Poggio and Vetter (1992) showed that learning one view of a bilaterally symmetric object could be sufficient for its recognition, if this view allows the computation of a symmetric, "virtual," view. Faces are roughly bilaterally symmetric objects. Learning a side-view--which always has a symmetric view--should allow for better generalization performances than learning the frontal view. Two psychophysical experiments tested these predictions. Stimuli were views of shaded 3D models of laser-scanned faces. The first experiment tested whether a particular view of a face was canonical. The second experiment tested which single views of a face give rise to best generalization performances. The results were compatible with the symmetry hypothesis: Learning a side view allowed better generalization performances than learning the frontal view.

None

Neural networks can be looked as a kind of intruments which is able to learn. For making the fruitful results of neural networks' learning possess parctical applied value, the thesis makes use of linear regression technics to strengthen the extended capability of RBF neural networks. The thesis researches the training methods of RBF neural networks, and retains the frame of OLS(orthogonal least square) learning rules which is published by Chen and Billings in 1992. Besides, aiming at the RBF's characteristics, the thesis brings up improved learning rules in first and second phases, and uses " early stop" to be the condition of training ceasing. To sum up, chiefly the thesis applies some technics of statistic linear regression to strenthen the extended capability of RBF, and using different methods to do computer simulation in different noise situations.

To understand remotely sensed data, one must understand the relationship between radiative transfer models and their predictions of the interaction of solar radiation on geophysical media. If it can be established that these models are indeed accurate, some form of evaluation has to be performed on these models, for users to choose the model that suits their requirements. This thesis focuses on the implementation of a variable linear kernel model, its validation, and to study its application in the prediction of BRDF effects using two different neural networks-- the backpropogation and the radial basis function neural network and finally to draw conclusions on which neural network is best suited for this model. Based on these results the optimum number of kernels for this model is derived.

No description available.

Nous avons dans ce travail testé avec succès la méthode RBF sur le problème raide, le problème de la concentration d'un contaminant, le modèle Black-Scholes et le modèle du champ classique d'un méson. Notre contribution a été importante lors de la résolution de l'équation non linéaire de Klein-Gordon. La convergence et l'efficacité de la méthode a été montré grâce au RMSE entre la solution analytique et la solution numérique. L'introduction mise à part, cette thèse a été composé de quatre chapitres. Le premier exprime l'interpolant radial dans la base du sous-espace des interpolés. Le second estime l'erreur d'interpolation dans des cas particuliers de la fonction radiale de base et fournit les meilleures constantes dans les majorations de l'erreur. Le troisième consacré au problème de la quasi-interpolation a aussi permis d'établir l'existence et l'unicité de la solution du champ classique d'un méson grâce à la théorie des semi-groupes et au théorème du point fixe de Banach. Le quatrième a été consacré aux applications numériques. Une simulation numérique a été faite pour le problème de la concentration d'un contaminant. Nous avons terminé par une conclusion et perspectives en désignant les futurs lignes de recherche sur le sujet.

No description available.

