Neural networks have become very important tools in many areas including economic researches. The objectives of this thesis are to examine the fundamental components, concepts and theory of neural network methods from econometric and statistic perspective, with particular focus on econometrically and statistically relevant models. In order to evaluate the relative effectiveness of econometric and neural network methods, two empirical studies are conducted by applying neural network methods in a methodological comparison fashion with traditional econometric models.Both neural networks and econometrics have similar models, common problems of modeling and interference. Neural networks and econometrics/statistics, particularly their discriminant methods, are two sides of the same coin in terms of the nature of modeling statistic issues. On one side, econometric models are sampling paradigm oriented methods, which estimate the distribution of the predictor variable separately for each class and combine these with the prior probabilities of each class occurring; while neural networks are one of the techniques based on diagnostic paradigm, which use theinformation from the samples to estimate the conditional probability of an observation belonging to each class, based on predictor variables. Hence, neural network and econometric/statistical methods (particularly, discriminant models) have the same properties, except that the natural parameterizations differ.The empirical studies indicate that neural network methods outperform or are as good as traditional econometric models including Multiple Regression Analysis, Linear Probability Model (LPM), and Logit model, in terms of minimizing the errors of in-sample predictions and out-of-sample forecasts. Although neural networks have some advantages over econometric methods, they have some limitations too. Hence, neural networks are perhaps best viewed as supplements to econometric methods in studying economic issues, and not necessarily as substitutes.

Artificial neural networks are demanding to computational power of a computer. Increasing their learning speed could mean new posibilities for research or aplication of the algorithm. And that is a purpose of this thesis. The usage of graphics processing units for neural networks learning is one way how to achieve above mentioned goals. This thesis is offering a survey of theoretical background and consequently implementation of a software library for neural networks learning with a Backpropagation algorithm with a support of acceleration on graphics processing unit.

Neural networks can perform an incredible array of complex tasks, but successfully training a network is difficult because it requires us to minimize a function about which we know very little. In practice, developing a good model requires both intuition and a lot of guess-and-check. In this dissertation, we study a type of fully-connected neural network that improves on standard rectifier networks while retaining their useful properties. We then examine this type of network and its loss function from a probabilistic perspective. This analysis leads to a new rule for parameter initialization and a new method for predicting effective learning rates for gradient descent. Experiments confirm that the theory behind these developments translates well into practice.

In the present study neural networks are investigated for use in fluid dynamics simulations. These range from static simulations for a simple 2D geometry like an airfoil section to dynamic simulations for a complicated 3D geometry like a model submarine. A detailed analysis of the application of neural networks for the case of vehicle trajectory determination is provided. This involves identifying the physics of the problem and tailoring it to a neural network architecture. The learning process involves training the neural network on a variety of maneuvers and the prediction process involves applying new maneuvers to the neural network. The results are compared to both experimental data and CFD data for the training sets and the prediction sets. The need and scope for parallelization in neural networks is also examined and the performance of pattern partitioning and vertical partitioning algorithms is studied.

Neural networks typically struggle with reasoning tasks on out of domain data, something that humans can more easily adapt to. Humans come with prior knowledge of concepts and can segment their environment into building blocks (such as objects) that allow them to reason effectively in unfamiliar situations. Using this intuition, we train a network that utilizes fixed embeddings from the CLIP (Contrastive Language--Image Pre-training) model to do a simple task that the original CLIP model struggles with. The network learns concepts (such as "collide" and "avoid") in a supervised source domain in such a way that the network can adapt and identify similar concepts in a target domain with never-before-seen objects. Without any training in the target domain, we show a 11% accuracy improvement in recognizing concepts compared to the baseline zero-shot CLIP model. When provided with a few labels, this accuracy gap widens to 20%.

