Dissertations / Theses on the topic 'Neural forms'

Acetylcholinesterase [ACNE) and butyrylcholinesterase [EChE) were studied in amniotic fluid in relation to the detection of neural tube defects CNTD), and in rectal tissue in the diagnosis of Hirschsprung's disease. An automated assay is described for measurement of AChE and BChE activity in amniotic Fluid, and an increase in both is found in the presence of NTD. Analysis of AChE molecular forms by sucrose density sedimentation revealed three species with differing sedimentation coefficients and molecular masses: monomeric G1[4. OS, 78KOa), dimeric G2[5.5S, 126KOa) and tetrameric G4(10.35,256KDa). The tetramer, G4 is NTD specific and is largely responsible for the increase in activity seen in the quantitative assessment of'total AChE and for the abnormal band identifiable by polyacrylamide gel electrophoresis in pregnancies affected by NTO. Evidence is presented which indicates that G4 is a soluble species secreted from nerve trunks exposed as a result of the lesion. SChE activity, the likely source of which is fetal plasma is shown to be a less specific indicator of NTD. These results represent the first description of the structural molecular heterogeneity of AChE and SChE forms in amniotic fluid. AChE activity was measured in rectal biopsy specimens from 213 patients in whom a diagnosis of Hirschsprung's disease was suspected. The results from this, the largest study so far reported, indicate the value of AChE measurement in the detection of the disease. The molecular forms of AChE and SChE were investigated in resected bowel segments from patients with Hirschsprung's disease. Four species of AChE were identified: G1[3.55,74KOa), G2[S. OS, 131KDa), 64(9.23,275KOa] and the asymmetric form A12(16.83,811KDa). In all cases there was an increase (4-14 fold] in G4-AChE activity in the aganglionic cola-rectum. The evidence indicates that this is derived from hypertrophied nerve trunks present in the affected zone. The increase in G4-AChE was largely responsible for the increase in total AChE activity in rectal biopsy specimens from patients with Hirschsprung's disease. BChE molecular forms showed no consistent changes in Hirschsprung's disease. Characterisation of the molecular forms of AChE by gel filtration and with respect to their thermal stability, sensitivity to Triton X-100 and response to substrate inhibition is also investigated.

Quality by Design (QbD) is a systematic approach that has been recommended as suitable for the development of quality pharmaceutical products. The QbD approach commences with the definition of a quality target drug profile and predetermined objectives that are then used to direct the formulation development process with an emphasis on understanding the pharmaceutical science and manufacturing principles that apply to a product. The design space is directly linked to the use of QbD for formulation development and is a multidimensional combination and interaction of input variables and process parameters that have been demonstrated to provide an assurance of quality. The objective of these studies was to apply the principles of QbD as a framework for the optimisation of a sustained release (SR) formulation of salbutamol sulphate (SBS), and for the establishment of a design space using Response Surface Methodology (RSM) and Artificial Neural Networks (ANN). SBS is a short-acting ♭₂ agonist that is used for the management of asthma and chronic obstructive pulmonary disease (COPD). The use of a SR formulation of SBS may provide clinical benefits in the management of these respiratory disorders. Ashtalin®8 ER (Cipla Ltd., Mumbai, Maharashtra, India) was selected as a reference formulation for use in these studies. An Ishikawa or Cause and Effect diagram was used to determine the impact of formulation and process factors that have the potential to affect product quality. Key areas of concern that must be monitored include the raw materials, the manufacturing equipment and processes, and the analytical and assessment methods employed. The conditions in the laboratory and manufacturing processes were carefully monitored and recorded for any deviation from protocol, and equipment for assessment of dosage form performance, including dissolution equipment, balances and hardness testers, underwent regular maintenance. Preliminary studies to assess the potential utility of Methocel® Kl OOM, alone and in combination with other matrix forming polymers, revealed that the combination of this polymer with xanthan gum and Carbopol® has the potential to modulate the release of SBS at a specific rate, for a period of 12 hr. A central composite design using Methocel® KlOOM, xanthan gum, Carbopol® 974P and Surelease® as the granulating fluid was constructed to fully evaluate the impact of these formulation variables on the rate and extent of SBS release from manufactured formulations. The results revealed that although Methocel® KlOOM and xanthan gum had the greatest retardant effect on drug release, interactions between the polymers used in the study were also important determinants of the measureable responses. An ANN model was trained for optimisation using the data generated from a central composite study. The efficiency of the network was optimised by assessing the impact of the number of nodes in the hidden layer using a three layer Multi Layer Perceptron (MLP). The results revealed that a network with nine nodes in the hidden layer had the best predictive ability, suitable for application to formulation optimisation studies. Pharmaceutical optimisation was conducted using both the RSM and the trained ANN models. The results from the two optimisation procedures yielded two different formulation compositions that were subjected to in vitro dissolution testing using USP Apparatus 3. The results revealed that, although the formulation compositions that were derived from the optimisation procedures were different, both solutions gave reproducible results for which the dissolution profiles were indeed similar to that of the reference formulation. RSM and ANN were further investigated as possible means of establishing a design space for formulation compositions that would result in dosage forms that have similar in vitro release test profiles comparable to the reference product. Constraint plots were used to determine the bounds of the formulation variables that would result in the manufacture of dosage forms with the desired release profile. ANN simulations with hypothetical formulations that were generated within a small region of the experimental domain were investigated as a means of understanding the impact of varying the composition of the formulation on resultant dissolution profiles. Although both methods were suitable for the establishment of a design space, the use of ANN may be better suited for this purpose because of the manner in which ANN handles data. As more information about the behaviour of a formulation and its processes is generated during the product Iifecycle, ANN may be used to evaluate the impact of formulation and process variables on measureable responses. It is recommended that ANN may be suitable for the optimisation of pharmaceutical formulations and establishment of a design space in line with ICH Pharmaceutical Development [1], Quality Risk Management [2] and Pharmaceutical Quality Systems [3]

A idéia de capacitar uma máquina a reconhecer o ambiente em que atua tem motivado pesquisadores a investir esforços no estudo do mais complexo dos sentidos humanos, a visão. A visão é, antes de tudo, uma tarefa de representação e processamento de informações, sendo portanto adequada ao tratamento computacional. Visto que ainda não se possuem métodos que tenham resultados equivalentes ao que seria obtido com um usuário humano, tem-se estudado intensamente a utilização de feições para um melhor aproveitamento de seu potencial. Dentre estas feições, a forma de um objeto proporciona um poderoso indício de sua identidade e funcionalidade, podendo ser utilizada para seu reconhecimento. Isso distingue a forma de outras feições visuais elementares, como a cor, o movimento ou a textura, que, apesar de igualmente importantes, normalmente não revelam a identidade de um objeto. Assim sendo, a possibilidade de avaliar a robustez e a estabilidade de técnicas alternativas para a representação de forma é vital para prever o desempenho de cada técnica na presença de alguma incerteza ou discrepância. Neste trabalho, alguns descritores de forma descritos na literatura foram implementados e utilizados em estudos de caso para avaliar sua eficácia. Estes estudos de caso foram realizados utilizando-se caracteres, todavia, com finalidades bastante distintas. O primeiro estudo de caso é voltado para aplicações como a robótica móvel, com reconhecimento de comandos localizados no ambiente por parte do robô. Já o estudo de caso principal está direcionado para aplicações de reconhecimento de placas de automóveis, que poderia tanto ser utilizado para monitoramento e controle do fluxo de trânsito, quanto para controle de infrações. Muitas aplicações, incluindo aquelas que envolvem a recuperação e indexação de objetos visuais, são apropriadas para a utilização de feições de forma. Outra característica importante do presente trabalho é a de realçar que a seleção de um bom descritor reduz o esforço necessário na etapa de classificação, o qual é computacionalmente elevado.
The idea of enabling a machine to recognize the environment with which it interacts has motivated researchers to dedicate efforts in studying the most complex of the human senses: vision. Vision is essentially a task of information representation and processing, what makes it suitable for computational treatment. Given that currently there are no methods that perform equivalently to humans, the use of features has been intensively studied in order to improve the performance of the existing methods. Among these features, the shape of an object provides a powerful sign of its identity and functionality, what enables the exploitation of this feature with the purpose of recognition. This evidence distinguishes shape from other visual features, such as color, motion or texture, which, although equally important, normally do not reveal the identity of an object. As a result, the possibility of evaluating the robustness and stability of alternate techniques for shape representation is essential in order to measure the performance of each technique in the presence of uncertainty. In this work, some shape descriptors available in the literature were implemented and used in case studies aiming at evaluating their effectiveness. These case studies were carried out using characters, although, with very different purposes. The first case study is geared towards applications such as mobile robotics, where the robot recognizes commands available in the environment. The main case study is focused on applications of license plate recognition, which could be used both in situations of surveillance and traffic control and in situations of infraction. Many applications, including those that involve the search and indexing of visual objects, are suited for the use of shape features. Another important characteristic of this work is that it emphasizes that the selection of a good shape description reduces the effort during the classification step, which is computationally elevated.

Thesis (Ph. D. in Neuroscience)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2013.
This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Cataloged from student-submitted PDF version of thesis.
Includes bibliographical references.
Connectivity restricts and defines the information that a network can process. It is the substance of information processing that underlies the patterns of functional activity in the brain. By combining diffusion-weighted imaging or DWI, with fMRI, we are able to non-invasively measure connectivity and neural responses in the same individuals and directly relate these two measures to one another. In Chapter 2, I first establish the proof-of-principle that anatomical connectivity alone can predict neural responses in cortex, specifically of face-selectivity in the fusiform gyrus. I then extend this novel approach to the rest of the brain and test whether connectivity can accurately predict neural responses to various visual categories in Chapter 3. Finally, in Chapter 4, I compare and contrast the resulting models, which are essentially networks of connectivity that are functionally-relevant to each visual category, and demonstrate the type of knowledge that can be uncovered by directly integrating structure and function.
by David Eugene Osher.
Ph.D.in Neuroscience

Due to the multiplicity of computer-aided engineering applications present in industry today, interoperability between programs has become increasingly important. A survey conducted among top engineering companies found that 82% of respondents reported using 3 or more CAD formats during the design process. A 1999 study by the National Institute for Standards and Technology (NIST) estimated that inadequate interoperability between the OEM and its suppliers cost the US automotive industry over $1 billion per year, with the majority spent fixing data after translations. The Neutral Parametric Canonical Form (NPCF) prototype standard developed by the NSF Center for e-Design, BYU Site offers a solution to the translation problem by storing feature data in a CAD-neutral format to offer higher-fidelity parametric transfer between CAD systems. This research has focused on expanding the definitions of the NPCF to enforce data integrity and to support associativity between features to preserved design intent through the neutralization process. The NPCF data structure schema was defined to support associativity while maintaining data integrity. Neutral definitions of new features was added including multiple types of coordinate systems, planes and axes. Previously defined neutral features were expanded to support new functionality and the software architecture was redefined to support new CAD systems. Complex models have successfully been created and exchanged by multiple people in real-time to validated the approach of preserving associativity and support for a new CAD system, PTC Creo, was added.