CoordenaÃÃo de AperfeiÃoamento de Pessoal de NÃvel Superior / O aumento da capacidade dos computadores nessas trÃs ultimas dÃcadas tem tornado possÃvel a soluÃÃo de problemas de engenharia cada vez mais complexos. Essa ampliaÃÃo na possibilidade de soluÃÃo de tais problemas à resultado do avanÃo nos mÃtodos numÃricos e do desenvolvimento de algoritmos eficientes. Entretanto, estes mÃtodos numÃricos sÃo baseados na construÃÃo de malhas de discretizaÃÃo e a geraÃÃo de malhas ainda representa o maior desafio desses mÃtodos. Por esse motivo, nos Ãltimos anos o foco dos estudos de modelagem de problemas relacionados Ãs Ãguas subterrÃneas està voltado para o desenvolvimento de âmÃtodos livres de malhasâ ( meslhess ou meshfree methods) que tÃm como objetivo eliminar ou, pelo menos aliviar os problemas associados à construÃÃo e/ou reconstruÃÃo de malhas. Em problemas transientes, nos mÃtodos numÃricos tradicionais, o espaÃo à discretizado e em seguida à feita uma nova discretizaÃÃo para o tempo que requer a escolha de uma relaÃÃo Ãtima entre o intervalo de tempo escolhido e a discretizaÃÃo do espaÃo. Esta tese propÃe o desenvolvimento de um modelo numÃrico-analÃtico para problemas transientes de fluxo e contaminaÃÃo de Ãgua subterrÃnea. à um mÃtodo numÃrico com relaÃÃo à descriÃÃo do espaÃo, onde serà utilizado o mÃtodo RBF livre de malha e à analÃtico com relaÃÃo ao tempo, onde serÃo geradas expressÃes matemÃticas para a parte transiente. TrÃs configuraÃÃes de problemas unidimensionais de Ãgua subterrÃnea foram modeladas pelo mÃtodo RBF livre de malha e pelo mÃtodo numÃrico-analÃtico (MNA), utilizando o MAPLE e ( versÃo 10.0) como linguagem de programaÃÃo. Nos trÃs casos estudados, o meio poroso à homogÃneo e saturado. Os resultados apresentados mostram a validaÃÃo de fÃsica do MNA, mas possuem algumas restriÃÃes em sua aplicaÃÃo, tais como domÃnios poucos discretizados e a escolha de um fator de forma Ãtimo. O presente trabalho mostra tambÃm que o modelo proposto acomoda condiÃÃes de contorno que variam com o tempo. / Improvements in computer capabilities in the last three decades make it possible to solve more and more complex engineering problems. The increase in possibilities for solving such problems has been due to advances in numerical methods and development of efficient algorithms. Nevertheless, these numerical methods are based on mesh discretization and it is widely acknowledged that mesh generation remains one of the biggest challenges in mesh-based methods. During recent years, groundwater problems modeling studies are focused on the development of meshless or mesh-free methods. The aim of the so-called mesh-free methods is to eliminate or at least minimize the problems associated with meshing and/or remeshing. Traditional numerical methods discretize space and then discretize time in order to solve transient problems. This procedure requires na optimal relationship between space and time discretizations. This work proposes the development of a numerical-analytical model for flow and contaminant transport groundwater transient problems. It is a numerical method with respect to space, with RBF meshfree method and it is analytical with respect to time, with mathematical expressions. Three onedimensional problems configurations were teste using RBF meshless method and numerical-analytical (MNA) method, in MAPLE program. In all three cases, porous media is homogeneous and saturated. Results show MNAâs physical validation, but there are some restrictions to its use, such as domain discretization, PÃclet number and optimal shape parameter. The present work also shows that MNA accommodates well varying boundary conditions.

The geotechnical conditions of the soil can fluctuate greatly across the wind farm. This is an issue since geotechnical modelling is the base of the structural design of an offshore wind farm, and the efficient installation of the wind turbines depends on its accuracy. This paper deals with the characterization of the seabed, predicting the soil properties over the total affected area by a wind farm, with the challenge to reduce the required data samples in the site investigation under the number of installed wind turbines, to reduce its cost. It is compared the prediction outcome from two different interpolation methods, kriging and radial basis function, assessing their accuracy by the Mean-Squared Error and the Goodness-of-Prediction Estimate, as well as with a visual examination of their mapping; obtaining higher accuracy for radial basis function and reducing to half the required sample points, from the initial value of installed wind turbines. In a second stage it is studied the soil effect over the foundation, analyzing the results from a FEA, where different geometries of the structure are compared submitted to different load cases to check its limit states. Those results show that the foundation cost can increase four times due to the soil conditions, taking into account only the steel volume, and demonstrating how important is the soil characterization in the foundation design, as it gives the chance to relocate those wind turbines that require more expensive foundations.

