Rozprawy doktorskie na temat „Linear perceptrons”

Dissertação (mestrado) - Universidade Federal de Santa Catarina, Centro Tecnológico. Programa de Pós-Graduação em Ciência da Computação.
Made available in DSpace on 2012-10-24T01:24:51Z (GMT). No. of bitstreams: 1 258459.pdf: 252997 bytes, checksum: a0457bb78b352c0aab2bb1f48ab79985 (MD5)
Este trabalho descreve a aplicação de uma técnica de regressão não linear (mínimos quadrados) para obter predições intervalares em redes neurais artificiais (RNA#s). Através de uma simulação de Monte Carlo é mostrada uma maneira de escolher um ajuste de parâmetros (pesos) para uma rede neural, de acordo com um critério de seleção que é baseado na magnitude dos intervalos de predição fornecidos pela rede. Com esta técnica foi possível obter as predições intervalares com amplitude desejada e com probabilidade de cobertura conhecida, de acordo com um grau de confiança escolhido. Os resultados e as discussões associadas indicam ser possível e factível a obtenção destes intervalos, fazendo com que a resposta das redes seja mais informativa e consequentemente aumentando sua aplicabilidade. A implementação computacional está disponível em www.inf.ufsc.br/~dandrade. This work describes the application of a nonlinear regression technique (least squares) to create prediction intervals on artificial neural networks (ANN´s). Through Monte Carlo#s simulations it is shown a way of choosing the set of parameters (weights) to a neural network, according to a selection criteria based on the magnitude of the prediction intervals provided by the net. With this technique it is possible to obtain the prediction intervals with the desired amplitude and with known coverage probability, according to the chosen confidence level. The associated results and discussions indicate to be possible and feasible to obtain these intervals, thus making the network response more informative and consequently increasing its applicability. The computational implementation is available in www.inf.ufsc.br/~dandrade.

Les travaux présentés dans cette thèse relèvent de l'étude des méthodes de classification linéaires, c'est à dire l'étude de méthodes ayant pour but la catégorisation de données en différents groupes à partir d'un jeu d'exemples, préalablement étiquetés, disponible en amont et appelés ensemble d'apprentissage. En pratique, l'acquisition d'un tel ensemble d'apprentissage peut être difficile et/ou couteux, la catégorisation d'un exemple étant de fait plus ardu que l'obtention de dudit exemple. Cette disparité entre la disponibilité des données et notre capacité à constituer un ensemble d'apprentissage étiqueté a été un des problèmes centraux de l'apprentissage automatique et ce manuscrit s’intéresse à deux solutions usuellement considérées pour contourner ce problème : l'apprentissage en présence de données bruitées et l'apprentissage actif
The works presented in this thesis fall within the general framework of linear classification, that is the problem of categorizing data into two or more classes based on on a training set of labelled data. In practice though acquiring labeled examples might prove challenging and/or costly as data are inherently easier to obtain than to label. Dealing with label scarceness have been a motivational goal in the machine learning literature and this work discuss two settings related to this problem: learning in the presence of noise and active learning

The electro-oculogram (EOG) is the most widely used technique for recording eye movements in clinical settings. It is inexpensive, practical, and non-invasive. Use of EOG is usually restricted to horizontal recordings as vertical EOG contains eyelid artefact (Oster & Stern, 1980) and blinks. The ability to analyse two dimensional (2D) eye movements may provide additional diagnostic information on pathologies, and further insights into the nature of brain functioning. Simultaneous recording of both horizontal and vertical EOG also introduces other difficulties into calibration of the eye movements, such as different gains in the two signals, and misalignment of electrodes producing crosstalk. These transformations of the signals create problems in relating the two dimensional EOG to actual rotations of the eyes. The application of an artificial neural network (ANN) that could map 2D recordings into 2D eye positions would overcome this problem and improve the utility of EOG. To determine whether ANNs are capable of correctly calibrating the saccadic eye movement data from 2D EOG (i.e. performing the necessary inverse transformation), the ANNs were first tested on data generated from mathematical models of saccadic eye movements. Multi-layer perceptrons (MLPs) with non-linear activation functions and trained with back propagation proved to be capable of calibrating simulated EOG data to a mean accuracy of 0.33° of visual angle (SE = 0.01). Linear perceptrons (LPs) were only nearly half as accurate. For five subjects performing a saccadic eye movement task in the upper right quadrant of the visual field, the mean accuracy provided by the MLPs was 1.07° of visual angle (SE = 0.01) for EOG data, and 0.95° of visual angle (SE = 0.03) for infrared limbus reflection (IRIS®) data. MLPs enabled calibration of 2D saccadic EOG to an accuracy not significantly different to that obtained with the infrared limbus tracking data.

