produziu melhores resultados, se comparado à rede neural retropropagação convencional. Essas arquiteturas proporcionam bons resultados, porém, são estruturas de redes neurais que possuem o problema de convergência. O problema de previsão de cargas elétricas a curto-prazo necessita de uma rede neural que forneça uma saída de forma rápida e eficaz. No intuito de solucionar os problemas encontrados com o algoritmo retropropagação foi desenvolvida/implementada uma rede neural baseada na arquitetura ART (Adaptive Rossonance Theory), denominada rede neural ART&ARTMAP nebulosa, aplicada ao problema de previsão de carga elétrica. Trata-se, por conseguinte, da principal contribuição desta tese. As redes neurais, baseadas na arquitetura ART, possuem duas características fundamentais que são de extrema importância para o desempenho da rede (estabilidade e plasticidade), que permite a implementação do treinamento de modo contínuo...(Resumo completo, clicar acesso eletrônico abaixo) / Abstract: Nowadays due to the deregulamentation it is very important to study the problems of analyzing, planning and operation of electric power systems. For a reliable operation it is necessary to know previously the behavior of the load to guarantee the energy providing to the users with security and continuity and in an economic way. This work proposes to develop artificial neural networks to solve the problem of electric load forecasting. First, it is introduced some improvements on the feedforward neural network, with the training effectuated with the backpropagation algorithm. The improvement was the adaptation of the inclination and translation parameters of the sigmoid function (activation function of the neural network). The inclusion of this new structure provides better results if compared to the conventional backpropagation algorithm. These architectures provide good results, although they are structures that have some convergence problems. The short term electric load forecasting problem needs a neural network that provide a fast and efficient output. To solve this problem a neural network based on the ART (Adaptive Ressonance Theory), called_ fuzzy ART&ARTMAP applied to the load-forecasting problem, was developed and implemented._This is one of the contributions of this work. Neural networks based on the ART architecture have two important characteristics for the network performance, which are stability and plasticity, allowing the continuous training. The fuzzy ART&ARTMAP neural network reduces the imprecision of the results by a mechanism that separates the binary and analogical data and processing them separately. This represents a quality and an improvement on the results (reduction of the processing time and better precision), if compared to the neural network with backpropagation training (often considered as a benchmark in precision by the specialized...(Complete abastract click electronic access below) / Doutor

Orientador: Marcelo Carvalho Minhoto Teixeira / Banca: Aparecido Augusto de Carvalho / Banca: Márcio Roberto Covacic / Resumo: Neste trabalho descreve-se o desenvolvimento de duas redes neurais que identificam e classificam dados da distribuição do peso corporal na região plantar de pessoas normais e hemiplégicas. Esses dados são experimentais e foram obtidos através da utilização de uma plataforma de força contendo 48 sensores. As arquiteturas utilizadas para esta aplicação foram as redes neurais MLP (Multilayer Perceptron) com o algoritmo retropropagação (backpropagation), e ARTMAP Nebulosa. A escolha de tais arquiteturas se deve ao treinamento (supervisionado) o qual associa de forma direta a distribuição de força plantar com os respectivos pacientes (normais e hemiplégicos). Ambas as arquiteturas, MLP e ARTMAP Nebulosa, conseguiram fazer a discriminação entre quase todas as pessoas normais e hemiplégicos. A rede neural ARTMAP Nebulosa possui a vantagem de efetuar a classificação de forma rápida e eficiente. Esta aplicação é importante nas áreas de Podologia, Posturologia e Podoposturologia, pois propicia ao profissional de saúde uma nova metodologia de diagnóstico / Abstract: This work describes the development of two neural networks that identify and classify data distribution of plantar body weight of normal or hemiplegic individuals. The architectures used for this application were, respectively, MLP neural networks (Multilayer Perceptron) with backpropagation algorithm, and Fuzzy ARTMAP. The choice of such architectures was due to the training (supervised training) which directly associates the distribution of plantar force with the patients (normal or hemiplegic). The input data used for training and diagnosis of the neural networks were obtained from a force plate, with 48 sensors, containing measurements of the weight distribution on the plantar region (right and left) of normal or hemiplegic patients. Both architectures, MLP and Fuzzy ARTMAP, were able to discriminate almost all normal and hemiplegic patients. The Fuzzy ARTMAP neural network was more efficient than MLP neural network in the classification of the patients. This application is important in areas of Podiatry, Posturology and Podoposturology because it can help the health care professionals / Mestre

