Neural networks are an attempt to build computer networks called artificial neurons, which imitate the activities of the human brain. Its origin dates back to 1943 when neurophysiologist Warren Me Cello and logician Walter Pits produced the first artificial neuron. Since then there has been tremendous development of neural networks and their applications to pattern and optical character recognition, speech processing, time series prediction, image processing and scattered data approximation. Since it has been shown that neural nets can approximate all but pathological functions, Neil Cotter considered neural network architecture based on Stone-Weierstrass Theorem. Using exponential functions, polynomials, rational functions and Boolean functions one can follow the method given by Cotter to obtain neural networks, which can approximate bounded measurable functions. Another problem of current research in computer graphics is to construct curves and surfaces from scattered spatial points by using B-Splines and NURBS or Bezier surfaces. Hoffman and Varady used Kohonen neural networks to construct appropriate grids. This thesis is concerned with two types of neural networks viz. those which satisfy the conditions of the Stone-Weierstrass theorem and Kohonen neural networks. We have used self-organizing maps for scattered data approximation. Neural network Tool Box from MATLAB is used to develop the required grids for approximating scattered data in one and two dimensions.

[pt] Esta tese apresenta os principais resultados de medidas de eficiência dos custos operacionais de 60 distribuidoras brasileiras de energia elétrica. Baseado no esquema yardstick competition, foi utilizado uma Rede Neural d e Kohonen (KNN) para identificar grupos de empresas similares. Os resultados obtidos pela KNN não são determinísticos, visto que os pesos sinápticos da rede são inicializados aleatoriamente. Então, é realizada uma simulação de Monte Carlo para encontrar os clusters mais frequentes. As medidas foram obtidas por modelos DEA (input oriented, com e sem restrições aos pesos) e modelos Bayesianos e frequencistas de fronteira estocástica (utilizando as funções Cobb-Douglas e Translog). Em todos os modelos, DEA e SFA, a única variável input refere-se ao custo operacional (OPEX). Os índices de eficiência destes modelos representam a potencial redução destes custos de acordo com cada concessionária avaliada. Os outputs são os cost drivers da variável OPEX: número de unidades consumidoras (uma proxy da quantidade de serviço), montante de energia distribuída (uma proxy do produto total) e a extensão da rede de distribuição (uma proxy da dispersão dos consumidores na área de concessão). Finalmente, vale registrar que estas técnicas podem mitigar a assimetria de informação e aprimorar a habilidade do agente regulador em comparar os desempenhos das distribuidoras em ambientes de regulação incentivada. / [en] This thesis presents the main results of the cost efficiency scores of 60 Brazilian electricity distribution utilities. Based on yardstick competition scheme, it was applied a Kohonen Neural Networks (KNN) to identify and to group the similar utilities. The KNN results are not deterministic, since the estimated weights are randomly initialized. Thus, a Monte Carlo simulation was used in order to find the most frequent clusters. Therefore was examined the use of the DEA methodology (input oriented, with and without weight constraints) and Bayesian and non- Bayesian Stochastic Frontier Analysis (centered on a Cobb- Douglas and Translog cost functions) to evaluate the cost efficiency scores of electricity distribution utilities. In both models the only input variable is operational cost (OPEX). The efficiency measures from these models reflect the potential of the reduction of operational costs of each utility. The outputs are the cost-drivers of the OPEX: the number of customers (a proxy for the amount of service), the total electric power supplied (a proxy for the amount of product delivered) and the distribution network size (a proxy of the customers scattering in the operating territory of each distribution utility). Finally, it is important to mention that these techniques can reduce the information assimetry to improve the regulator´s skill to compare the performance of the utilities in incentive regulation environments.

This thesis discusses the design and the utilization of the artificial neural networks as ECG classifiers and the detectors of heart diseases in ECG signal especially myocardial ischaemia. The changes of ST-T complexes are the important indicator of ischaemia in ECG signal. Different types of ischaemia are expressed particularly by depression or elevation of ST segments and changes of T wave. The first part of this thesis is orientated towards the theoretical knowledges and describes changes in the ECG signal rising close to different types of ischaemia. The second part deals with to the ECG signal pre-processing for the classification by neural network, filtration, QRS detection, ST-T detection, principal component analysis. In the last part there is described design of detector of myocardial ischaemia based on artificial neural networks with utilisation of two types of neural networks back – propagation and self-organizing map and the results of used algorithms. The appendix contains detailed description of each neural networks, description of the programme for classification of ECG signals by ANN and description of functions of programme. The programme was developed in Matlab R2007b.