Introdução: A hanseníase é a principal causa infecciosa de neuropatia periférica e consequentes incapacidades em todo o mundo. Seu diagnóstico, na maioria das vezes é simples, especialmente quando as clássicas lesões cutâneas estão presentes. No entanto, alguns pacientes apresentam apenas envolvimento neural (forma neural pura - PNL) transformando o seu diagnóstico em um grande desafio. Nesses casos, mesmo quando essa possibilidade é aventada, sua confirmação pode ser extremamente difícil e muitos pacientes só serão corretamente diagnosticados tardiamente, quando uma neuropatia grave e irreversível já está estabelecida. Objetivos: Analisar as características de uma série de pacientes com diagnóstico definitivo ou provável de PNL seguidos no setor de doenças neuromusculares e dermatologia no HCRP em um período de 12 anos e reconhecer o padrão de apresentação mais frequente, suas manifestações clínicas e o padrão eletroneuromiográfico. Métodos: Estudo retrospectivo, observacional, cujos critérios de inclusão foram: evidência clínica de comprometimento de nervos periféricos na ausência de lesões de pele. O diagnóstico definitivo foi estabelecido quando o Mycobacterium leprae foi identificado na biópsia de nervo, e provável quando um quadro clínico sugestivo foi associado a pelo menos um dos seguintes: anti PGL1 positivo, padrão sugestivo na biópsia (neurite granulomatosa epitelióide, infiltrado linfomomononuclear, fibrose) e/ou padrão eletroneuromiográfico sugestivo. Para avaliar a importância da duração da doença na apresentação clínica, foram considerados dois grupos de acordo com o tempo da doença, 12 meses ou menos (grupo 1) e mais de 12 meses (grupo 2). Foram comparados os sinais, os sintomas, a gravidade da doença e o padrão da EMG para delinear o quadro de apresentação. Resultados: Dos 34 pacientes incluídos no estudo, 7 tinham diagnóstico definitivo e 24 diagnóstico provável. Os sintomas de início mais frequentes foram alterações sensitivas (91,2%), em 70,6% dos casos iniciaram nos membros superiores, sendo o nervo ulnar o local mais frequente. O padrão de distribuição intradérmico exclusivo foi observado apenas no grupo 1. A alteração da sensibilidade vibratória (p=0,07), a presença de alterações motoras (p=0,03) e hipo ou areflexia em 1 ou mais nervos (p=0,03) foram mais frequentemente observadas no grupo 2. Os nervos sensitivos mais frequentemente envolvidos foram o ulnar e fibular superficial. O nervo motor mais frequentemente afetado foi o ulnar. O padrão eletroneuromiográfico mais frequente foi de uma neuropatia sensitivo motora assimétrica com reduções focais da velocidade de condução e franco predomínio sensitivo. Conclusões: A PNL se apresenta invariavelmente de maneira assimétrica e com franco predomínio sensitivo. Na maioria das vezes o início ocorre nos membros superiores, especificamente no território do nervo ulnar. Há uma predisposição ao acometimento das fibras finas nos estágios iniciais e com a evolução da doença as fibras grossas passam a também serem afetadas. Os nervos sensitivos mais frequentemente envolvidos são o ulnar seguido pelo fibular superficial.
Backgrounds: Leprosy is the main infectious cause of peripheral neuropathy and disabilities in the world. Its diagnosis is straightforward when the classical skin lesions are present. However, some patients present only neural involvement (pure neural form-PNL) turning its diagnosis on a great challenge. Additionally, even when this possibility is suspected, confirmation may be extremely difficult and many patients are only correctly diagnosed late on the clinical course of the disease when a severe and irreversible neuropathy is already established. Objectives: To review the characteristics of a series of PNL patients followed in our institution in the last 12 years and recognize the clinical manifestations. Methods: Inclusion criteria: Clinical evidence of peripheral nerve impairment with no skin lesions. PNL diagnose were classified as definitive when the Mycobacterium leprae was identified under nerve biopsy, and probable when a suggestive clinical picture was associated to at least one of the following: positive anti PGL1, suggestive pattern biopsy represented by the presence of epithelioid granulomatous neuritis, mononuclear cell endoneuritis and fibrose and/or an EMG pattern showing a predominantly sensory mononeuritis multiplex pattern. Exclusion criteria: Two patients were excluded because of associated diabetes mellitus, one because had CMT1A and another had HNPP. To evaluate the importance of disease duration in clinical presentation, we considered two groups according to the time course, first that disease duration of 12 or fewer months (group 1) and those with disease duration over 12 months (group 2). Results: We reviewed 34 patients with PNL, including 7 with a definite and 24 with probable diagnosis. The most common onset symptoms were sensory (91.2 %), in 70.6 % of cases symptoms started in the upper limbs, the ulnar nerve being the most frequent site. Intradermal pattern was observed only in group 1. Vibration was altered more frequent in group 2 (p=0.07), the presence of motor abnormalities (p = 0.03) and deep tendon reflexes reduced or absent in 1 or more nerves (p = 0.03) were more frequently observed in group 2. Sensory nerves most frequently involved were the ulnar and superficial peroneal. The motor nerve most often affected was the ulnar. The most frequent EMG pattern was an asymmetrical sensory and motor neuropathy with focal slowing of conduction velocity. Conclusions: PNL is an asymmetrical sensory or sensory motor neuropathy. Upper limbs are most frequent affected with frequent ulnar nerve territory involvement. Small fibers seem to be affected at early stages. Larger fibers are affected with disease progression. It is unclear whether the PNL represents a stage prior to the appearance of typical skin lesions or whether it represents a different and more aggressive leprosy type. Phenotype characterization from early signs and symptoms its a powerful tool to PNL early diagnosis.

Orientador: Caroline Joy Steel
Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia de Alimentos
Made available in DSpace on 2018-08-24T06:50:38Z (GMT). No. of bitstreams: 1 Marangoni_AndreLuis_D.pdf: 1214013 bytes, checksum: a252edf1c913d6c97e6f9530017bb9cd (MD5) Previous issue date: 2014
Resumo: A interesterificação é uma ferramenta fundamental para o desenvolvimento de gorduras "zero trans"; entretanto, comparada ao processo de hidrogenação, esta apresenta limitações, sobretudo no desenvolvimento de gorduras para uso em panificação. De acordo com a literatura, na produção de pão de forma, a gordura exerce diversas funções, como a lubrificação e o aumento da extensibilidade da massa, e o aumento do volume e do sabor do pão. A gordura afeta a textura, mantendo os pães macios por mais tempo; isto se deve possivelmente à sua interação com o amido da farinha, retardando o processo de retrogradação e, assim, estendendo a vida de prateleira do pão. O objetivo deste trabalho foi aplicar a tecnologia de Redes Neurais Artificiais (RNA) na formulação de gorduras "zero trans" à base de óleo de soja e gorduras interesterificadas de soja para facilitar o processo de formulação por blending, específicas para produtos de panificação, e determinar a influência das mesmas na qualidade dos pães de forma e nas interações entre as gorduras e o amido da farinha. Para tanto, foram produzidos pré-misturas e pães de forma com a adição de 4% de gordura. Como padrões, foram utilizadas gorduras comerciais, hidrogenada (GHS) e low trans (GLT), além de óleo de soja (OLS). Também foram utilizados os blends de gordura formulados através da RNA (BL1, BL2, BL3 e BL4). Para efeito de controle, foi produzido um pão sem adição de gordura (C). A análise farinográfica mostrou que a absorção de água (ABS) da farinha de trigo pura (59,0%) foi em média 6,5% maior que a das pré-misturas adicionadas de gordura. O tempo de desenvolvimento (Td) foi menor para as amostras GHS, GLT e BL4. A extensografia mostrou que, dentre todas as amostras, a BL4 foi a mais resistente (980 UE) e a menos extensível (114 mm). Isto provavelmente ocorreu devido ao menor teor de óleo de soja em sua constituição (54%), o que pode ter contribuído para uma massa de maior consistência. A análise dos pães produzidos revelou que apenas os volumes específicos das amostras OLS (3,46 mL/g) e BL4 (4,07 mL/g) diferiram significativamente entre si. A análise de firmeza dos pães mostrou que ao longo da estocagem houve diferença significativa entre a firmeza dos pães com gordura e a amostra controle (1005,75 gf), sendo este valor 13% superior ao da amostra GHS - a mais firme dentre os pães com adição de gordura. A uniformidade do miolo foi maior com a utilização de gordura. Nos pães controle (C), a porosidade (26,73%) foi quase 3 vezes superior ao das amostras com a adição dos blends. Os miolos dos pães BL1, BL2, BL3 e BL4 apresentaram alvéolos pequenos e espalhados mais uniformemente, quando comparados aos pães C, GHS, GLT e OLS. Quanto à umidade, os pães com gordura apresentaram um menor teor em relação ao da amostra controle (35%), pois as suas massas absorveram menos água durante a mistura. A análise térmica através de DSC sugeriu um efeito da gordura sobre o envelhecimento dos pães, uma vez que as variações de entalpia de retrogradação foram menores para os pães com gorduras. Os blends de gordura desenvolvidos usando a RNA e aqui utilizados, além do baixo teor de ácidos graxos trans (1,18% em média), apresentaram-se viáveis para aplicação em panificação, sobretudo o BL4
Abstract: Interesterification is a fundamental tool in the development of "zero trans" fats; however, when compared to the hydrogenation process, it presents limitations, especially when developing shortenings for bakery products. According to literature, in the production of pan bread, fat has several functions, such as lubrication and an increase in dough extensibility, and an increase in bread volume and flavor. Fat affects texture, maintaining breads soft for a longer period of time; this is possibly due to its interaction with starch in flour, retarding the retrogradation process and, thus, extending bread shelf-life. The aim of this study was to apply Artificial Neural Network (ANN) technology in the formulation of "zero trans" fats based on soybean oil and soybean interesterified fats to ease the formulation process through blending, for use in bakery products, and determine their influence on the quality of pan bread and on their interaction with starch in flour. For this, pre-mixes and breads with the addition of 4% fat were produced. As standards, commercial fats (hydrogenated soybean fat ¿ GHS and low trans fat ¿ GLT) were used, as well as soybean oil (OLS). The fat blends formulated using the ANN (BL1, BL2, BL3 e BL4) were also used. As control (C), bread without fat addition was prepared. The farinographic analysis showed that water absorption (ABS) of pure wheat flour (59.0%) was in average 6.5% higher than that of the pre-mixes of flour and fats. Dough development time (Td) was lower for the samples GHS, GLT and BL4. The extensographic analysis showed that, amongst all samples, BL4 showed the highest resistance to extension (980 EU) and the lowest extensibility (114 mm). This probably occurred due to the lower soybean oil content in its constitution (54%) that could have contributed to a more consistent dough. The analysis of the breads produced revealed that only the specific volumes of the samples OLS (3.46 mL/g) and BL4 (4.07 mL/g) differed significantly. Firmness analysis of breads showed that throughout the storage period studied there was a significant difference between the firmness of the breads with fats and the control sample (1005.75 gf), being this value 13% higher than that of GHS ¿ the firmest amongst samples with fat. Crumb uniformity was greater with the use of fat. In the control breads (C), porosity (26.73%) was almost 3 times greater than that of the samples with the addition of the blends. The crumbs of breads BL1, BL2, BL3 and BL4 presented small and more uniformly distributed alveoli, when compared to breads C, GHS, GLT and OLS. As to moisture content, breads with fat presented lower values when compared to the control sample (35%), as their doughs absorbed less water during mixing. Thermal analysis through DSC suggested an effect of fat on bread staling, once retrogradation enthalpy changes were lower for breads with fats. The fat blends developed using the ANN and used in this study, as well as having a low trans fatty acid content (1.18% in average), showed feasibility for application in pan bread, especially BL4
Doutorado
Tecnologia de Alimentos
Doutor em Tecnologia de Alimentos

Thesis (M.S.)--University of Missouri-Columbia, 2006.
Title from title screen of research.pdf file (viewed on December 22, 2006). The entire dissertation/thesis text is included in the research.pdf file; the official abstract appears in the short.pdf file (which also appears in the research.pdf); a non-technical general description, or public abstract, appears in the public.pdf file. "May 2006" Includes bibliographical references.