A utilização de Redes Neurais Artificias para fins de identificação e controle de sistemas dinâmicos têm recebido atenção especial de muitos pesquisadores, principalmente no que se refere a sistemas não lineares. Neste trabalho é apresentado um estudo sobre a utilização de um tipo em particular de Rede Neural Artificial, uma Perceptron Multicamadas com Atraso de Tempo, na estimação de estados da etapa fermentativa do processo de Reichstein para produção de vitamina C. A aplicação de Redes Neurais Artificiais a este processo pode ser justificada pela existência de problemas associados à esta etapa, como variáveis de estado não mensuráveis e com incertezas de medida e não linearidade do processo fermentativo, além da dificuldade em se obter um modelo convencional que contemple todas as fases do processo. É estudado também a eficácia do algoritmo de Levenberg-Marquadt, na aceleração do treinamento da Rede Neural Artificial, além de uma comparação do desempenho de estimação de estados das Redes Neurais Artificiais estudadas com o filtro estendido de Kalman, baseado em um modelo não estruturado do processo fermentativo. A análise do desempenho das Redes Neurais Artificiais estudadas é avaliada em termos de uma figura de mérito baseada no erro médio quadrático sendo feitas considerações quanto ao tipo da função de ativação e o número de unidades da camada oculta. Os dados utilizados para treinamento e avaliação da Redes Neurais Artificiais foram obtidos de um conjunto de ensaios interpolados para o intervalo de amostragem desejado.
ldentification and Control of dynamic systems using Artificial Neural Networks has been widely investigated by many researchers in the last few years, with special attention to the application of these in nonlinear systems. ls this works, a study on the utilization of a particular type of Artificial Neural Networks, a Time Delay Multi Layer Perceptron, in the state estimation of the fermentative phase of the Reichstein process of the C vitamin production. The use of Artificial Neural Networks can be justified by the presence of problems, such as uncertain and unmeasurable state variables and process non-linearity, and by the fact that a conventional model that works on all phases of the fermentative processes is very difficult to obtain. The efficiency of the Levenberg Marquadt algorithm on the acceleration of the training process is also studied. Also, a comparison is performed between the studied Artificial Neural Networks and an extended Kalman filter based on a non-structured model for this fermentative process. The analysis of lhe Artificial Neural Networks is carried out using lhe mean square errors taking into consideration lhe activation function and the number of units presents in the hidden layer. A set of batch experimental runs, interpolated to the desired time interval, is used for training and validating the Artificial Neural Networks.