Submitted by PPG Engenharia El?trica (engenharia.pg.eletrica@pucrs.br) on 2017-09-18T13:18:06Z No. of bitstreams: 1 Diss_Helder_Robelcio_Junior.pdf: 2504827 bytes, checksum: 6687d65850121aa2895c1fa972749363 (MD5) / Rejected by Caroline Xavier (caroline.xavier@pucrs.br), reason: Devolvido devido ? falta de capa institucional no arquivo PDF. on 2017-09-21T14:34:31Z (GMT) / Submitted by PPG Engenharia El?trica (engenharia.pg.eletrica@pucrs.br) on 2017-09-21T18:31:36Z No. of bitstreams: 1 Diss_Helder_Robelcio_Junior.pdf: 2504827 bytes, checksum: 6687d65850121aa2895c1fa972749363 (MD5) / Approved for entry into archive by Caroline Xavier (caroline.xavier@pucrs.br) on 2017-09-26T13:13:22Z (GMT) No. of bitstreams: 1 Diss_Helder_Robelcio_Junior.pdf: 2504827 bytes, checksum: 6687d65850121aa2895c1fa972749363 (MD5) / Made available in DSpace on 2017-09-26T13:17:46Z (GMT). No. of bitstreams: 1 Diss_Helder_Robelcio_Junior.pdf: 2504827 bytes, checksum: 6687d65850121aa2895c1fa972749363 (MD5) Previous issue date: 2017-08-30 / Conselho Nacional de Pesquisa e Desenvolvimento Cient?fico e Tecnol?gico - CNPq / The wireless communication channel has severe signal degradation effects resulting from the usual multiplicity of propagation paths originated by reflection of the electromagnetic wave at specific points along the path of the digital transmitter-receiver link. Metallic building structures, for example, are points of reflection of the wave. This multiplicity of propagation paths, called multipath, generates signal interference on itself when multiple signals arrive at the receiver, degrading signal intelligibility, which increases the bit error rates of the link, reducing a reliability. This interference is called intersymbol interference (ISI) because, in the baseband signal at the receiver, overlapping of the digital modulation symbols occurs, resulting in a failure to detect the binary words associated with the symbols. In this context, the channel estimation and compensation process plays an important role in the wireless receiver. The increasing demand for systems with higher transmission capacities, robustness and less computational complexity, has driven several researches in the scope of the algorithms used in the channel estimation process. In recent years, data transmission technique through Orthogonal Frequency Division Multiplexing (OFDM) has been highlighted by resistance to ISI, good spectral efficiency and transmission capacity of high data rates. OFDM is a multi-carrier modulation technique that consists of dividing the total bandwidth into smaller subchannels by using orthogonal subcarriers spectrally superimposed. Despite its robustness, it is still required channel estimation techniques, in OFDM receiver, due to the multipath effect characteristic of a wireless communication channel. This work proposes the implementation of a channel estimator, based on a complex Radial Basis Function (RBF) network. The proposed network is trained from the impulse response of the channel obtained through the pilot carriers sent and known by the receiver. The simulation results show that the proposed network obtained better results than the classical estimators used for channel estimation in OFDM systems. / O canal de comunica??o sem fio (wireless) apresenta severos efeitos de degrada??o de sinal resultantes da usual multiplicidade de caminhos de propaga??o originados por reflex?o da onda eletromagn?tica em pontos espec?ficos ao longo do caminho do enlace entre transmissor e receptor digital. Estruturas met?licas de constru??es civis, por exemplo, constituem pontos de reflex?o da onda. Esta multiplicidade de caminhos de propaga??o, denominada multipercurso, gera interfer?ncia do sinal sobre ele mesmo quando os m?ltiplos sinais chegam ao receptor, degradando a inteligibilidade do sinal recebido em consequ?ncia dos m?ltiplos ecos do sinal, o que aumenta a taxa de erro de bits do enlace, reduzindo a confiabilidade. Denomina-se esta interfer?ncia de interfer?ncia intersimb?lica (ISI, do ingl?s intersymbol interference) porque, no sinal em banda-base no receptor, ocorre superposi??o dos s?mbolos da modula??o digital, resultando em falha na detec??o das palavras bin?rias associadas aos s?mbolos. Neste contexto, o processo de estima??o e compensa??o dos efeitos do canal desempenha um papel importante em um receptor de comunica??o wireless. A crescente demanda por sistemas de maiores capacidades de transmiss?o, robustez e menor complexidade computacional, tem impulsionado v?rias pesquisas no ?mbito dos algoritmos utilizados no processo de estima??o de canal. Nos ?ltimos anos a t?cnica de transmiss?o de dados atrav?s da Multiplexa??o por Divis?o Ortogonal de Frequ?ncia (OFDM) tem ganhado destaque por apresentar resist?ncia ? ISI, boa efici?ncia espectral e capacidade de transmiss?o de altas taxas de dados. OFDM ? uma t?cnica de modula??o por multiportadoras que consiste na divis?o da largura de banda total em subcanais menores, utilizando subportadoras ortogonais sobrepostas espectralmente. N?o obstante a sua robustez, faz-se ainda necess?rio aplicar t?cnicas de estima??o de canal no receptor OFDM, devido ao efeito de m?ltipercurso caracter?stico de um canal de comunica??o wireless. Neste trabalho ? proposto a implementa??o de um estimador de canal, baseado em uma rede neural com fun??o de base radial (RBF, do ingl?s Radial Basis Function) complexa. A rede proposta ? treinada a partir da resposta ao impulso do canal obtida atrav?s de portadoras piloto enviadas e conhecidas pelo receptor. Os resultados da simula??o mostram que a rede proposta obteve melhores resultados do que os estimadores cl?ssicos utilizados para estima??o de canal em sistemas OFDM.