La transcription automatique du texte dans les documents manuscrits a de nombreuses applications, allant du traitement automatique des documents à leur indexation ou leur compréhension. L'une des approches les plus populaires de nos jours consiste à parcourir l'image d'une ligne de texte avec une fenêtre glissante, de laquelle un certain nombre de caractéristiques sont extraites, et modélisées par des Modèles de Markov Cachés (MMC). Quand ils sont associés à des réseaux de neurones, comme des Perceptrons Multi-Couches (PMC) ou Réseaux de Neurones Récurrents de type Longue Mémoire à Court Terme (RNR-LMCT), et à un modèle de langue, ces modèles produisent de bonnes transcriptions. D'autre part, dans de nombreuses applications d'apprentissage automatique, telles que la reconnaissance de la parole ou d'images, des réseaux de neurones profonds, comportant plusieurs couches cachées, ont récemment permis une réduction significative des taux d'erreur.Dans cette thèse, nous menons une étude poussée de différents aspects de modèles optiques basés sur des réseaux de neurones profonds dans le cadre de systèmes hybrides réseaux de neurones / MMC, dans le but de mieux comprendre et évaluer leur importance relative. Dans un premier temps, nous montrons que des réseaux de neurones profonds apportent des améliorations cohérentes et significatives par rapport à des réseaux ne comportant qu'une ou deux couches cachées, et ce quel que soit le type de réseau étudié, PMC ou RNR, et d'entrée du réseau, caractéristiques ou pixels. Nous montrons également que les réseaux de neurones utilisant les pixels directement ont des performances comparables à ceux utilisant des caractéristiques de plus haut niveau, et que la profondeur des réseaux est un élément important de la réduction de l'écart de performance entre ces deux types d'entrées, confirmant la théorie selon laquelle les réseaux profonds calculent des représentations pertinantes, de complexités croissantes, de leurs entrées, en apprenant les caractéristiques de façon automatique. Malgré la domination flagrante des RNR-LMCT dans les publications récentes en reconnaissance d'écriture manuscrite, nous montrons que des PMCs profonds atteignent des performances comparables. De plus, nous avons évalué plusieurs critères d'entrainement des réseaux. Avec un entrainement discriminant de séquences, nous reportons, pour des systèmes PMC/MMC, des améliorations comparables à celles observées en reconnaissance de la parole. Nous montrons également que la méthode de Classification Temporelle Connexionniste est particulièrement adaptée aux RNRs. Enfin, la technique du dropout a récemment été appliquée aux RNR. Nous avons testé son effet à différentes positions relatives aux connexions récurrentes des RNRs, et nous montrons l'importance du choix de ces positions.Nous avons mené nos expériences sur trois bases de données publiques, qui représentent deux langues (l'anglais et le français), et deux époques, en utilisant plusieurs types d'entrées pour les réseaux de neurones : des caractéristiques prédéfinies, et les simples valeurs de pixels. Nous avons validé notre approche en participant à la compétition HTRtS en 2014, où nous avons obtenu la deuxième place. Les résultats des systèmes présentés dans cette thèse, avec les deux types de réseaux de neurones et d'entrées, sont comparables à l'état de l'art sur les bases Rimes et IAM, et leur combinaison dépasse les meilleurs résultats publiés sur les trois bases considérées
The automatic transcription of text in handwritten documents has many applications, from automatic document processing, to indexing and document understanding. One of the most popular approaches nowadays consists in scanning the text line image with a sliding window, from which features are extracted, and modeled by Hidden Markov Models (HMMs). Associated with neural networks, such as Multi-Layer Perceptrons (MLPs) or Long Short-Term Memory Recurrent Neural Networks (LSTM-RNNs), and with a language model, these models yield good transcriptions. On the other hand, in many machine learning applications, including speech recognition and computer vision, deep neural networks consisting of several hidden layers recently produced a significant reduction of error rates. In this thesis, we have conducted a thorough study of different aspects of optical models based on deep neural networks in the hybrid neural network / HMM scheme, in order to better understand and evaluate their relative importance. First, we show that deep neural networks produce consistent and significant improvements over networks with one or two hidden layers, independently of the kind of neural network, MLP or RNN, and of input, handcrafted features or pixels. Then, we show that deep neural networks with pixel inputs compete with those using handcrafted features, and that depth plays an important role in the reduction of the performance gap between the two kinds of inputs, supporting the idea that deep neural networks effectively build hierarchical and relevant representations of their inputs, and that features are automatically learnt on the way. Despite the dominance of LSTM-RNNs in the recent literature of handwriting recognition, we show that deep MLPs achieve comparable results. Moreover, we evaluated different training criteria. With sequence-discriminative training, we report similar improvements for MLP/HMMs as those observed in speech recognition. We also show how the Connectionist Temporal Classification framework is especially suited to RNNs. Finally, the novel dropout technique to regularize neural networks was recently applied to LSTM-RNNs. We tested its effect at different positions in LSTM-RNNs, thus extending previous works, and we show that its relative position to the recurrent connections is important. We conducted the experiments on three public databases, representing two languages (English and French) and two epochs, using different kinds of neural network inputs: handcrafted features and pixels. We validated our approach by taking part to the HTRtS contest in 2014. The results of the final systems presented in this thesis, namely MLPs and RNNs, with handcrafted feature or pixel inputs, are comparable to the state-of-the-art on Rimes and IAM. Moreover, the combination of these systems outperformed all published results on the considered databases

Made available in DSpace on 2014-06-11T19:22:34Z (GMT). No. of bitstreams: 0 Previous issue date: 2010-08-20Bitstream added on 2014-06-13T20:48:00Z : No. of bitstreams: 1 amaral_fcl_me_bauru.pdf: 878445 bytes, checksum: 4f078f3133073c4bc5a0882fbff35667 (MD5)
Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)
A proposta deste trabalho consiste em investigar o comportamento de transformadores de distribuição de 25 KVA e de 45 KVA submetidos a ensaios de impulsos atmosféricos. Essa investigação consistiu da verificação da influência dos valores de tempos de frente e de cauda, da forma de onda do impulso, na amplitude das correntes produzidas nos enrolamentos do transformador durante o ensaio. Tais correntes são usualmente empregadas para avaliar o desempenho elétrico dos transformadores e, nessa dissertação, foram usadas para o treinamento e tese de Redes Neurais Artificiais desenvolvidas como ferramentas inteligentes computacionais. Neste contexto, o desempenho de duas Redes Neurais foi avaliado. A primeira rede usou como variável de entrada, os valores de tempo de frente, de caula e da tensão máxima (crista) e, como saída, a corrente máxima no transformador. Na segunda rede neural, a entrada correspondente ao valor da tensão máxima, da primeira rede neural, é substituída pelo valor da taxa de crescimento da tensão. Com base nos resultados obtidos, pode-se verificar que, para determinados valores de tempos de frente e de cauda, a amplitude da corrente máxima, aumenta ou diminui, apresentando um comportamento não-linear. A utilização das Redes Neurais desenvolvidas neste trabalho poderá auxiliar na escolha das características das formas de onda de impulso que tornem mais sensíveis os ensaios de impulsos atmosféricos em transformadores de distribuição. O objetivo é que esse aumento da sensibilidade do ensaio minimize o empirismo e erros de avaliação, contribuindo para tornar mínima a taxa de falha em transformadores
The proposal of this work is to investigate and to analyze the behavior of 25 kVA and 45 KVA distribution transformers under impulses/surge tests. This research consited in a verification of the influence of front and and tail time values, from surge waverfom, in the current magnitude produced in transformer windings during the test. Such currents are usually employed to evalute the perfomers and, in this dissertation, were used for the training and test the Artificial Neural Networks, developed as intelligent computational tools. In this context, the performance of two Neural Networks was evaluated. This first network has used as entry variables: front and tail time and the maximun voltage (crest) values and, as an exist, the maximum current in transformer. In the second neural network, the entry corresponding to the maximum voltage value, from the first neural network, is replaced by the value of the rate of growth of the voltage. Based on the obtained results, one may find out, for certain values front and tail times values, the amplitude of maximum current, increases or decreases, presenting a non-linear behavior. The use of Neural Networks developed in this work can help someone to choose the best impulse waveform characteristics which make the impulse test in distribution transformers more sinsitive. The objective is that the rising the test sensitivity will minimize the empiricism and errors of assessment, helping to reduce the failure rate in transformers

The challenge of interoperability is to retain model integrity when different software applications exchange and interpret model data. Transferring CAD data between heterogeneous CAD systems is a challenge because of differences in feature representation. A study by the National Institute for Standards and Technology (NIST) performed in 1999 made a conservative estimate that inadequate interoperability in the automotive industry costs them $1 billion per year. One critical part of eliminating the high costs due to poor interoperability is a neutral format between heterogeneous CAD systems. An effective neutral CAD format should include a current-state data store, be associative, include the union of CAD features across an arbitrary number of CAD systems, maintain design history, maintain referential integrity, and support multi-user collaboration. This research has focused on extending an existing synchronous collaborative CAD software tool to allow for a neutral, current-state data store. This has been accomplished by creating a Neutral Parametric Canonical Form (NPCF) which defines the neutral data structure for many basic CAD features to enable translation between heterogeneous CAD systems. The initial architecture developed begins to define a new standard for storing CAD features neutrally. The NPCF's for many features have been implemented in a multi-user interoperability program and work between NX and CATIA CAD systems. The 2D point, 2D line, 2D arc, 2D circle, 2D spline, 3D point, extrude, and revolve NPCF's will be specifically defined. Complex models have successfully been modeled and exchanged in real time and have validated the NPCF approach. Multiple users can be in the same part at the same time in different CAD systems and create and update models in real time.

Submitted by Angelo Silva (asilva@icict.fiocruz.br) on 2016-07-13T18:45:41Z No. of bitstreams: 2 license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5) 71837.pdf: 4020666 bytes, checksum: d3470ad1e5a8fb4a1a611d711dc45510 (MD5)
Approved for entry into archive by Anderson Silva (avargas@icict.fiocruz.br) on 2016-07-25T13:25:23Z (GMT) No. of bitstreams: 2 71837.pdf: 4020666 bytes, checksum: d3470ad1e5a8fb4a1a611d711dc45510 (MD5) license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5)
Made available in DSpace on 2016-07-25T13:25:23Z (GMT). No. of bitstreams: 2 71837.pdf: 4020666 bytes, checksum: d3470ad1e5a8fb4a1a611d711dc45510 (MD5) license.txt: 1748 bytes, checksum: 8a4605be74aa9ea9d79846c1fba20a33 (MD5) Previous issue date: 2016
Fundação Oswaldo Cruz. Instituto Oswaldo Cruz. Rio de Janeiro, RJ, Brasil
A forma neural pura da hanseníase (FNP) está caracterizada pela presença da neuropatia periférica na ausência de lesões dermatológicas. O diagnóstico se baseia na análise histopatológica do nervo periférico em conjunto com os dados clínicos e a eletroneuromiografia. No entanto os achados histopatológicos muitas vezes são inespecíficos e o Mycobacterium leprae raramente é detectado. Diferente da biópsia de nervo, a biópsia cutânea pode ser realizada em unidades básicas de saúde visto se tratar de um exame pouco invasivo e de fácil realização. O objetivo deste estudo foi verificar através da análise histopatológica básica, o valor das biópsias cutâneas como ferramenta para o diagnóstico de pacientes com FNP. Cinquenta fragmentos cutâneos e de nervos periféricos correspondentes de pacientes com FNP e 50 fragmentos cutâneos de pacientes com neuropatias não hansênicas foram examinados através das colorações de hematoxilina&eosina, tricrômico de Gomori e Wade. Nossos resultados indicam que filetes nervosos foram observados em 91% das amostras cutâneas do total de pacientes com neuropatias periféricas. Destes, 33% pertenciam a pacientes com FNP e apresentaram alguma alteração morfológica nos nervos. As características confirmativas (presença de bacilo álcool-ácido resistente- BAAR) ou sugestivas de hanseníase (granuloma epitelióide, inflamação linfo-histiocítica endoneural e fibrose endoneural) foram observadas em 28% das amostras cutâneas. A fibrose foi a principal alteração histopatológica (14/50) e mesmo quando analisada individualmente pode ser sugestiva da doença. A formação de microfascículos foi observado em 4% das amostras sugerindo o papel da biópsia cutânea na identificação do processo regenerativo do nervo. Em 34% dos fragmentos de pele os filetes nervosos estavam normais Quando as amostras cutâneas foram comparadas com as amostras dos nervos periféricos correspondentes, houve concordância em 86% quanto à ausência do BAAR. A inflamação endoneural e o granuloma epitelióide estiveram mais presentes no nervo do que na pele, não sendo um bom parâmetro de análise quando observados individualmente na pele. Outras alterações, consideradas inespecíficas, estiveram presentes em 38% das amostras cutâneas dos pacientes com FNP, como a laminação do perineuro e a proliferação das células de Schwann. O valor preditivo positivo do fragmento cutâneo coletado por procedimento de biópsia para o diagnóstico da FNP foi de 68,2%. Podemos concluir que a análise morfológica detalhada dos filetes nervosos dos fragmentos cutâneos coletados no dermátomo do nervo afetado pode ser uma importante ferramenta de auxílio ao diagnóstico da FNP quando analisadas por patologistas experientes nos centros de referência para o tratamento da hanseníase
Abstract: Pure neural leprosy (PNL), characterized by peripheral neuropathy in the absence of dermatological alterations, is diagnosed based on the nerve biopsy results and the clinic and eletroneuromyograph data. However, histological findings are nonspecific and Mycobacterium leprae detection is rare. Unlike nerve biopsy, a skin biopsy may be performed in basic health units since it is minimally invasive and easy access procedure. The aim of this study was to analyze the value of dermal nerves in routine histopathology of skin samples as a tool for diagnosis. Fifty skin and nerves samples from patients with PNL and fifty skin samples from patients with non-leprosy peripheral neuropathies were examined with hematoxilin&eosin, Gomori\2019s trichrome and Wade stain. Our results showed that nerves are seen in 91% of skin fragments from pacients with leprosy and others neuropathies. Thirty-three per cent were from PNL and had morfological alterations. Confirmatory findings such as acid-fast bacilli (AFB) and other strong sugestive leprosy findings were observed in 28% of the skin samples. Endoneural fibrosis was the major histopathologic alteration (14/50) and can be considered sugestive of leprosy even when seen alone Microfasciculation was observed in 4% of skin biopsies sugesting it\2019s value as a tool for identification of regenerative nerve process. Thirty-four per cent of the skin fragments had normal appearance. When the skin fragments were compared with the corresponding peripheral nerves, there was agreement in 86% regarding the absence of AFB. Endoneurial inflamatory and epithelioid granuloma were more frequent in nerves than in skin fragments. Unspecific alterations such as perineural thickening and Schwann cell proliferation were showed in 38% of skin fragments. The positive predictive value for diagnosis of PNL were 68,2% in skin fragments. We conclude that histopathologic analysis of skin fragments can be useful as a diagnostic tool for PNL when analized by experienced pathologists in referral centers for leprosy treatment

For decades electrophysiologists have recorded and characterized the biophysical properties of a rich diversity of neuron types. This diversity of neuron types is critical for generating functionally important patterns of brain activity and implementing neural computations. In this thesis, I developed computational methods towards quantifying neuron diversity and applied these methods for understanding the functional implications of within-type neuron variability and across-type neuron diversity. First, I developed a means for defining the functional role of differences among neurons of the same type. Namely, I adapted statistical neuron models, termed generalized linear models, to precisely capture how the membranes of individual olfactory bulb mitral cells transform afferent stimuli to spiking responses. I then used computational simulations to construct virtual populations of biophysically variable mitral cells to study the functional implications of within-type neuron variability. I demonstrate that an intermediate amount of intrinsic variability enhances coding of noisy afferent stimuli by groups of biophysically variable mitral cells. These results suggest that within-type neuron variability, long considered to be a disadvantageous consequence of biological imprecision, may serve a functional role in the brain. Second, I developed a methodology for quantifying the rich electrophysiological diversity across the majority of the neuron types throughout the mammalian brain. Using semi-automated text-mining, I built a database, Neuro- Electro, of neuron type specific biophysical properties extracted from the primary research literature. This data is available at http://neuroelectro.org, which provides a publicly accessible interface where this information can be viewed. Though the extracted physiological data is highly variable across studies, I demonstrate that knowledge of article-specific experimental conditions can significantly explain the observed variance. By applying simple analyses to the dataset, I find that there exist 5-7 major neuron super-classes which segregate on the basis of known functional roles. Moreover, by integrating the NeuroElectro dataset with brain-wide gene expression data from the Allen Brain Atlas, I show that biophysically-based neuron classes correlate highly with patterns of gene expression among voltage gated ion channels and neurotransmitters. Furthermore, this work lays the conceptual and methodological foundations for substantially enhanced data sharing in neurophysiological investigations in the future.