Orientador: Peter Sussner
Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matemática Estatística e Computação Científica
Made available in DSpace on 2018-08-26T15:06:30Z (GMT). No. of bitstreams: 1 Bueno_FelipeRoberto_M.pdf: 1499339 bytes, checksum: 85b58d8b856fafa47974349e80c1729e (MD5) Previous issue date: 2015
Resumo: Perceptrons morfológicos (MPs) pertencem à classe de redes neurais morfológicas (MNNs). Estas redes representam uma classe de redes neurais artificiais que executam operações de morfologia matemática (MM) em cada nó, possivelmente seguido pela aplicação de uma função de ativação. Vale ressaltar que a morfologia matemática foi concebida como uma teoria para processamento e análise de objetos (imagens ou sinais), por meio de outros objetos chamados elementos estruturantes. Embora inicialmente desenvolvida para o processamento de imagens binárias e posteriormente estendida para o processamento de imagens em tons de cinza, a morfologia matemática pode ser conduzida de modo mais geral em uma estrutura de reticulados completos. Originalmente, as redes neurais morfológicas empregavam somente determinadas operações da morfologia matemática em tons de cinza, denominadas de erosão e dilatação em tons de cinza, segundo a abordagem umbra. Estas operações podem ser expressas em termos de produtos máximo aditivo e mínimo aditivo, definidos por meio de operações entre vetores ou matrizes, da álgebra minimax. Recentemente, as operações da morfologia matemática fuzzy surgiram como funções de agregação das redes neurais morfológicas. Neste caso, falamos em redes neurais morfológicas fuzzy. Perceptrons híbridos lineares/morfológicos fuzzy foram inicialmente projetados como uma generalização dos perceptrons lineares/morfológicos existentes, ou seja, os perceptrons lineares/morfológicos fuzzy podem ser definidos por uma combinação convexa de uma parte morfológica fuzzy e uma parte linear. Nesta dissertação de mestrado, introduzimos uma rede neural artificial alimentada adiante, representando um perceptron híbrido linear/morfológico fuzzy chamado F-DELP (do inglês fuzzy dilation/erosion/linear perceptron), que ainda não foi considerado na literatura de redes neurais. Seguindo as ideias de Pessoa e Maragos, aplicamos uma suavização adequada para superar a não-diferenciabilidade dos operadores de dilatação e erosão fuzzy utilizados no modelo F-DELP. Em seguida, o treinamento é realizado por intermédio de um algoritmo de retropropagação de erro tradicional. Desta forma, aplicamos o modelo F-DELP em alguns problemas de classificação conhecidos e comparamos seus resultados com os produzidos por outros classificadores
Abstract: Morphological perceptrons (MPs) belong to the class of morphological neural networks (MNNs). These MNNs represent a class of artificial neural networks that perform operations of mathematical morphology (MM) at every node, possibly followed by the application of an activation function. Recall that mathematical morphology was conceived as a theory for processing and analyzing objects (images or signals), by means of other objects called structuring elements. Although initially developed for binary image processing and later extended to gray-scale image processing, mathematical morphology can be conducted very generally in a complete lattice setting. Originally, morphological neural networks only employed certain operations of gray-scale mathematical morphology, namely gray-scale erosion and dilation according to the umbra approach. These operations can be expressed in terms of (additive maximum and additive minimum) matrix-vector products in minimax algebra. It was not until recently that operations of fuzzy mathematical morphology emerged as aggregation functions of morphological neural networks. In this case, we speak of fuzzy morphological neural networks. Hybrid fuzzy morphological/linear perceptrons was initially designed by generalizing existing morphological/linear perceptrons, in other words, fuzzy morphological/linear perceptrons can be defined by a convex combination of a fuzzy morphological part and a linear part. In this master's thesis, we introduce a feedforward artificial neural network representing a hybrid fuzzy morphological/linear perceptron called fuzzy dilation/erosion/linear perceptron (F-DELP), which has not yet been considered in the literature. Following Pessoa's and Maragos' ideas, we apply an appropriate smoothing to overcome the non-differentiability of the fuzzy dilation and erosion operators employed in the proposed F-DELP models. Then, training is achieved using a traditional backpropagation algorithm. Finally, we apply the F-DELP model to some well-known classification problems and compare the results with the ones produced by other classifiers
Mestrado
Matematica Aplicada
Mestre em Matemática Aplicada