O objetivo deste trabalho foi desenvolver um algoritmo baseado em redes neurais para o reconhecimento facial. O algoritmo contém dois módulos principais, um módulo para a extração de características e um módulo para o reconhecimento facial, sendo aplicado sobre imagens digitais nas quais a face foi previamente detectada. O método utilizado para a extração de características baseia-se na aplicação de assinaturas horizontais e verticais para localizar os componentes faciais (olhos e nariz) e definir a posição desses componentes. Como entrada foram utilizadas imagens faciais de três bancos distintos: PICS, ESSEX e AT&T. Para esse módulo, a média de acerto foi de 86.6%, para os três bancos de dados. No módulo de reconhecimento foi utilizada a arquitetura perceptron multicamadas (MLP), e para o treinamento dessa rede foi utilizado o algoritmo de aprendizagem backpropagation. As características faciais extraídas foram aplicadas nas entradas dessa rede neural, que realizou o reconhecimento da face. A rede conseguiu reconhecer 97% das imagens que foram identificadas como pertencendo ao banco de dados utilizado. Apesar dos resultados satisfatórios obtidos, constatou-se que essa rede não consegue separar adequadamente características faciais com valores muito próximos, e portanto, não é a rede mais eficiente para o reconhecimento facial / This research presents a facial recognition algorithm based in neural networks. The algorithm contains two main modules: one for feature extraction and another for face recognition. It was applied in digital images from three database, PICS, ESSEX and AT&T, where the face was previously detected. The method for feature extraction was based on previously knowledge of the facial components location (eyes and nose) and on the application of the horizontal and vertical signature for the identification of these components. The mean result obtained for this module was 86.6% for the three database. For the recognition module it was used the multilayer perceptron architecture (MLP), and for training this network it was used the backpropagation algorithm. The extracted facial features were applied to the input of the neural network, that identified the face as belonging or not to the database with 97% of hit ratio. Despite the good results obtained it was verified that the MLP could not distinguish facial features with very close values. Therefore the MLP is not the most efficient network for this task

A demanda e, consequentemente, a geração de energia elétrica são questões de suma importância para o desenvolvimento econômico e social dos países. Modelos para previsão destes parâmetros no longo e médio prazo são empregados com a finalidade de antever possíveis cenários e propor estratégias para a realização de um planejamento energético adequado. Neste contexto, o presente estudo tem como objetivo realizar a previsão da geração de energia elétrica no estado do Rio Grande do Sul (RS) em um horizonte de médio prazo (um ano), utilizando Redes Neurais Artificiais (RNA’s) do tipo feedforward com algoritmo de aprendizado supervisionado backpropagation. Para o desenvolvimento deste trabalho elaborou-se um script para executar as simulações necessárias, as quais foram realizadas através do software Matlab®. As variáveis de influência selecionadas como entradas do modelo de previsão referem-se à economia (estadual e nacional), ao balanço de energia elétrica e à meteorologia do estado, durante o período de janeiro de 2009 a março de 2016. Para realizar o treinamento da rede neural, adicionou-se a matriz de entrada este conjunto de dados, com frequência mensal, referentes a janeiro de 2009 a março de 2015 e para previsão foram inseridos dados de abril de 2015 a março de 2016. Por fim, depois de realizada a simulação completa da RNA, comparou-se o resultado observado da geração de energia elétrica do estado com o obtido através do modelo de previsão, indicando um erro percentual absoluto médio (MAPE) de 5,86% e um desvio absoluto médio (MAD) de 134,15 MW médio. Os resultados obtidos neste trabalho mostram-se promissores, além de semelhantes aos encontrados na literatura, demonstrando assim confiabilidade e eficácia do método empregado. / The demand and, consequently, the generation of electric power are very important issues for social and economic development of countries. Models to forecast these parameters in long and medium terms are used to anticipate possible sceneries and propose strategies for the energy planning of countries. In this context, the present study aims to forecast the generation of electric energy in Rio Grande do Sul State (RS) in a medium-term horizon (one year) using, Artificial Neural Networks (ANNs) of the feedforward type with algorithm of supervised learning backpropagation. For the development of this work, a script was elaborated in order to execute the necessary simulations, which were carried out through Matlab® software. The selected variables of influence as inputs of forecasting model refer to economy (State and National), to the electric energy balance and to the meteorology State, during the period from January, 2009 to March, 2016. In order to train the neural network, this data set was added to the entrance matrix, with monthly frequency, from January, 2009 to March, 2015 and for prediction, data were inserted from April, 2015 to March, 2016. Finally, after RNA complete simulation, the observed result of the electric power generation of the State was compared with the one obtained through the prediction model, indicating a mean absolute percent error (MAPE) of 5.86% and a mean absolute deviation (MAD) of 134.15 average MW. The obtained results in this work are promising, besides; they are similar to those found in literature, in this way demonstrating the reliability and efficacy of the using method.