[ES] El problema conocido como de secuencia a secuencia consiste en transformar una secuencia de entrada en una secuencia de salida. Bajo esta perspectiva se puede atacar una amplia cantidad de problemas, entre los cuales destacan la traducción automática o la descripción automática de objetos multimedia. La aplicación de redes neuronales profundas ha revolucionado esta disciplina, y se han logrado avances notables. Pero los sistemas automáticos todavía producen predicciones que distan mucho de ser perfectas. Para obtener predicciones de gran calidad, los sistemas automáticos se utilizan bajo la supervisión de un humano, quien corrige los errores. Esta tesis se centra principalmente en el problema de la traducción del lenguaje natural, usando modelos enteramente neuronales. Nuestro objetivo es desarrollar sistemas de traducción neuronal más eficientes. asentándonos sobre dos pilares fundamentales: cómo utilizar el sistema de una forma más eficiente y cómo aprovechar datos generados durante la fase de explotación del mismo. En el primer caso, aplicamos el marco teórico conocido como predicción interactiva a la traducción automática neuronal. Este proceso consiste en integrar usuario y sistema en un proceso de corrección cooperativo, con el objetivo de reducir el esfuerzo humano empleado en obtener traducciones de alta calidad. Desarrollamos distintos protocolos de interacción para dicha tecnología, aplicando interacción basada en prefijos y en segmentos, implementados modificando el proceso de búsqueda del sistema. Además, ideamos mecanismos para obtener una interacción con el sistema más precisa, manteniendo la velocidad de generación del mismo. Llevamos a cabo una extensa experimentación, que muestra el potencial de estas técnicas: superamos el estado del arte anterior por un gran margen y observamos que nuestros sistemas reaccionan mejor a las interacciones humanas. A continuación, estudiamos cómo mejorar un sistema neuronal mediante los datos generados como subproducto de este proceso de corrección. Para ello, nos basamos en dos paradigmas del aprendizaje automático: el aprendizaje muestra a muestra y el aprendizaje activo. En el primer caso, el sistema se actualiza inmediatamente después de que el usuario corrige una frase, aprendiendo de una manera continua a partir de correcciones, evitando cometer errores previos y especializándose en un usuario o dominio concretos. Evaluamos estos sistemas en una gran cantidad de situaciones y dominios diferentes, que demuestran el potencial que tienen los sistemas adaptativos. También llevamos a cabo una evaluación humana, con traductores profesionales. Éstos quedaron muy satisfechos con el sistema adaptativo. Además, fueron más eficientes cuando lo usaron, comparados con un sistema estático. El segundo paradigma lo aplicamos en un escenario en el que se deban traducir grandes cantidades de frases, siendo inviable la supervisión de todas. El sistema selecciona aquellas muestras que vale la pena supervisar, traduciendo el resto automáticamente. Aplicando este protocolo, redujimos de aproximadamente un cuarto el esfuerzo humano necesario para llegar a cierta calidad de traducción. Finalmente, atacamos el complejo problema de la descripción de objetos multimedia. Este problema consiste en describir en lenguaje natural un objeto visual, una imagen o un vídeo. Comenzamos con la tarea de descripción de vídeos pertenecientes a un dominio general. A continuación, nos movemos a un caso más específico: la descripción de eventos a partir de imágenes egocéntricas, capturadas a lo largo de un día. Buscamos extraer relaciones entre eventos para generar descripciones más informadas, desarrollando un sistema capaz de analizar un mayor contexto. El modelo con contexto extendido genera descripciones de mayor calidad que un modelo básico. Por último, aplicamos la predicción interactiva a estas tareas multimedia, disminuyendo el esfuerzo necesa
[CAT] El problema conegut com a de seqüència a seqüència consisteix en transformar una seqüència d'entrada en una seqüència d'eixida. Seguint aquesta perspectiva, es pot atacar una àmplia quantitat de problemes, entre els quals destaquen la traducció automàtica, el reconeixement automàtic de la parla o la descripció automàtica d'objectes multimèdia. L'aplicació de xarxes neuronals profundes ha revolucionat aquesta disciplina, i s'han aconseguit progressos notables. Però els sistemes automàtics encara produeixen prediccions que disten molt de ser perfectes. Per a obtindre prediccions de gran qualitat, els sistemes automàtics són utilitzats amb la supervisió d'un humà, qui corregeix els errors. Aquesta tesi se centra principalment en el problema de la traducció de llenguatge natural, el qual s'ataca emprant models enterament neuronals. El nostre objectiu principal és desenvolupar sistemes més eficients. Per a aquesta tasca, les nostres contribucions s'assenten sobre dos pilars fonamentals: com utilitzar el sistema d'una manera més eficient i com aprofitar dades generades durant la fase d'explotació d'aquest. En el primer cas, apliquem el marc teòric conegut com a predicció interactiva a la traducció automàtica neuronal. Aquest procés consisteix en integrar usuari i sistema en un procés de correcció cooperatiu, amb l'objectiu de reduir l'esforç humà emprat per obtindre traduccions d'alta qualitat. Desenvolupem diferents protocols d'interacció per a aquesta tecnologia, aplicant interacció basada en prefixos i en segments, implementats modificant el procés de cerca del sistema. A més a més, busquem mecanismes per a obtindre una interacció amb el sistema més precisa, mantenint la velocitat de generació. Duem a terme una extensa experimentació, que mostra el potencial d'aquestes tècniques: superem l'estat de l'art anterior per un gran marge i observem que els nostres sistemes reaccionen millor a les interacciones humanes. A continuació, estudiem com millorar un sistema neuronal mitjançant les dades generades com a subproducte d'aquest procés de correcció. Per a això, ens basem en dos paradigmes de l'aprenentatge automàtic: l'aprenentatge mostra a mostra i l'aprenentatge actiu. En el primer cas, el sistema s'actualitza immediatament després que l'usuari corregeix una frase. Per tant, el sistema aprén d'una manera contínua a partir de correccions, evitant cometre errors previs i especialitzant-se en un usuari o domini concrets. Avaluem aquests sistemes en una gran quantitat de situacions i per a dominis diferents, que demostren el potencial que tenen els sistemes adaptatius. També duem a terme una avaluació amb traductors professionals, qui varen quedar molt satisfets amb el sistema adaptatiu. A més, van ser més eficients quan ho van usar, si ho comparem amb el sistema estàtic. Pel que fa al segon paradigma, l'apliquem per a l'escenari en el qual han de traduir-se grans quantitats de frases, i la supervisió de totes elles és inviable. En aquest cas, el sistema selecciona les mostres que paga la pena supervisar, traduint la resta automàticament. Aplicant aquest protocol, reduírem en aproximadament un quart l'esforç necessari per a arribar a certa qualitat de traducció. Finalment, ataquem el complex problema de la descripció d'objectes multimèdia. Aquest problema consisteix en descriure, en llenguatge natural, un objecte visual, una imatge o un vídeo. Comencem amb la tasca de descripció de vídeos d'un domini general. A continuació, ens movem a un cas més específic: la descripció d''esdeveniments a partir d'imatges egocèntriques, capturades al llarg d'un dia. Busquem extraure relacions entre ells per a generar descripcions més informades, desenvolupant un sistema capaç d'analitzar un major context. El model amb context estés genera descripcions de major qualitat que el model bàsic. Finalment, apliquem la predicció interactiva a aquestes tasques multimèdia, di
[EN] The sequence-to-sequence problem consists in transforming an input sequence into an output sequence. A variety of problems can be posed in these terms, including machine translation, speech recognition or multimedia captioning. In the last years, the application of deep neural networks has revolutionized these fields, achieving impressive advances. However and despite the improvements, the output of the automatic systems is still far to be perfect. For achieving high-quality predictions, fully-automatic systems require to be supervised by a human agent, who corrects the errors. This is a common procedure in the translation industry. This thesis is mainly framed into the machine translation problem, tackled using fully neural systems. Our main objective is to develop more efficient neural machine translation systems, that allow for a more productive usage and deployment of the technology. To this end, we base our contributions on two main cornerstones: how to better use of the system and how to better leverage the data generated along its usage. First, we apply the so-called interactive-predictive framework to neural machine translation. This embeds the human agent and the system into a cooperative correction process, that seeks to reduce the human effort spent for obtaining high-quality translations. We develop different interactive protocols for the neural machine translation technology, namely, a prefix-based and a segment-based protocols. They are implemented by modifying the search space of the model. Moreover, we introduce mechanisms for achieving a fine-grained interaction while maintaining the decoding speed of the system. We carried out a wide experimentation that shows the potential of our contributions. The previous state of the art is overcame by a large margin and the current systems are able to react better to the human interactions. Next, we study how to improve a neural system using the data generated as a byproduct of this correction process. To this end, we rely on two main learning paradigms: online and active learning. Under the first one, the system is updated on the fly, as soon as a sentence is corrected. Hence, the system is continuously learning from the corrections, avoiding previous errors and specializing towards a given user or domain. A large experimentation stressed the adaptive systems under different conditions and domains, demonstrating the capabilities of adaptive systems. Moreover, we also carried out a human evaluation of the system, involving professional users. They were very pleased with the adaptive system, and worked more efficiently using it. The second paradigm, active learning, is devised for the translation of huge amounts of data, that are infeasible to being completely supervised. In this scenario, the system selects samples that are worth to be supervised, and leaves the rest automatically translated. Applying this framework, we obtained reductions of approximately a quarter of the effort required for reaching a desired translation quality. The neural approach also obtained large improvements compared with previous translation technologies. Finally, we address another challenging problem: visual captioning. It consists in generating a description in natural language from a visual object, namely an image or a video. We follow the sequence-to-sequence framework, under a a multimodal perspective. We start by tackling the task of generating captions of videos from a general domain. Next, we move on to a more specific case: describing events from egocentric images, acquired along the day. Since these events are consecutive, we aim to extract inter-eventual relationships, for generating more informed captions. The context-aware model improved the generation quality with respect to a regular one. As final point, we apply the intractive-predictive protocol to these multimodal captioning systems, reducing the effort required for correcting the outputs.
Section 5.4 describes an user evaluation of an adaptive translation system. This was done in collaboration with Miguel Domingo and the company Pangeanic, with funding from the Spanish Center for Technological and Industrial Development (Centro para el Desarrollo Tecnológico Industrial). [...] Most of Chapter 6 is the result of a collaboration with Marc Bolaños, supervised by Prof. Petia Radeva, from Universitat de Barcelona/CVC. This collaboration was supported by the R-MIPRCV network, under grant TIN2014-54728-REDC.
Peris Abril, Á. (2019). Interactivity, Adaptation and Multimodality in Neural Sequence-to-sequence Learning [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/134058
TESIS

The action of postponing an intended plan is often referred to as procrastination. Research on procrastination generally views the phenomenon as a form of self-regulation failure. Self-regulation refers to the conscious and non-conscious processes that enable individuals to guide their thoughts, feelings, and behaviors purposefully. Research indicates correlations between self-regulation and executive functions providing a fruitful integration. From a neuroscientific perspective, this integration generally associates the prefrontal cortex with top-down control whenever successful self-regulation is achieved. On the contrary, self-regulation failure appears to involve a bottom-up control, in which subcortical regions have greater influence on behavioral outcomes. Subcortical regions involved in emotional and rewarding processes, such as the amygdala and nucleus accumbens appears to lie at the coreof self-regulation failure, whereas cortical executive functions of regulating emotion and impulsive behaviors may contribute to successful self-regulation, thus overcoming procrastination. This thesis aims to obtain a deeper understanding of the mechanisms of procrastination, specifically investigating self-regulation failure and its relationship with executive functions and the neural underpinnings of self-regulation.