The subject of this thesis is the original study of the application of the multi-layer perceptron architecture to channel equalization in digital communications systems. Both theoretical analyses and simulations were performed to explore the performance of the perceptron-based equalizer (including the decision feedback equalizer). Topics covered include the factors that affect performance of the structures including, the parameters (learning gain and momentum parameter) in the learning algorithm, the network topology (input dimension, number of neurons and the number of hidden layers), and the power metrics on the error cost function. Based on the geometric hyperplane analysis of the multi-layer perceptron, the results offer valuable insight into the properties and complexity of the network. Comparisons of the bit error rate performance and the dynamic behaviour of the decision boundary of the perceptron-based equalizer with both the optimal non-linear equalizer and the optimal linear equalizer are provided. Through comparisons, some asymptotic results for the performance in the perceptron-based equalizer are obtained. Furthermore, a comparison of the performance of the perceptron-based equalizer (including the decision feedback equalizer) with the least mean squares linear transversal equalizer (including decision feedback equalizer) indicates that the former offers significant reduction in the bit error rate. This is because it has the ability to form highly nonlinear decision regions, in contrast with the linear equalizer which only forms linear decision regions. The linearity of the decision regions limits the performance of the conventional linear equalizer.

This thesis describes the development and implementation of an on-line optimal predictive controller incorporating a neural network model of a non-linear process. The scheme is based on a Multi-Layer Perceptron neural net-work as a modelling tool for a real non-linear, dual tank, liquid level process. A neural network process model is developed and evaluated firstly in simulation studies and then subsequently on the real process. During the development of the network model, the ability of the network to predict the process output multiple time steps ahead was investigated. This led to investigations into a number of important aspects such as the network topology, training algorithms, period of network training, model validation and conditioning of the process data. Once the development of the neural network model had been achieved, it was included into a predictive control scheme where an on-line comparison with a conventional three term controller was undertaken. Improvements in process control performance that can be achieved in practice using a neural control scheme are illustrated. Additionally, an insight into the dynamics and stability of the neural control scheme was obtained in a novel application of linear system identification techniques. The research shows that a technique of conditioning the process data, called spread encoding, enabled a neural network to accurately emulate the real process using only process input information and this facilitated accurate multi-step-ahead predictive control to be performed.

Dissertação (mestrado) - Pontifícia Universidade Católica do Paraná, Curitiba, 2006
Inclui bibliografia
A identificação de sistemas dinâmicos não-lineares multivariáveis é uma área importante em várias áreas da Engenharia. Esta dissertação apresenta o estudo de uma metodologia baseada em redes neurais artificiais para identificação de sistemas não-lineares

The increased use of digital media to store legal, as well as illegal data, has created the need for specialized tools that can monitor, control and even recover this data. An important task in computer forensics and security is to identify the true le type to which a computer le or computer le fragment belongs. File type identi cation is traditionally done by means of metadata, such as le extensions and le header and footer signatures. As a result, traditional metadata-based le object type identi cation techniques work well in cases where the required metadata is available and unaltered. However, traditional approaches are not reliable when the integrity of metadata is not guaranteed or metadata is unavailable. As an alternative, any pattern in the content of a le object can be used to determine the associated le type. This is called content-based le object type identi cation. Supervised learning techniques can be used to infer a le object type classi er by exploiting some unique pattern that underlies a le type's common le structure. This study builds on existing literature regarding the use of supervised learning techniques for content-based le object type identi cation, and explores the combined use of multilayer perceptron neural network classi ers and linear programming-based discriminant classi ers as a solution to the multiple class le fragment type identi cation problem. The purpose of this study was to investigate and compare the use of a single multilayer perceptron neural network classi er, a single linear programming-based discriminant classi- er and a combined ensemble of these classi ers in the eld of le type identi cation. The ability of each individual classi er and the ensemble of these classi ers to accurately predict the le type to which a le fragment belongs were tested empirically. The study found that both a multilayer perceptron neural network and a linear programming- based discriminant classi er (used in a round robin) seemed to perform well in solving the multiple class le fragment type identi cation problem. The results of combining multilayer perceptron neural network classi ers and linear programming-based discriminant classi ers in an ensemble were not better than those of the single optimized classi ers.
MSc (Computer Science), North-West University, Potchefstroom Campus, 2013