A demanda e, consequentemente, a geração de energia elétrica são questões de suma importância para o desenvolvimento econômico e social dos países. Modelos para previsão destes parâmetros no longo e médio prazo são empregados com a finalidade de antever possíveis cenários e propor estratégias para a realização de um planejamento energético adequado. Neste contexto, o presente estudo tem como objetivo realizar a previsão da geração de energia elétrica no estado do Rio Grande do Sul (RS) em um horizonte de médio prazo (um ano), utilizando Redes Neurais Artificiais (RNA’s) do tipo feedforward com algoritmo de aprendizado supervisionado backpropagation. Para o desenvolvimento deste trabalho elaborou-se um script para executar as simulações necessárias, as quais foram realizadas através do software Matlab®. As variáveis de influência selecionadas como entradas do modelo de previsão referem-se à economia (estadual e nacional), ao balanço de energia elétrica e à meteorologia do estado, durante o período de janeiro de 2009 a março de 2016. Para realizar o treinamento da rede neural, adicionou-se a matriz de entrada este conjunto de dados, com frequência mensal, referentes a janeiro de 2009 a março de 2015 e para previsão foram inseridos dados de abril de 2015 a março de 2016. Por fim, depois de realizada a simulação completa da RNA, comparou-se o resultado observado da geração de energia elétrica do estado com o obtido através do modelo de previsão, indicando um erro percentual absoluto médio (MAPE) de 5,86% e um desvio absoluto médio (MAD) de 134,15 MW médio. Os resultados obtidos neste trabalho mostram-se promissores, além de semelhantes aos encontrados na literatura, demonstrando assim confiabilidade e eficácia do método empregado. / The demand and, consequently, the generation of electric power are very important issues for social and economic development of countries. Models to forecast these parameters in long and medium terms are used to anticipate possible sceneries and propose strategies for the energy planning of countries. In this context, the present study aims to forecast the generation of electric energy in Rio Grande do Sul State (RS) in a medium-term horizon (one year) using, Artificial Neural Networks (ANNs) of the feedforward type with algorithm of supervised learning backpropagation. For the development of this work, a script was elaborated in order to execute the necessary simulations, which were carried out through Matlab® software. The selected variables of influence as inputs of forecasting model refer to economy (State and National), to the electric energy balance and to the meteorology State, during the period from January, 2009 to March, 2016. In order to train the neural network, this data set was added to the entrance matrix, with monthly frequency, from January, 2009 to March, 2015 and for prediction, data were inserted from April, 2015 to March, 2016. Finally, after RNA complete simulation, the observed result of the electric power generation of the State was compared with the one obtained through the prediction model, indicating a mean absolute percent error (MAPE) of 5.86% and a mean absolute deviation (MAD) of 134.15 average MW. The obtained results in this work are promising, besides; they are similar to those found in literature, in this way demonstrating the reliability and efficacy of the using method.