The neuronal ceroid lipofuscinoses (NCLs) are a group of inherited lysosomal storage disorders that cause profound neurodegeneration and predominantly affect children. Currently, there are no effective treatments available for any form of NCL. This thesis focuses on novel applications of gene therapy for CLN1 disease, which is caused by a deficiency in the lysosomal enzyme palmitoyl protein thioesterase-1 (PPT1) and CLN3 disease, which results from mutations in CLN3, which encodes a lysosomal transmembrane protein of unknown function. Therapy for transmembrane protein-deficient NCLs is complicated because uptake of therapeutically-delivered enzyme or “cross-correction” is not possible. Gene therapy could theoretically correct the CLN3 defect, but would require widespread transduction, and overexpression of CLN3 may have safety concerns. We used AAV2/9 vectors to deliver either mouse or human Cln3/CLN3 to the brains of wild- type and Cln3Adeficient mice, and assessed the short-and long-term response. Although a neuroinflammatory response to Cln3 over-expression was initially seen, this declined over time, and no obvious toxicity was associated with this overexpression. Compared to the brain, little is known about NCL systemic pathology, but we show that both Cln1A and Cln3Adeficient mice display functional and pathological phenotypes including heart rate abnormalities, sinoatrial node pathology, left ventricular hypertrophy and cardiac remodeling, thus providing a more complete picture of NCL pathology and where to target therapy. Although gene therapy has been partly successful for treating murine CLN1 disease, it has shown limited ability to extend lifespan. One way to enhance this efficacy would be to combine CNS gene therapy with treating systemic pathology, and we used neonatal intracranial AAV2/9Amediated gene therapy combined with enzyme replacement therapy (ERT) to treat the brain and body, respectively. While gene therapy alone significantly improved neuropathological changes, ERT conferred little additional benefit, and appeared to negate the positive effects of gene therapy.

The measurement of the π0 electromagnetic transition form factor (TFF) slope a is performed in the time-like region of momentum transfer using a sample of 1.1 x 106 π0→ e+e-y Dalitz decay collected at the NA62-RK experiment in 2007. The event selection, the fit procedure and the study of the systematic effects are presented. The final result obtained a = (3.68 ± 0.51stat ± 0.25syst) X 10- 2 is the most precise to date and represents the first evidence of a non-zero π0 TFF slope with more than 3σ. The NA62 experiment based at the CERN SPS is currently taking data and aims at measuring the branching fraction of the K→ πvv ultra-rare decay with 10% precision and less than 10% background. A complex trigger and data acquisition system is in place to record the data collected by the various detectors in use to reach this goal. The Run Control system of the experiment is meant to supervise and control them in a simple transparent way. The choices made to address the requirements for the system and the most important aspects of its implementation are discussed.

Ces travaux de thèse, réalisés dans une équipe concevant des circuits analogiques neuromimétiques suivant le modèle d’Hodgkin-Huxley, concernent la modélisation de neurones biologiques, plus précisément, l’estimation des paramètres de modèles de neurones. Une première partie de ce manuscrit s’attache à faire le lien entre la modélisation neuronale et l’optimisation. L’accent est mis sur le modèle d’Hodgkin- Huxley pour lequel il existait déjà une méthode d’extraction des paramètres associée à une technique de mesures électrophysiologiques (le voltage-clamp) mais dont les approximations successives rendaient impossible la détermination précise de certains paramètres. Nous proposons dans une seconde partie une méthode alternative d’estimation des paramètres du modèle d’Hodgkin-Huxley s’appuyant sur l’algorithme d’évolution différentielle et qui pallie les limitations de la méthode classique. Cette alternative permet d’estimer conjointement tous les paramètres d’un même canal ionique. Le troisième chapitre est divisé en trois sections. Dans les deux premières, nous appliquons notre nouvelle technique à l’estimation des paramètres du même modèle à partir de données biologiques, puis développons un protocole automatisé de réglage de circuits neuromimétiques, canal ionique par canal ionique. La troisième section présente une méthode d’estimation des paramètres à partir d’enregistrements de la tension de membrane d’un neurone, données dont l’acquisition est plus aisée que celle des courants ioniques. Le quatrième et dernier chapitre, quant à lui, est une ouverture vers l’utilisation de petits réseaux d’une centaine de neurones électroniques : nous réalisons une étude logicielle de l’influence des propriétés intrinsèques de la cellule sur le comportement global du réseau dans le cadre des oscillations gamma
These works, which were conducted in a research group designing neuromimetic integrated circuits based on the Hodgkin-Huxley model, deal with the parameter estimation of biological neuron models. The first part of the manuscript tries to bridge the gap between neuron modeling and optimization. We focus our interest on the Hodgkin-Huxley model because it is used in the group. There already existed an estimation method associated to the voltage-clamp technique. Nevertheless, this classical estimation method does not allow to extract precisely all parameters of the model, so in the second part, we propose an alternative method to jointly estimate all parameters of one ionic channel avoiding the usual approximations. This method is based on the differential evolution algorithm. The third chaper is divided into three sections : the first two sections present the application of our new estimation method to two different problems, model fitting from biological data and development of an automated tuning of neuromimetic chips. In the third section, we propose an estimation technique using only membrane voltage recordings – easier to mesure than ionic currents. Finally, the fourth and last chapter is a theoretical study preparing the implementation of small neural networks on neuromimetic chips. More specifically, we try to study the influence of cellular intrinsic properties on the global behavior of a neural network in the context of gamma oscillations

The neuronal ceroid lipofuscinosis (NCLs, Batten Disease) are inherited, fatal neurodegenerative disorders of childhood. In all forms of NCL, astrocyte activation occurs early in the disease and precedes neuronal loss. However, in the most common juvenile form (JNCL), which is caused by a mutation in the Cln3 gene, this astrocyte response appears to be compromised. Since astrocytes are crucial for the functioning and survival of neurons, and emerging evidence highlights the pivotal role that reactive astrocytosis plays in the pathogenesis of CNS diseases, any deficits in the biology of these cells could significantly impact neuronal health. In order to study the functioning of JNCL astrocytes, these cells were isolated from a well-characterised mouse model of the disease, Cln3 deficient mice (Cin3-l- mice), and their basic biology characterised. These studies revealed that Cln3-l- astrocytes have a disrupted actin and intermediate filament cytoskeleton. Possibly due to these defects, Cln3-l- astrocytes have an attenuated ability to response to an activation stimulus, just as observed in vivo, and to divide and migrate. They also display pronounced defects in their ability to take-up glutamate and to secrete a range of proteins, including cytokines, neuroprotective factors and the anti-oxidant glutathione, that become even more evident upon stimulation. Additionally, their impaired calcium signalling suggests that communication might be altered in these cells. Most importantly, using a co-culture system, these Cln3-l- glia were shown to negatively impact the health of both Cln3-l- and wild-type neurons, with the mutant neurons being the most severely affected, probably because of their own compromised biology. This includes a reduction in neurite complexity and displacement of the axon initial segment (AIS), which modulates neuronal excitability and the initiation of axon potentials. Thus, these data show, for the first time, that JNCL astrocytes are functionally compromised and might play an active role in the neurodegeneration observed in JNCL. Further, this information raises the possibility that, in future, astrocytes should be considered as targets for therapeutic interventions.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior
O diagnóstico da hanseníase neural pura baseia-se em dados clínicos e laboratoriais do paciente, incluindo a histopatologia de espécimes de biópsia de nervo e detecção de DNA de Mycobacterium leprae (M. leprae) pelo PCR. Como o exame histopatológico e a técnica PCR podem não ser suficientes para confirmar o diagnóstico, a imunomarcação de lipoarabinomanana (LAM) e/ou Glicolipídio fenólico 1 (PGL1) - componentes de parede celular de M. leprae foi utilizada na primeira etapa deste estudo, na tentativa de detectar qualquer presença vestigial do M. leprae em amostras de nervo sem bacilos. Além disso, sabe-se que a lesão do nervo na hanseníase pode diretamente ser induzida pelo M. leprae nos estágios iniciais da infecção, no entanto, os mecanismos imunomediados adicionam severidade ao comprometimento da função neural em períodos sintomáticos da doença. Este estudo investigou também a expressão imuno-histoquímica de marcadores envolvidos nos mecanismos de patogenicidade do dano ao nervo na hanseníase. Os imunomarcadores selecionados foram: quimiocinas CXCL10, CCL2, CD3, CD4, CD8, CD45RA, CD45RO, CD68, HLA-DR, e metaloproteinases 2 e 9. O estudo foi desenvolvido em espécimes de biópsias congeladas de nervo coletados de pacientes com HNP (n=23 / 6 BAAR+ e 17 BAAR - PCR +) e pacientes diagnosticados com outras neuropatias (n=5) utilizados como controle. Todas as amostras foram criosseccionadas e submetidas à imunoperoxidase. Os resultados iniciais demonstraram que as 6 amostras de nervos BAAR+ são LAM+/PGL1+. Já entre as 17 amostras de nervos BAAR-, 8 são LAM+ e/ou PGL1+. Nas 17 amostras de nervos BAAR-PCR+, apenas 7 tiveram resultados LAM+ e/ou PGL1+. A detecção de imunorreatividade para LAM e PGL1 nas amostras de nervo do grupo HNP contribuiu para a maior eficiência diagnóstica na ausência recursos a diagnósticos moleculares. Os resultados da segunda parte deste estudo mostraram que foram encontradas imunoreatividade para CXCL10, CCL2, MMP2 e MMP9 nos nervos da hanseníase, mas não em amostras de nervos com outras neuropatias. Além disso, essa imunomarcação foi encontrada predominantemente em células de Schwann e em macrófagos da população celular inflamatória nos nervos HNP. Os outros marcadores de ativação imunológica foram encontrados em leucócitos (linfócitos T e macrófagos) do infiltrado inflamatório encontrados nos nervos. A expressão de todos os marcadores, exceto CXCL10, apresentou associação com a fibrose, no entanto, apenas a CCL2, independentemente dos outros imunomarcadores, estava associada a esse excessivo depósito de matriz extracelular. Nenhuma diferença na frequência da imunomarcação foi detectada entre os subgrupos BAAR+ e BAAR-, exceção feita apenas às células CD68+ e HLA-DR+, que apresentaram discreta diferença entre os grupos BAAR + e BAAR- com granuloma epitelioide. A expressão de MMP9 associada com fibrose é consistente com os resultados anteriores do grupo de pesquisa. Estes resultados indicam que as quimiocinas CCL2 e CXCL10 não são determinantes para o estabelecimento das lesões com ou sem bacilos nos em nervo em estágios avançados da doença, entretanto, a CCL2 está associada com o recrutamento de macrófagos e com o desenvolvimento da fibrose do nervo na lesão neural da hanseníase.
The diagnosis of pure neural leprosy (PNL) is based on clinical and laboratory data, including the histopathology of nerve biopsy specimens and detection of M. leprae DNA by polymerase chain reaction (PCR). Given that histopathological examination and PCR methods may not be sufficient to confirm diagnosis, immunolabeling of lipoarabinomanan (LAM) and/or phenolic glycolipid 1 (PGL1) M. leprae wall components were utilized in the first step of this investigation in an attempt to detect any vestigial presence of M. leprae in AFB- nerve samples. Furthermore, its well known that nerve damage in leprosy can be directly induced by Mycobacterium leprae in the early stages of infection; however, immunomediated mechanisms add gravity to the impairment of neural function in symptomatic periods of the disease. Therefore, this study also investigated the immunohistochemical expression of immunomarkers involved in the pathogenic mechanisms of leprosy nerve damage. These markers selected were CXCL10, CCL2 chemokines and CD3, CD4, CD8, CD45RA, CD45RO, CD68, HLA-DR, metalloproteinases 2 and 9 in nerve biopsy specimens collected from leprosy (23) and nonleprosy patients (5) suffering peripheral neuropathy. Twenty-three PNL nerve samples (6 AFB+ and 17 AFB-PCR+) were cryosectioned and submitted to LAM and PGL1 immunohistochemical staining by immunoperoxidase; 5 nonleprosy nerve samples were used as controls. The 6 AFB-positive samples showed LAM/PGL1 immunoreactivity. Among the 17 AFB- samples, only 8 revealed LAM and/or PGL1 immunoreactivity. In 17 AFB-PCR+ patients, just 7 had LAM and/or PGL1-positive nerve results. In the PNL cases, the detection of immunolabeled LAM and PGL1 in the nerve samples would have contributed to enhanced diagnostic efficiency in the absence of molecular diagnostic facilities. The results of the second part of this study showed that CXCL10-, CCL2-, MMP2- and MMP9-immunoreactivities were found in the leprosy nerves but not in nonleprosy samples. Immunolabeling was predominantly found in recruited macrophages and Schwann cells composing the inflammatory cellular population in the leprosy-affected nerves. The immunohistochemical expression of all the markers, but CXCL10, was associated with fibrosis; however, only CCL2 was, independently from the other markers, associated with this excessive deposit of extracellular matrix. No difference in the frequency of the immunolabeling was detected between the AFB+ and AFB- leprosy subgroups of nerves, exception made to some statistical tendency to difference in regard to CD68+ and HLA-DR+ cells in the AFB- nerves exhibiting epithelioid granuloma. MMP9 expression associated with fibrosis is consistent with previous results of this research group. The findings conveys the idea that CCL2 and CXCL10 chemokines at least in advanced stages of leprosy nerve lesions are not determinant for the establishment of AFB+ or AFB- leprosy lesions, however, CCL2 is associated with macrophage recruitment and fibrosis.