Dissertação (mestrado) - Pontifícia Universidade Católica do Paraná, Curitiba, 2011
Bibliografia: f.89-95
A modelagem matemática baseada em leis elementares da física nem sempre é uma tarefa possível se a complexidade do sistema estudado for considerada alta. Porém, utilizando técnicas de identificação de sistemas, é possível obter modelos matemáticos a parti
Mathematical modelling based on elementary laws of physics is not always possible if the complexity of the studied system is high. However, using system identification techniques, it is possible to build mathematical models based on real observed data fro

Modern methods of speech synthesis are not suitable for restoration of song signals due to lack of vitality and intonation in the resulted sounds. The aim of presented work is to synthesize melismas met in Lithuanian folk songs, by applying Artificial Neural Networks. An analytical survey of rather a widespread literature is presented. First classification and comprehensive discussion of melismas are given. The theory of dynamic systems which will make the basis for studying melismas is presented and finally the relationship for modeling a melisma with nonlinear and dynamic systems is outlined. Investigation of the most widely used Linear Prediction Coding method and possibilities of its improvement. The modification of original Linear Prediction method based on dynamic LPC frame positioning is proposed. On its basis, the new melisma synthesis technique is presented. Developed flexible generalized melisma model, based on two Artificial Neural Networks – a Multilayer Perceptron and Adaline – as well as on two network training algorithms – Levenberg- Marquardt and the Least Squares error minimization – is presented. Moreover, original mathematical models of Fortis, Gruppett, Mordent and Trill are created, fit for synthesizing melismas, and their minimal sizes are proposed. The last chapter concerns experimental investigation, using over 500 melisma records, and corroborates application of the new mathematical models to melisma synthesis of one performer.