Master of Science / Department of Electrical and Computer Engineering / Anil Pahwa / Electricity generation capacity from different renewable sources has been significantly growing worldwide in recent years, specially wind power. Fast dispatch of wind power provides flexibility for spinning reserve. However, wind is intermittent in nature. Thus, stable grid operations and energy management are becoming more challenging with the increasing penetration of wind in power systems. Efficient forecast methods can help the scenario. Many wind forecast models have been developed over the years. Highly effective models with the combination of numerical weather prediction and statistical models also exist at present. This study intends to develop a model to forecast hourly wind speed using an artificial neural network (ANN) approach for effective and fast operation with minimum data. The procedure is outlined in this work and the performance of the ANN model is compared with the persistence forecast model.

The effects of noise correlations on neuronal stimulus discrimination have been the subject of sustained debate. Both experimental and computational work suggest beneficial and detrimental contributions of noise correlations. The aim of this study is to develop an analytically tractable model of stimulus discrimination that reveals the conditions leading to improved or impaired performance from model parameters and levels of noise correlation. We begin with a mean firing rate integrator model as an approximation of underlying spiking activity in neuronal circuits. We consider two independent units receiving constant input and time fluctuating noise whose correlation across units can be tuned independently of firing rate. We implement a perceptron-like readout with Fisher Linear Discriminant Analysis (LDA). We exploit its closed form solution to find explicit expressions for discrimination error as a function of network parameters (leak, shared inputs, and noise gain) as well as the strength of noise correlation. First, we derive equations for discrimination error as a function of noise correlation. We find that four qualitatively different sets of results exist, based on the ratios of the difference of means and variance of the distributions of neural activity. From network parameters, we find the conditions for which an increase in noise correlation can lead to monotonic decrease or monotonic increase of error, as well as conditions for which error evolves non-monotonically as a function of correlations. These results provide a potential explanation for previously reported contradictory effects of noise correlation. Second, we expand on the dependency of the quantitative behaviour of the error curve on the tuning of specific subsets of network parameters. Particularly, when the noise gain of a pair of units is increased, the error rate as a function of noise correlation increases multiplicatively. However, when the noise gain of a single unit is increased, under certain conditions, the effect of noise can be beneficial to stimulus discrimination. In sum, we present a framework of analysis that explains a series of non-trivial properties of neuronal discrimination via a simple linear classifier. We show explicitly how different configurations of parameters can lead to drastically different conclusions on the impact of noise correlations. These effects shed light on abundant experimental and computational results reporting conflicting effects of noise correlations. The derived analyses rely on few assumptions and may therefore be applicable to a broad class of neural models whose activity can be approximated by a multivariate distribution.

A solution to distinguish ventricular fibrillation and ventricular flutter from other arrhythmias and from disturbances caused by body motion or muscle activity with the use of a neural network has been investigated. Ventricular fibrillation and ventricular flutter occurs when the cardiac muscle cells are not triggered by the cardiac conduction system, but rather by ectopic foci preventing a synchronized contraction of the cardiac muscle cells and therefore inhibiting the hearts capability to properly pump blood. Two different methods, gradient descent and quasi-Newton, used by the network for learning was tested and preprocessing methods used on the input data before introducing it to the network was evaluated. Gradient descent makes use of the gradient to the error function with regards to its weights and updates the network in the direction which the output error by the network decreases the most. Quasi-Newton update the network roughly in the Newton direction by iteratively build up an approximation to the Hessian of the error function with the use of information from the gradient. The preprocessing methods used were: Threshold Crossing Intervals (TCI) which looks at the time between baseline crossings of the ECG signal. Mean Absolute Value (MAV) which computes the mean absolute value of the normalized ECG signal. Spectral Analysis which takes into account different properties of the frequency spectrum of ventricular fibrillation and normal sinus rhythm. VF-filter which assumes VF to be sinusoidal and computes the leakage after the ECG signal has been bandstop filtered around the mean frequency. Period and Amplitude Information of the maximum amplitude of the input frequency spectrum and its period. It was found that the networks that used the preprocessed signal was a poor classifier for the arrhythmias partially because ventricular fibrillation was not easily separable from the arrhythmias by the implementaion of the preprocessed inputs given.