Le sujet de thèse porte sur la maîtrise et la mesure des formes neuronales, “maîtrise” du fait de l’emploi de micropatterns adhésifs permettant un contrôle des formes cellulaires en deux dimensions, “mesure” du fait de notre volonté d’accéder au volume ainsi qu’à la masse sèche de la cellule par l’emploi de deux techniques complémentaires faisant appel à l’interférométrie ou à des mesures de fluorescence en espace confiné.La question biologique au cœur de cette thèse est celle de la régulation par le neurone de diverses caractéristiques morphologiques comme sa longueur, son volume en lien avec l’établissement de la polarité axo-dendritique. Ces aspects sont développés et approfondis au cours de cette thèse des points de vue expérimentaux mais aussi théoriques (coll. Nir Gov, Institut Weizmann).Ce sujet de thèse multidisciplinaire porte ainsi des aspects de biologie et d’instrumentation physique
The thesis deals with the control and the measurement of neuronal shapes, "control" by using adhesive micropatterns allowing to constrain cells shape in two dimensions, "measurement" by using either interferometry or fluorescence measurements in confined spaces to gain knowledge on cell dry mass and volume.The biological question at the heart of this thesis is the regulation by the neuron of its various morphological characteristics such as length, volume, in association with the establishment of the axo-dendritic polarity. These aspects are developed and deepened in the course of this thesis on experimental but also theoretical (coll. Nir Gov, Weizmann Institute) point of views.This multidisciplinary thesis topic thus builds on biological aspects and physical instrumentation

This study investigated the neural bases underlying representation of nouns and verbs at the semantic, lexical form, and morpho-syntactic levels in Mandarin Chinese, a language with little inflectional morphology. Compared with other studies employing European languages with rich inflections, examination of Chinese would allow the separation of conceptual and morpho-syntactic operations based on different stimulus formats and experimental paradigms. To deal with both the theoretical and design issues in previous studies, several additional measures were taken. First, at each cognitive level, two experiments, one receptive and one expressive, were conducted. Moreover, convergence across experiments at the same cognitive level was computed in order to search for taskindependent grammatical class effects. Second, both concrete and abstract nouns and verbs were included, and conjunction analyses across the two concreteness levels were employed to ensure the generalizability of the findings to all nouns and verbs. Results revealed greater activation for verbs in the left posterior lateral temporal gyri in experiments at both semantic and morpho-syntactic levels, and stronger responses in the prefrontal cortex, including left BA47 and the supplementary motor area, only for morpho-syntactic processing associated with nominal grammatical morphemes, namely, classifiers. No differential levels of activation for nouns and verbs were observed in tasks emphasizing word form representation. While greater activation for processing of nominal classifiers in prefrontal areas may reflect differences in computational complexity associated with selection of grammatical morphemes, the involvement of left posterior lateral temporal cortex has been interpreted as reflecting semantic processing of verbs. The nature of processes represented in each of these regions was further discussed with findings from previous relevant studies. Finally, future studies are proposed for further exploration into the neural mechanisms underlying presentation of nouns and verbs using more recently developed methods of analyses.
published_or_final_version
Speech and Hearing Sciences
Doctoral
Doctor of Philosophy

This master thesis deals with the principles of form design, form printing and form processing. Three different types of forms and applications for their detection are created with the reference of these principles. This application provides to create a new type of form and to print out a form. The application itself is implemented in C++ with the use of OpenCV library. This work describes the classification methods of direction finding marks, identification numbers and submission numbers, bar codes EAN-13, page numbers, answer fields and single answers. The classification of all the handwritten numbers is implemented by neural nets.

La présence de porosités dans les lingots métalliques représente un problème majeur dans l’industrie des matériaux. En effet, ces porosités altèrent significativement les caractéristiques mécaniques du matériau (ductilité notamment), et sont des sources d’apparition de défauts en mise en forme ou en tenue en service. Pour éliminer ces porosités, les industriels utilisent souvent des procédés de mise forme à chaud tels que le forgeage ou le laminage, mais il est souvent difficile de définir le taux de déformation à appliquer pour refermer entièrement ces porosités. La modélisation numérique s’avère donc être un outil particulièrement intéressant afin d’étudier l’impact des paramètres procédé sur le taux de refermeture de porosités. Dans ce travail, nous avons développé une méthodologie de calibration basée sur des algorithmes d’optimisation et une base de données de 800 simulations à champ complet sur VER, où les paramètres influents sur la refermeture des porosités sont variés (mécaniques et géométriques). Le premier modèle proposé est un modèle scalaire qui s’affranchit de l’hypothèse de chargement axisymétrique, largement utilisée dans la littérature. Le paramètre de Lode a permis avec l’utilisation de la triaxialité des contraintes de définir l’état de contraintes d’une manière unique. Les comparaisons de ce nouveau modèle à trois autres modèles de refermeture de la littérature montrent le gain de précision de ce nouveau modèle scalaire de refermeture. Le deuxième modèle est un modèle tensoriel adapté aux procédés multipasses grâce à l’analyse de la matrice d’inertie de la porosité. Cette matrice sert pour calculer le volume, la forme et l’orientation de la porosité. Ce modèle a été calibré en utilisant une approche basée sur les réseaux de neurones artificiels. La comparaison avec le modèle scalaire et la modélisation en champ complet a montré un gain en précision jusqu’à 35%. Il s’agit là par ailleurs du premier modèle tensoriel proposé dans la littérature
The presence of voids in ingots is a major issue in the casting industry. These voids decrease materials properties (in particular ductility) and may induce premature failure during metal forming or service life. Hot metal forming processes are therefore used to close these voids and obtain a sound product. However, the amount of deformation required to close these voids is difficult to estimate.Numerical modeling is an interesting tool to study the influence of process parameters on void closure rate. In this work, an optimization-based strategy has been developed to identify the parameters of a mean-field model based on a database of 800 full-field REV simulations with various loading conditions and voids geometry and orientations. The first void closure model is a scalar model that gets rid of the axisymmetric loading hypothesis considered in most models in the literature. The Lode angle, coupled with the stress triaxiality ratio enables to identify the stress state in a unique way. Comparisons of this new model with three other models fromthe literature show the accuracy increase for general loading conditions. In order to address multistages processes, a second model is defined in a tensor version. The ellipsoid void inertia matrix is used to define void’s morphology, orientation and volume. The tensor model predicts the evolution of the inertia terms and its calibration is based on the full-field REV database and on a new Artificial Neural Networks approach. Comparisons were carried out between this tensor model, the scalar model and full-field simulations for multi-stages configurations. These comparisons showed up to 35% accuracy improvement with the tensor model. It is worth mentioning that this is the first attempt to define a void closure tensor model in the literature

For social robots to move and behave appropriately in dynamic and complex social contexts they need to be flexible in their movement behaviors. The natural complexity of social interaction makes this a difficult property to encode programmatically. Instead of programming these algorithms by hand it could be preferable to have the system learn these behaviors. In this project a framework is created in which an agent, through deep reinforcement learning, can learn how to mimic poses, here defined as the most basic case of social movements. The framework aimed to be as agent agnostic as possible and suitable for both real life robots and virtual agents through an approach called "dancer in the mirror". The framework utilized a learning algorithm called PPO and trained agents, as a proof of concept, on both a virtual environment for the humanoid robot Pepper and for virtual agents in a physics simulation environment. The framework was meant to be a simple starting point that could be extended to incorporate more and more complex tasks. This project shows that this framework was functional for agents to learn to mimic poses on a simplified environment.

L'estimation de profondeur à partir d'une seule image est maintenant cruciale pour plusieurs applications, de la robotique à la réalité virtuelle. Les approches par apprentissage profond dans les tâches de vision par ordinateur telles que la reconnaissance et la classification d'objets ont également apporté des améliorations au domaine de l'estimation de profondeur. Dans cette thèse, nous développons des méthodes pour l'estimation en profondeur avec un réseau de neurones profond en explorant différents indices, tels que le flou de défocalisation et la sémantique. Nous menons également plusieures expériences pour comprendre la contribution de chaque indice à la performance du modèle et sa capacité de généralisation. Dans un premier temps, nous proposons un réseau de neurones convolutif efficace pour l'estimation de la profondeur ainsi qu'une stratégie d'entraînement basée sur les réseaux génératifs adversaires conditionnels. Notre méthode permet d'obtenir des performances parmis les meilleures sur les jeux de données standard. Ensuite, nous proposons d'explorer le flou de défocalisation, une information optique fondamentalement liée à la profondeur. Nous montrons que ces modèles sont capables d'apprendre et d'utiliser implicitement cette information pour améliorer les performances et dépasser les limitations connues des approches classiques d'estimation de la profondeur par flou de défocalisation. Nous construisons également une nouvelle base de données avec de vraies images focalisées et défocalisées que nous utilisons pour valider notre approche. Enfin, nous explorons l'utilisation de l'information sémantique, qui apporte une information contextuelle riche, en apprenant à la prédire conjointement avec la profondeur par une approache multi-tâche
Depth estimation from a single image is a key instrument for several applications from robotics to virtual reality. Successful Deep Learning approaches in computer vision tasks as object recognition and classification also benefited the domain of depth estimation. In this thesis, we develop methods for monocular depth estimation with deep neural network by exploring different cues: defocus blur and semantics. We conduct several experiments to understand the contribution of each cue in terms of generalization and model performance. At first, we propose an efficient convolutional neural network for depth estimation along with a conditional Generative Adversarial framework. Our method achieves performances among the best on standard datasets for depth estimation. Then, we propose to explore defocus blur cues, which is an optical information deeply related to depth. We show that deep models are able to implicitly learn and use this information to improve performance and overcome known limitations of classical Depth-from-Defocus. We also build a new dataset with real focused and defocused images that we use to validate our approach. Finally, we explore the use of semantic information, which brings rich contextual information while learned jointly to depth on a multi-task approach. We validate our approaches with several datasets containing indoor, outdoor and aerial images

Submitted by Mayara Nascimento (mayara.nascimento@ufba.br) on 2016-05-31T14:15:01Z No. of bitstreams: 1 Igor Bastos - Dissertação VFinal.pdf: 3703212 bytes, checksum: a00013910865dacb8025d56659076abb (MD5)
Approved for entry into archive by Alda Lima da Silva (sivalda@ufba.br) on 2016-06-03T23:17:29Z (GMT) No. of bitstreams: 1 Igor Bastos - Dissertação VFinal.pdf: 3703212 bytes, checksum: a00013910865dacb8025d56659076abb (MD5)
Made available in DSpace on 2016-06-03T23:17:29Z (GMT). No. of bitstreams: 1 Igor Bastos - Dissertação VFinal.pdf: 3703212 bytes, checksum: a00013910865dacb8025d56659076abb (MD5)
Gestos são ações corporais não-verbais voltadas para a expressão de algum significado. Estes incluem movimentos de mãos, face, braços, dedos, entre outros, sendo abordados por trabalhos que visam reconhecê-los para promover interações humanas com sistemas computacionais. Devido à grande aplicabilidade do reconhecimento de gestos, tem-se notado que estes trabalhos estão se tornando mais comuns, utilizando técnicas e metodologias mais elaboradas e capazes de prover resultados cada vez melhores. A opção por quais técnicas aplicar para o reconhecimento de gestos varia de acordo com a estratégia empregada em cada trabalho e quais aspectos são utilizados para este reconhecimento. Tem-se, por exemplo, trabalhos baseados no uso de modelos estatísticos. Outros optam pela aquisição de características geométricas de mãos e partes do corpo, enquanto outros, dentre os quais se enquadra o presente trabalho, optam pelo uso de descritores e classificadores, responsáveis por extrair características das imagens relevantes para o seu reconhecimento e; por realizar a classificação efetiva dos gestos baseado nestas informações. Neste âmbito, o presente trabalho visa elaborar, aplicar e apresentar uma abordagem para o reconhecimento de gestos, embasando-se em uma revisão da literatura a respeito das principais técnicas e metodologias empregadas para este fim e escolhendo como campo prático, a Língua Brasileira de Sinais (Libras). Para a extração de informações das imagens, optou-se pelo uso de um vetor de características resultante da aplicação dos descritores Histograma de Gradientes Orientados (HOG) e Momentos Invariantes de Zernike (MIZ), os quais voltam-se para as formas e contornos presentes nas imagens. Para o reconhecimento, foi utilizado o classificador Perceptron Multicamada, sendo este disposto em uma arquitetura onde o processo de classificação é dividido em 2 estágios. Devido à inexistência de datasets públicos da Libras, fez-se necessária, com o auxílio de especialistas da língua e alunos surdos, a criação de um dataset de 9600 imagens, as quais referem-se a 40 sinais da Libras. Isso fez com que a presente abordagem partisse desta criação do dataset até a etapa final de classificação dos sinais. Por fim, testes foram realizados e obteve-se 96,77% de taxa de acerto, evidenciando um alto índice de acerto. Este resultado foi validado considerando possíveis ameaças à abordagem, como a realização de testes considerando um indivíduo não-presente no conjunto de treinamento do classificador e a aplicação da abordagem em um dataset público de gestos.