In Sweden, there is a strong and growing interest in solar power. In recent years, photovoltaic (PV) system installations have increased dramatically and a large part are distributed grid connected PV systems i.e. rooftop installations. Currently the electricity export rate is significantly lower than the import rate which has made the amount of self-consumed PV electricity a critical factor when assessing the system profitability. Self-consumption (SC) is calculated using hourly or sub-hourly timesteps and is highly dependent on the solar patterns of the location of interest, the PV system configuration and the building load. As this varies for all potential installations it is difficult to make estimations without having historical data of both load and local irradiance, which is often hard to acquire or not available. A method to predict SC using commonly available information at the planning phase is therefore preferred.  There is a scarcity of documented SC data and only a few reports treating the subject of mapping or predicting SC. Therefore, this thesis is investigating the possibility of utilizing machine learning to create models able to predict the SC using the inputs: Annual load, annual PV production, tilt angle and azimuth angle of the modules, and the latitude. With the programming language Python, seven models are created using regression techniques, using real load data and simulated PV data from the south of Sweden, and evaluated using coefficient of determination (R2) and mean absolute error (MAE). The techniques are Linear Regression, Polynomial regression, Ridge Regression, Lasso regression, K-Nearest Neighbors (kNN), Random Forest, Multi-Layer Perceptron (MLP), as well as the only other SC prediction model found in the literature. A parametric analysis of the models is conducted, removing one variable at a time to assess the model’s dependence on each variable.  The results are promising, with five out of eight models achieving an R2 value above 0.9 and can be considered good for predicting SC. The best performing model, Random Forest, has an R2 of 0.985 and a MAE of 0.0148. The parametric analysis also shows that while more input data is helpful, using only annual load and PV production is sufficient to make good predictions. This can only be stated for model performance for the southern region of Sweden, however, and are not applicable to areas outside the latitudes or country tested.
I Sverige finns ett starkt och växande intresse för solenergi. De senaste åren har antalet solcellsanläggningar ökat dramatiskt och en stor del är distribuerade nätanslutna solcellssystem, dvs takinstallationer. För närvarande är elexportpriset betydligt lägre än importpriset, vilket har gjort mängden egenanvänd solel till en kritisk faktor vid bedömningen av systemets lönsamhet. Egenanvändning (EA) beräknas med tidssteg upp till en timmes längd och är i hög grad beroende av solstrålningsmönstret för platsen av intresse, PV-systemkonfigurationen och byggnadens energibehov. Eftersom detta varierar för alla potentiella installationer är det svårt att göra uppskattningar utan att ha historiska data om både energibehov och lokal solstrålning, vilket ofta inte är tillgängligt. En metod för att förutsäga EA med allmän tillgänglig information är därför att föredra.  Det finns en brist på dokumenterad EA-data och endast ett fåtal rapporter som behandlar kartläggning och prediktion av EA. I denna uppsats undersöks möjligheten att använda maskininlärning för att skapa modeller som kan förutsäga EA. De variabler som ingår är årlig energiförbrukning, årlig solcellsproduktion, lutningsvinkel och azimutvinkel för modulerna och latitud. Med programmeringsspråket Python skapas sju modeller med hjälp av olika regressionstekniker, där energiförbruknings- och simulerad solelproduktionsdata från södra Sverige används. Modellerna utvärderas med hjälp av determinationskoefficienten (R2) och mean absolute error (MAE). Teknikerna som används är linjär regression, polynomregression, Ridge regression, Lasso regression, K-nearest neighbor regression, Random Forest regression, Multi-Layer Perceptron regression. En additionell linjär regressions-modell skapas även med samma metodik som används i en tidigare publicerad rapport. En parametrisk analys av modellerna genomförs, där en variabel exkluderas åt gången för att bedöma modellens beroende av varje enskild variabel.  Resultaten är mycket lovande, där fem av de åtta undersökta modeller uppnår ett R2-värde över 0,9. Den bästa modellen, Random Forest, har ett R2 på 0,985 och ett MAE på 0,0148. Den parametriska analysen visar också att även om ingångsdata är till hjälp, är det tillräckligt att använda årlig energiförbrukning och årlig solcellsproduktion för att göra bra förutsägelser. Det måste dock påpekas att modellprestandan endast är tillförlitlig för södra Sverige, från var beräkningsdata är hämtad, och inte tillämplig för områden utanför de valda latituderna eller land.

Engine mapping is the process of empirically modelling engine behaviour as a function of adjustable engine parameters, predicting the output of the engine. The aim is to calibrate the electronic engine controller to meet decreasing emission requirements and increasing fuel economy demands. Modern engines have an increasing number of control parameters that are having a dramatic impact on time and e ort required to obtain optimal engine calibrations. These are further complicated due to transient engine operating mode. A new model-based transient calibration method has been built on the application of hierarchical statistical modelling methods, and analysis of repeated experiments for the application of engine mapping. The methodology is based on two-stage regression approach, which organise the engine data for the mapping process in sweeps. The introduction of time-dependent covariates in the hierarchy of the modelling led to the development of a new approach for the problem of transient engine calibration. This new approach for transient engine modelling is analysed using a small designed data set for a throttle body inferred air ow phenomenon. The data collection for the model was performed on a transient engine test bed as a part of this work, with sophisticated software and hardware installed on it. Models and their associated experimental design protocols have been identi ed that permits the models capable of accurately predicting the desired response features over the whole region of operability. Further, during the course of the work, the utility of multi-layer perceptron (MLP) neural network based model for the multi-covariate case has been demonstrated. The MLP neural network performs slightly better than the radial basis function (RBF) model. The basis of this comparison is made on assessing relevant model selection criteria, as well as internal and external validation ts. Finally, the general ability of the model was demonstrated through the implementation of this methodology for use in the calibration process, for populating the electronic engine control module lookup tables.