The way humans listen to music and perceive its structure isautomatic. In an attempt by Friberg et al. (2011) to model thishuman perception mechanism, a set of nine perceptual features wasselected to describe the overall properties of music. By letting atest group rate the perceptual features in a data set of musicalpieces, they discovered that the factor with most importance fordescribing the emotions happy and sad was the perceptual featuremodality. Modality in music denotes whether the key of a musicalpiece is in major or minor.This thesis aims to predict the modality in a continuous scale (0-10) from chord analysis with multiple linear regression and a NeuralNetwork (NN) in a computational model using a custom set offeatures. The model was able to predict the modality with anexplained variability of 64 % using a NN. The results clearlyindicated that the approach of using chords as features to predictmodality, is appropriate for music data sets that consisted of tonalmusic. / Computational Modelling of Perceptual Music Features

Research on modeling trees and plants has attracted a great deal of attention in recent years. Early procedural tree modeling can be divided into four main categories: rule-based algorithms, repetitive patterns, cellular automata, and particle systems. These methods offer a very high level of realism; however, creating millions of varied tree datasets manually is not logistically possible, even for professional 3D modeling artists. Trees created using these previous methods are typically static and the controllability of these procedural tree models is low. Deep generative models are capable of generating any type of shape automatically, making it possible to create 3D models at large scale. In this paper, we introduce a novel deep generative model that generates 3D (botanical) tree models, which are not only edible, but also have diverse shapes. Our proposed network, denoted BranchNet, trains the tree branch structures on a hierarchical Variational Autoencoder (VAE) that learns new generative model structures. By directly encoding shapes into a hierarchy graph, BranchNet can generate diverse, novel, and realistic tree structures. To assist the creation of tree models, we create a domain-specific language with a GUI for modeling 3D shape structures, in which the continuous parameters can be manually edited in order to produce new tree shapes. The trees are interpretable and the GUI can be edited to capture the subset of shape variability.

Ulcerative colitis (UC) is a chronic inflammatory disease characterized by periods of relapses and remissions affecting more than 500,000 people in the United States. To achieve the therapeutic goals of UC, which are to first induce and then maintain disease remission, doctors need to evaluate the severity of UC of a patient. However, it is very difficult to evaluate the severity of UC objectively because of non-uniform nature of symptoms and large variations in their patterns. To address this, in our previous works, we developed two different approaches in which one is using the image textures, and the other is using CNN (convolutional neural network) to measure and classify objectively the severity of UC presented in optical colonoscopy video frames. But, we found that the image texture based approach could not handle larger number of variations in their patterns, and the CNN based approach could not achieve very high accuracy. In this paper, we improve our CNN based approach in two ways to provide better accuracy for the classification. We add more thorough and essential preprocessing, and generate more classes to accommodate large variations in their patterns. The experimental results show that the proposed preprocessing can improve the overall accuracy of evaluating the severity of UC.

It is well known that the stock market is highly volatile, so stock price prediction is a very challenging task. However, in order to make a profit or to understand the equity market, many investors and researchers use various statistical, econometric, and neural network models to make the best stock price predictions possible. In this thesis the aim is to compare the predictability of two econometric models, the exponential moving average (EMA) and auto regressive integrated moving average (ARIMA) models, and two neural network models, a simple recurrent neural network (RNN) and the long short term memory model (LSTM) model. The comparison is primarily made using the Tesla company as the underlying stock. While using mean square error (MSE) as a measure of performance, the LSTM model consistently outperformed the other three models.