O uso de dados biométricos da face para verificação automática de identidade é um dos maiores desafios em sistemas de controle de acesso seguro. O processo é extremamente complexo e influenciado por muitos fatores relacionados à forma, posição, iluminação, rotação, translação, disfarce e oclusão de características faciais. Hoje existem muitas técnicas para se reconhecer uma face. Esse trabalho apresenta uma investigação buscando identificar uma face no banco de dados ORL com diferentes grupos de treinamento. É proposto um algoritmo para o reconhecimento de faces baseado na técnica de subespaço LDA (PCA + LDA) utilizando uma rede neural SOM para representar cada classe (face) na etapa de classificação/identificação. Aplicando o método do subespaço LDA busca-se extrair as características mais importantes na identificação das faces previamente conhecidas e presentes no banco de dados, criando um espaço dimensional menor e discriminante com relação ao espaço original. As redes SOM são responsáveis pela memorização das características de cada classe. O algoritmo oferece maior desempenho (taxas de reconhecimento entre 97% e 98%) com relação às adversidades e fontes de erros que prejudicam os métodos de reconhecimento de faces tradicionais. / The use of biometric technique for automatic personal identification is one of the biggest challenges in the security field. The process is complex because it is influenced by many factors related to the form, position, illumination, rotation, translation, disguise and occlusion of face characteristics. Now a days, there are many face recognition techniques. This work presents a methodology for searching a face in the ORL database with some different training sets. The algorithm for face recognition was based on sub-space LDA (PCA + LDA) technique using a SOM neural net to represent each class (face) in the stage of classification/identification. By applying the sub-space LDA method, we extract the most important characteristics in the identification of previously known faces that belong to the database, creating a reduced and more discriminated dimensional space than the original space. The SOM nets are responsible for the memorization of each class characteristic. The algorithm offers great performance (recognition rates between 97% and 98%) considering the adversities and sources of errors inherent to the traditional methods of face recognition.

O uso de dados biométricos da face para verificação automática de identidade é um dos maiores desafios em sistemas de controle de acesso seguro. O processo é extremamente complexo e influenciado por muitos fatores relacionados à forma, posição, iluminação, rotação, translação, disfarce e oclusão de características faciais. Hoje existem muitas técnicas para se reconhecer uma face. Esse trabalho apresenta uma investigação buscando identificar uma face no banco de dados ORL com diferentes grupos de treinamento. É proposto um algoritmo para o reconhecimento de faces baseado na técnica de subespaço LDA (PCA + LDA) utilizando uma rede neural SOM para representar cada classe (face) na etapa de classificação/identificação. Aplicando o método do subespaço LDA busca-se extrair as características mais importantes na identificação das faces previamente conhecidas e presentes no banco de dados, criando um espaço dimensional menor e discriminante com relação ao espaço original. As redes SOM são responsáveis pela memorização das características de cada classe. O algoritmo oferece maior desempenho (taxas de reconhecimento entre 97% e 98%) com relação às adversidades e fontes de erros que prejudicam os métodos de reconhecimento de faces tradicionais. / The use of biometric technique for automatic personal identification is one of the biggest challenges in the security field. The process is complex because it is influenced by many factors related to the form, position, illumination, rotation, translation, disguise and occlusion of face characteristics. Now a days, there are many face recognition techniques. This work presents a methodology for searching a face in the ORL database with some different training sets. The algorithm for face recognition was based on sub-space LDA (PCA + LDA) technique using a SOM neural net to represent each class (face) in the stage of classification/identification. By applying the sub-space LDA method, we extract the most important characteristics in the identification of previously known faces that belong to the database, creating a reduced and more discriminated dimensional space than the original space. The SOM nets are responsible for the memorization of each class characteristic. The algorithm offers great performance (recognition rates between 97% and 98%) considering the adversities and sources of errors inherent to the traditional methods of face recognition.