Agriculture is becoming more dependent on computerized solutions to make thefarmer’s job easier. The big step that many companies are working towards is fullyautonomous vehicles that work the fields. To that end, the equipment fitted to saidvehicles must also adapt and become autonomous. Making this equipment autonomoustakes many incremental steps, one of which is developing an accurate and reliable tanklevel measurement system. In this thesis, a system for tank level measurement in a seedplanting machine is evaluated. Traditional systems use load cells to measure the weightof the tank however, these types of systems are expensive to build and cumbersome torepair. They also add a lot of weight to the equipment which increases the fuel consump-tion of the tractor. Thus, this thesis investigates the use of radar sensors together witha number of Machine Learning algorithms. Fourteen radar sensors are fitted to a tankat different positions, data is collected, and a preprocessing method is developed. Then,the data is used to test the following Machine Learning algorithms: Bagged RegressionTrees (BG), Random Forest Regression (RF), Boosted Regression Trees (BRT), LinearRegression (LR), Linear Support Vector Machine (L-SVM), Multi-Layer Perceptron Re-gressor (MLPR). The model with the best 5-fold crossvalidation scores was Random For-est, closely followed by Boosted Regression Trees. A robustness test, using 5 previouslyunseen scenarios, revealed that the Boosted Regression Trees model was the most robust.The radar position analysis showed that 6 sensors together with the MLPR model gavethe best RMSE scores.In conclusion, the models performed well on this type of system which shows thatthey might be a competitive alternative to load cell based systems.

Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics
The purpose of this thesis is to conduct comparative research between Genetic Programming (GP) and Geometric Semantic Genetic Programming (GSGP), with different initialization (RHH and EDDA) and selection (Tournament and Epsilon-Lexicase) strategies, in the context of a model-ensemble in order to solve regression optimization problems. A model-ensemble is a combination of base learners used in different ways to solve a problem. The most common ensemble is the mean, where the base learners are combined in a linear fashion, all having the same weights. However, more sophisticated ensembles can be inferred, providing higher generalization ability. GSGP is a variant of GP using different genetic operators. No previous research has been conducted to see if GSGP can perform better than GP in model-ensemble learning. The evolutionary process of GP and GSGP should allow us to learn about the strength of each of those base models to provide a more accurate and robust solution. The base-models used for this analysis were Linear Regression, Random Forest, Support Vector Machine and Multi-Layer Perceptron. This analysis has been conducted using 7 different optimization problems and 4 real-world datasets. The results obtained with GSGP are statistically significantly better than GP for most cases.
O objetivo desta tese é realizar pesquisas comparativas entre Programação Genética (GP) e Programação Genética Semântica Geométrica (GSGP), com diferentes estratégias de inicialização (RHH e EDDA) e seleção (Tournament e Epsilon-Lexicase), no contexto de um conjunto de modelos, a fim de resolver problemas de otimização de regressão. Um conjunto de modelos é uma combinação de alunos de base usados de diferentes maneiras para resolver um problema. O conjunto mais comum é a média, na qual os alunos da base são combinados de maneira linear, todos com os mesmos pesos. No entanto, conjuntos mais sofisticados podem ser inferidos, proporcionando maior capacidade de generalização. O GSGP é uma variante do GP usando diferentes operadores genéticos. Nenhuma pesquisa anterior foi realizada para verificar se o GSGP pode ter um desempenho melhor que o GP no aprendizado de modelos. O processo evolutivo do GP e GSGP deve permitir-nos aprender sobre a força de cada um desses modelos de base para fornecer uma solução mais precisa e robusta. Os modelos de base utilizados para esta análise foram: Regressão Linear, Floresta Aleatória, Máquina de Vetor de Suporte e Perceptron de Camadas Múltiplas. Essa análise foi realizada usando 7 problemas de otimização diferentes e 4 conjuntos de dados do mundo real. Os resultados obtidos com o GSGP são estatisticamente significativamente melhores que o GP na maioria dos casos.