Ces travaux de thèse, réalisés dans une équipe concevant des circuits analogiques neuromimétiques suivant le modèle d'Hodgkin-Huxley, concernent la modélisation de neurones biologiques, plus précisément, l'estimation des paramètres de modèles de neurones. Une première partie de ce manuscrit s'attache à faire le lien entre la modélisation neuronale et l'optimisation. L'accent est mis sur le modèle d'Hodgkin- Huxley pour lequel il existait déjà une méthode d'extraction des paramètres associée à une technique de mesures électrophysiologiques (le voltage-clamp) mais dont les approximations successives rendaient impossible la détermination précise de certains paramètres. Nous proposons dans une seconde partie une méthode alternative d'estimation des paramètres du modèle d'Hodgkin-Huxley s'appuyant sur l'algorithme d'évolution différentielle et qui pallie les limitations de la méthode classique. Cette alternative permet d'estimer conjointement tous les paramètres d'un même canal ionique. Le troisième chapitre est divisé en trois sections. Dans les deux premières, nous appliquons notre nouvelle technique à l'estimation des paramètres du même modèle à partir de données biologiques, puis développons un protocole automatisé de réglage de circuits neuromimétiques, canal ionique par canal ionique. La troisième section présente une méthode d'estimation des paramètres à partir d'enregistrements de la tension de membrane d'un neurone, données dont l'acquisition est plus aisée que celle des courants ioniques. Le quatrième et dernier chapitre, quant à lui, est une ouverture vers l'utilisation de petits réseaux d'une centaine de neurones électroniques : nous réalisons une étude logicielle de l'influence des propriétés intrinsèques de la cellule sur le comportement global du réseau dans le cadre des oscillations gamma.

L’objectif est de préciser les relations entre traitement du mouvement visuel et reconnaissance de forme lors de la perception d’objets 3D dynamique. Nous testons l’hypothèse d’une séquentialité des traitements dans des tâches d’identification. Nos résultats sont compatibles avec un accès préalable à l’information de mouvement. Puis, nous utilisons un protocole de PSVR pour tester les interactions entre les attributs de mouvement et de forme, avec l’hypothèse qu’un traitement perturbé du mouvement détériore la perception de la forme. Contrairement à nos attentes, la forme qui influence l’identification du mouvement. L’ordre des traitements est ainsi remis en cause. Enfin, en MEG, nous précisons le décours temporel des activités des aires impliquées dans le traitement des structures 3D quand l’attention est portée soit sur la forme soit sur le mouvement. Nous distinguons la construction de l’information 3D à partir de mouvement et le traitement attentionnel des attributs de l’objet.

Le travail présenté traite de l'application des techniques d'apprentissage symbolique automatique à la reconnaissance d'images. Le but est d'obtenir de façon automatique des fonctions de reconnaissance d'objets présents sur des images, en généralisant un ensemble d'images données en exemple. Le processus de généralisation est conduit par niveaux de détails croissants en prenant en compte la forme grossière d'un objet, puis ses détails si nécessaire. Un lien est établi entre apprentissage symbolique automatique et reconnaissance à l'aide d'un réseau neuronique.

This paper attempts to develop a mathematically rigid and unified framework for neural spatial interaction modeling. Families of classical neural network models, but also less classical ones such as product unit neural network ones are considered for the cases of unconstrained and singly constrained spatial interaction flows. Current practice appears to suffer from least squares and normality assumptions that ignore the true integer nature of the flows and approximate a discrete-valued process by an almost certainly misrepresentative continuous distribution. To overcome this deficiency we suggest a more suitable estimation approach, maximum likelihood estimation under more realistic distributional assumptions of Poisson processes, and utilize a global search procedure, called Alopex, to solve the maximum likelihood estimation problem. To identify the transition from underfitting to overfitting we split the data into training, internal validation and test sets. The bootstrapping pairs approach with replacement is adopted to combine the purity of data splitting with the power of a resampling procedure to overcome the generally neglected issue of fixed data splitting and the problem of scarce data. In addition, the approach has power to provide a better statistical picture of the prediction variability, Finally, a benchmark comparison against the classical gravity models illustrates the superiority of both, the unconstrained and the origin constrained neural network model versions in terms of generalization performance measured by Kullback and Leibler's information criterion.

Dissertação (mestrado) - Universidade Federal de Santa Catarina, Centro Tecnológico. Programa de Pós-Graduação em Ciência da Computação
Made available in DSpace on 2012-10-20T14:47:33Z (GMT). No. of bitstreams: 1 198727.pdf: 2500158 bytes, checksum: b788709e982b206f1cb7dd8c4c82b191 (MD5)
Reconhecimento de faces humanas é uma área de grande interesse no mundo científico. A maioria das tecnologias desenvolvidas utiliza imagens com informação 2D. Este trabalho contou com um método inédito de processamento para a obtenção da forma 3D de uma face. Por meio dele se produziu várias bases de dados, com diferentes resoluções e níveis de ruído aceitáveis. Elas foram utilizadas na construção de um sistema de reconhecimento de face baseado em redes neurais artificiais. A vantagem de se utilizar a forma 3D da face está na exclusão de problemas ocasionados pela iluminação e desalinhamento.

The world of Machine Learning and Computer Vision has experienced a revolution since the last years. The appearance of Deep Learning algorithms and Convolutional Neural Networks, altogether with the increased processing capabilities provided by modern GPUs and the enormous amounts of annotated data publicly available, have allowed a boost in the field as never seen before. These notable improvements achieved in the Machine Learning world have led to the appearance of new fields like the Multimodal Learning, which encompasses and learns from many subfields. Additionally, new applications have taken profit of these advancements in order to reach high levels of performance. The huge results improvement of the currently available algorithms have allowed not only revolutionizing the academic world, but also bringing AI-based solutions to the market that looked like science fiction barely 10 years ago. This thesis, which is written as a papers compendium, focuses on delving deeper into the novel topic of Deep Multimodal Learning by proposing new algorithms and solutions for both already existing and newly defined problems. From the applications perspective, most of the papers presented can be divided in two areas of applicability. From the one hand, Egocentric Vision and Storytelling, which consists in acquiring images from the daily life of a person in order to analyse its behaviour patterns like social interactions, activities and events, interactions with objects, etc. And on the other hand, Food Recognition and Analysis, which consists in visually analysing and recognizing the food appearing on images in multiple contexts and with different levels of complexity, from food groups recognition to nutritional analysis. In both applications, the final purpose of the proposed papers is building tools that provide information that could lead to a better quality of life of the users.
El mundo del Machine Learning y la Visión por Computador ha experimentado una revolución los últimos años. La aparición de algoritmos de Deep Learning y Convolutional Neural Networks, junto con las mayores capacidades de procesamiento proporcionadas por GPU modernas y las enormes cantidades de datos anotados disponibles públicamente, han permitió un impulso en el campo como nunca antes se había visto.Estas notables mejoras logradas en el mundo del Machine Learning han llevado a la aparición de nuevos campos como el Aprendizaje Multimodal, que engloba y aprende de muchos subcampos. Además, nuevas aplicaciones han aprovechado estos avances para alcanzar altos niveles de rendimiento. La enorme mejora en los resultados de los algoritmos disponibles actualmente ha permitido no solo revolucionar el mundo académico, sino también llevar al mercado soluciones basadas en IA que parecían ciencia ficción hace apenas 10 años.Esta tesis, que está escrita como un compendio de artículos, se enfoca en profundizar en el novedoso tema del Aprendizaje Multimodal Profundo al proponer nuevos algoritmos y soluciones para problemas ya existentes y recientemente definidos. Desde la perspectiva de las aplicaciones, la mayoría de los trabajos presentados se pueden dividir en dos áreas de aplicabilidad. Por un lado, la Visión Egocéntrica y el Storytelling, que consiste en la adquisición de imágenes de la vida diaria de una persona para analizar su comportamiento y extraer patrones asociadas a estos como por ejemplo interacciones sociales, actividades y eventos, interacciones con objetos, etc. Y por otro lado, el Reconocimiento y Análisis de Alimentos, que consiste en visualmente analizar y reconocer la comida que aparece en imágenes en múltiples contextos y con diferentes niveles de complejidad, desde el reconocimiento de grupos de alimentos hasta el análisis nutricional.En ambas aplicaciones, el propósito final de los artículos propuestos es construir herramientas que brinden información que pueda conducir a una mejor calidad de vida de los usuarios.

Tesis para optar al grado de Magíster en Lingüistica mención Lengua Inglesa
Studying children’s language acquisition in natural settings is not cost and time effective. Therefore, language acquisition may be studied in an artificial setting reducing the costs related to this type of research. By artificial, I do not mean that children will be placed in an artificial setting, first because this would not be ethical and second because the problem of the time needed for this research would still be present. Thus, by artificial I mean that the tools of simulation found in artificial intelligence can be used. Simulators as artificial neural networks (ANNs) possess the capacity to simulate different human cognitive skills, as pattern or speech recognition, and can also be implemented in personal computers with software such as MATLAB, a numerical computing software. ANNs are computer simulation models that try to resemble the neural processes behind several human cognitive skills. There are two main types of ANNs: supervised and unsupervised. The learning processes in the first are guided by the computer programmer, while the learning processes of the latter are random.