A demanda e, consequentemente, a geração de energia elétrica são questões de suma importância para o desenvolvimento econômico e social dos países. Modelos para previsão destes parâmetros no longo e médio prazo são empregados com a finalidade de antever possíveis cenários e propor estratégias para a realização de um planejamento energético adequado. Neste contexto, o presente estudo tem como objetivo realizar a previsão da geração de energia elétrica no estado do Rio Grande do Sul (RS) em um horizonte de médio prazo (um ano), utilizando Redes Neurais Artificiais (RNA’s) do tipo feedforward com algoritmo de aprendizado supervisionado backpropagation. Para o desenvolvimento deste trabalho elaborou-se um script para executar as simulações necessárias, as quais foram realizadas através do software Matlab®. As variáveis de influência selecionadas como entradas do modelo de previsão referem-se à economia (estadual e nacional), ao balanço de energia elétrica e à meteorologia do estado, durante o período de janeiro de 2009 a março de 2016. Para realizar o treinamento da rede neural, adicionou-se a matriz de entrada este conjunto de dados, com frequência mensal, referentes a janeiro de 2009 a março de 2015 e para previsão foram inseridos dados de abril de 2015 a março de 2016. Por fim, depois de realizada a simulação completa da RNA, comparou-se o resultado observado da geração de energia elétrica do estado com o obtido através do modelo de previsão, indicando um erro percentual absoluto médio (MAPE) de 5,86% e um desvio absoluto médio (MAD) de 134,15 MW médio. Os resultados obtidos neste trabalho mostram-se promissores, além de semelhantes aos encontrados na literatura, demonstrando assim confiabilidade e eficácia do método empregado. / The demand and, consequently, the generation of electric power are very important issues for social and economic development of countries. Models to forecast these parameters in long and medium terms are used to anticipate possible sceneries and propose strategies for the energy planning of countries. In this context, the present study aims to forecast the generation of electric energy in Rio Grande do Sul State (RS) in a medium-term horizon (one year) using, Artificial Neural Networks (ANNs) of the feedforward type with algorithm of supervised learning backpropagation. For the development of this work, a script was elaborated in order to execute the necessary simulations, which were carried out through Matlab® software. The selected variables of influence as inputs of forecasting model refer to economy (State and National), to the electric energy balance and to the meteorology State, during the period from January, 2009 to March, 2016. In order to train the neural network, this data set was added to the entrance matrix, with monthly frequency, from January, 2009 to March, 2015 and for prediction, data were inserted from April, 2015 to March, 2016. Finally, after RNA complete simulation, the observed result of the electric power generation of the State was compared with the one obtained through the prediction model, indicating a mean absolute percent error (MAPE) of 5.86% and a mean absolute deviation (MAD) of 134.15 average MW. The obtained results in this work are promising, besides; they are similar to those found in literature, in this way demonstrating the reliability and efficacy of the using method.

O objetivo deste trabalho foi desenvolver um algoritmo baseado em redes neurais para o reconhecimento facial. O algoritmo contém dois módulos principais, um módulo para a extração de características e um módulo para o reconhecimento facial, sendo aplicado sobre imagens digitais nas quais a face foi previamente detectada. O método utilizado para a extração de características baseia-se na aplicação de assinaturas horizontais e verticais para localizar os componentes faciais (olhos e nariz) e definir a posição desses componentes. Como entrada foram utilizadas imagens faciais de três bancos distintos: PICS, ESSEX e AT&T. Para esse módulo, a média de acerto foi de 86.6%, para os três bancos de dados. No módulo de reconhecimento foi utilizada a arquitetura perceptron multicamadas (MLP), e para o treinamento dessa rede foi utilizado o algoritmo de aprendizagem backpropagation. As características faciais extraídas foram aplicadas nas entradas dessa rede neural, que realizou o reconhecimento da face. A rede conseguiu reconhecer 97% das imagens que foram identificadas como pertencendo ao banco de dados utilizado. Apesar dos resultados satisfatórios obtidos, constatou-se que essa rede não consegue separar adequadamente características faciais com valores muito próximos, e portanto, não é a rede mais eficiente para o reconhecimento facial / This research presents a facial recognition algorithm based in neural networks. The algorithm contains two main modules: one for feature extraction and another for face recognition. It was applied in digital images from three database, PICS, ESSEX and AT&T, where the face was previously detected. The method for feature extraction was based on previously knowledge of the facial components location (eyes and nose) and on the application of the horizontal and vertical signature for the identification of these components. The mean result obtained for this module was 86.6% for the three database. For the recognition module it was used the multilayer perceptron architecture (MLP), and for training this network it was used the backpropagation algorithm. The extracted facial features were applied to the input of the neural network, that identified the face as belonging or not to the database with 97% of hit ratio. Despite the good results obtained it was verified that the MLP could not distinguish facial features with very close values. Therefore the MLP is not the most efficient network for this task