Submitted by Alessandra Portugal (alessandradf@ioc.fiocruz.br) on 2013-09-24T15:19:37Z No. of bitstreams: 1 Fernando Ricardo Serejo de Castro.pdf: 2042022 bytes, checksum: 4c1258f6d4ddae7e1ae9a1ecaf90f618 (MD5)
Made available in DSpace on 2013-09-24T15:19:37Z (GMT). No. of bitstreams: 1 Fernando Ricardo Serejo de Castro.pdf: 2042022 bytes, checksum: 4c1258f6d4ddae7e1ae9a1ecaf90f618 (MD5) Previous issue date: 2012
Fundação Oswaldo Cruz. Instituto Oswaldo Cruz. Rio de Janeiro, RJ, Brasil
A hanseníase neural pura (NP) caracteriza-se pelo comprometimento nervoso sem o aparecimento de lesões cutâneas. O objetivo do estudo foi descrever as alterações neurológicas observadas na forma NP e caracterizar a funcionalidade e as incapacidades encontradas nestes pacientes. Trata-se de um estudo retrospectivo, realizado com dados de 79 pacientes que confirmaram diagnóstico de NP acompanhados no Ambulatório Souza Araújo (ASA), no período de 2000 a 2010. As informações foram obtidas a partir de prontuários médicos e de bancos de dados. A Classificação Internacional de Funcionalidade, Incapacidade e Saúde (CIF) foi aplicada a partir da análise das avaliações neurológicas e fisioterapêuticas. Foi realizada uma análise de frequência para verificar a distribuição dos casos de NP em relação às variáveis demográficas, socioeconômicas e epidemiológicas (no diagnóstico), às variáveis do exame neurológico, ao grau de incapacidade física (GIF) e aos códigos da CIF (no diagnóstico e na alta). Para avaliar as características demográficas, socioeconômicas e epidemiológicas, foi realizada uma comparação entre o grupo de pacientes NP com o grupo de 634 pacientes com hanseníase paucibacilar (PB), acompanhados no ASA, no mesmo período dos pacientes NP. Além disso, foi investigada a relação do grau de incapacidade física no diagnóstico com as variáveis socioeconômicas e demográficas dos dois grupos de pacientes, através do teste de qui-quadrado (2). A maioria dos pacientes NP (71%) e PB (57,5%) era do sexo masculino e em relação à faixa etária, os pacientes NP foram acometidos mais tardiamente pela hanseníase do que os pacientes PB. A maior proporção dos pacientes NP (55%) e PB (47,5%) estava empregada e trabalhava no mercado informal. A renda familiar mensal não diferiu significativamente entre os grupos. Quanto à escolaridade, a maior parte dos pacientes apresentava até oito anos de estudo. Em relação ao modo de detecção, a maior proporção dos pacientes foi encaminhada ao ASA por meio algum serviço público de saúde. Quanto às alterações neurológicas nos pacientes NP, o sintoma inicial relatado com mais frequência foi a parestesia (52%) e o sinal predominante observado no exame diagnóstico foi a alteração de sensibilidade (87%). Na avaliação de alta, houve uma menor proporção de todos os sinais e sintomas neurológicos. Em relação ao GIF, 44,3% dos pacientes apresentaram algum GIF e na alta, 24% dos indivíduos permaneceram com algum grau de incapacidade. No tocante à CIF, dentro do domínio ―Estruturas do Corpo‖, o código mais frequente foi ―nervos raquidianos‖ (s1201) e no domínio ―Função do Corpo‖, o código mais frequente foi ―sensibilidade à temperatura‖ (b2700) no diagnóstico e ―sensibilidade a estímulos nocivos‖ (b2703) na alta. Este estudo apontou uma associação entre baixas condições socioeconômicas e um pior escore de GIF e caracterizou funcionalmente os pacientes NP.
The Pure Neural Leprosy (PNL) is characterized by neural involvement without the appearance of skin lesions. The goal of the study was to describe the neurological changes observed in the PNL and to characterize the functionality and the disabilities found in these patients. This is a retrospective study, conducted with 79 patients who confirmed diagnosis of PNL accompanied at Souza Araújo Clinic (ASA), from 2000 to 2010.The information was obtained from patient records and databases. The International Classification of Functioning, Disability and Health (ICF) was applied from the neurological and physiotherapeutic evaluations. A frequency analysis was conducted to verify the distribution of cases of PNL in relation to demographic and socioeconomic variables (diagnosis), neurological examination variables, grade of disability (GD) and ICF codes (upon diagnosis and release from treatment). In order to assess demographic and socioeconomic characteristics, a comparative study was undertaken between the PNL patients and the 634 paucibacillary leprosy (PB) patients, accompanied at ASA, during the same period as the PNL patients. Furthermore, the relationship between the GD at diagnosis with socioeconomic and demographic variables of two groups of patients was investigated using the chi-squared test, (2). The majority of PNL patients (71%) and PB (57.5%) were male and PNL patients were affected by leprosy at a later age than PB patients. The largest proportion of PNL patients (55%) and PB (47.5%) was employed in the informal market. The monthly household income did not differ significantly between groups. With regard to schooling, most patients had up to eight years of study. In relation to detection, the largest proportion of patients was referred to ASA from other public health services. With regard to neurological alterations in PNL patients, the most frequently reported initial symptom was the paresthesia (52%) and the predominant sign observed in the examination diagnosis was sensory alteration (87%). At evaluation after release from treatment, there was an overall reduction of all neurological signs and symptoms. Regarding GD, 44.3% of the patients had some GD, and at release from treatment, while 24% of individuals remained with some grade of disability. Regarding ICF, within the domain "Body Structures", the most frequent code was "spinal nerves" (s1201) and in the field "Body Functions", the most frequent code was "Sensitivity to Temperature" (b2700) at diagnosis and "Sensitivity to Noxious Stimuli" (b2703) at release from treatment. This study noted an association between low socio-economic conditions and a worse ICF score and characterized PNL patients functionally.

Since the default of Lehman Brothers in 2008, it has become increasingly important to measure, manage and price the default risk in financial derivatives. Default risk in financial derivatives is referred to as counterparty credit risk (CCR). The price of CCR is captured in Credit Valuation Adjustment (CVA). This adjustment should in principle always enter the valuation of a derivative traded over-the-counter (OTC). To calculate CVA, one needs to know the probability of default of the counterparty. Since CVA is a price, what one needs is the risk-neutral probability of default. The typical way of obtaining risk-neutral default probabilities is to build credit curves calibrated using Credit Default Swaps (CDS). However, for a majority of a bank's counterparties there are no CDSs liquidly traded. This constitutes a major challenge. How does one model the risk-neutral default probability in the absence of observable CDS spreads? A number of methods for constructing proxy credit curves have been proposed previously. A particularly popular choice is the so-called Nomura (or cross-section) model. In studying this model, we find some weaknesses, which in some instances lead to degenerate proxy credit curves. In this thesis we propose an altered model, where the modelling quantity is changed from the CDS spread to the hazard rate. This ensures that the obtained proxy curves are valid by construction. We find that in practice, the Nomura model in many cases gives degenerate proxy credit curves. We find no such issues for the altered model. In some cases, we see that the differences between the models are minor. The conclusion is that the altered model is a better choice since it is theoretically sound and robust.
Sedan Lehman Brothers konkurs 2008 har det blivit allt viktigare att mäta, hantera och prissätta kreditrisken i finansiella derivat. Kreditrisk i finansiella derivat benämns ofta motpartsrisk (CCR). Priset på motpartsrisk fångas i kreditvärderingsjustering (CVA). Denna justering bör i princip alltid ingå i värderingen av ett derivat som handlas över disk (eng. over-the-counter, OTC). För att beräkna CVA behöver man veta sannolikheten för fallissemang (konkurs) hos motparten. Eftersom CVA är ett pris, behöver man den riskneutrala sannolikheten för fallissemang. Det typiska tillvägagångsättet för att erhålla riskneutrala sannolikheter är att bygga kreditkurvor kalibrerade med hjälp av kreditswappar (CDS:er). För en majoritet av en banks motparter finns emellertid ingen likvid handel i CDS:er. Detta utgör en stor utmaning. Hur ska man modellera riskneutrala fallissemangssannolikheter vid avsaknad av observerbara CDS-spreadar? Ett antal metoder för att konstruera proxykreditkurvor har föreslagits tidigare. Ett särskilt populärt val är den så kallade Nomura- (eller cross-section) modellen. När vi studerar denna modell hittar vi ett par svagheter, som i vissa fall leder till degenererade proxykreditkurvor. I den här uppsatsen föreslår vi en förändrad modell, där den modellerade kvantiteten byts från CDS-spreaden till riskfrekvensen (eng. hazard rate). Därmed säkerställs att de erhållna proxykurvorna är giltiga, per konstruktion. Vi finner att Nomura-modellen i praktiken i många fall ger degenererade proxykreditkurvor. Vi finner inga sådana problem för den förändrade modellen. I andra fall ser vi att skillnaderna mellan modellerna är små. Slutsatsen är att den förändrade modellen är ett bättre val eftersom den är teoretiskt sund och robust.

La vision par ordinateur est un domaine stratégique, du fait du nombre potentiel d'applications avec un impact important sur la société. Ce secteur a rapidement progressé au cours de ces dernières années, notamment grâce aux avancées en intelligence artificielle et plus particulièrement l'avènement de l'apprentissage profond. Cependant, ces méthodes présentent deux défauts majeurs face au cerveau biologique : ils sont extrêmement énergivores et requièrent de gigantesques bases d'apprentissage étiquetées. Les réseaux de neurones à impulsions sont des modèles alternatifs qui permettent de répondre à la problématique de la consommation énergétique. Ces modèles ont la propriété de pouvoir être implémentés de manière très efficace sur du matériel, afin de créer des architectures très basse consommation. En contrepartie, ces modèles imposent certaines contraintes, comme l'utilisation uniquement de mémoire et de calcul locaux. Cette limitation empêche l'utilisation de méthodes d'apprentissage traditionnelles, telles que la rétro-propagation du gradient. La STDP est une règle d'apprentissage, observée dans la biologie, qui peut être utilisée dans les réseaux de neurones à impulsions. Cette règle renforce les synapses où des corrélations locales entre les temps d'impulsions sont détectées, et affaiblit les autres synapses. La nature locale et non-supervisée permet à la fois de respecter les contraintes des architectures neuromorphiques, et donc d'être implémentable de manière efficace, mais permet également de répondre aux problématiques d'étiquetage des bases d'apprentissage. Cependant, les réseaux de neurones à impulsions entraînés grâce à la STDP souffrent pour le moment de performances inférieures aux méthodes d'apprentissage profond. La littérature entourant la STDP utilise très majoritairement des données simples mais le comportement de cette règle n'a été que très peu étudié sur des données plus complexes, tel que sur des bases avec une variété d'images importante.L'objectif de ce manuscrit est d'étudier le comportement des modèles impulsionnels, entraîné via la STDP, sur des tâches de classification d'images. Le but principal est d'améliorer les performances de ces modèles, tout en respectant un maximum les contraintes imposées par les architectures neuromorphiques. Une première partie des contributions proposées dans ce manuscrit s'intéresse à la simulation logicielle des réseaux de neurones impulsionnels. L'implémentation matérielle étant un processus long et coûteux, l'utilisation de simulation est une bonne alternative pour étudier plus rapidement le comportement des différents modèles. La suite des contributions s'intéresse à la mise en place de réseaux impulsionnels multi-couches. Les réseaux composés d'un empilement de couches, tel que les méthodes d'apprentissage profond, permettent de traiter des données beaucoup plus complexes. Un des chapitres s'articule autour de la problématique de perte de fréquence observée dans les réseaux de neurones à impulsions. Ce problème empêche l'empilement de plusieurs couches de neurones impulsionnels. Une autre partie des contributions se concentre sur l'étude du comportement de la STDP sur des jeux de données plus complexes, tels que les images naturelles en couleur. Plusieurs mesures sont utilisées, telle que la cohérence des filtres ou la dispersion des activations, afin de mieux comprendre les raisons de l'écart de performances entre la STDP et les méthodes plus traditionnelles. Finalement, la réalisation de réseaux multi-couches est décrite dans la dernière partie des contributions. Pour ce faire, un nouveau mécanisme d'adaptation des seuils est introduit ainsi qu'un protocole permettant l'apprentissage multi-couches. Il est notamment démontré que de tels réseaux parviennent à améliorer l'état de l'art autour de la STDP
Computer vision is a strategic field, in consequence of its great number of potential applications which could have a high impact on society. This area has quickly improved over the last decades, especially thanks to the advances of artificial intelligence and more particularly thanks to the accession of deep learning. Nevertheless, these methods present two main drawbacks in contrast with biological brains: they are extremely energy intensive and they need large labeled training sets. Spiking neural networks are alternative models offering an answer to the energy consumption issue. One attribute of these models is that they can be implemented very efficiently on hardware, in order to build ultra low-power architectures. In return, these models impose certain limitations, such as the use of only local memory and computations. It prevents the use of traditional learning methods, for example the gradient back-propagation. STDP is a learning rule, observed in biology, which can be used in spiking neural networks. This rule reinforces the synapses in which local correlations of spike timing are detected. It also weakens the other synapses. The fact that it is local and unsupervised makes it possible to abide by the constraints of neuromorphic architectures, which means it can be implemented efficiently, but it also provides a solution to the data set labeling issue. However, spiking neural networks trained with the STDP rule are affected by lower performances in comparison to those following a deep learning process. The literature about STDP still uses simple data but the behavior of this rule has seldom been used with more complex data, such as sets made of a large variety of real-world images.The aim of this manuscript is to study the behavior of these spiking models, trained through the STDP rule, on image classification tasks. The main goal is to improve the performances of these models, while respecting as much as possible the constraints of neuromorphic architectures. The first contribution focuses on the software simulations of spiking neural networks. Hardware implementation being a long and costly process, using simulation is a good alternative in order to study more quickly the behavior of different models. Then, the contributions focus on the establishment of multi-layered spiking networks; networks made of several layers, such as those in deep learning methods, allow to process more complex data. One of the chapters revolves around the matter of frequency loss seen in several spiking neural networks. This issue prevents the stacking of multiple spiking layers. The center point then switches to a study of STDP behavior on more complex data, especially colored real-world image. Multiple measurements are used, such as the coherence of filters or the sparsity of activations, to better understand the reasons for the performance gap between STDP and the more traditional methods. Lastly, the manuscript describes the making of multi-layered networks. To this end, a new threshold adaptation mechanism is introduced, along with a multi-layer training protocol. It is proven that such networks can improve the state-of-the-art for STDP

Submitted by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2015-01-07T11:19:58Z No. of bitstreams: 2 rafael_ribeiro_souza.pdf: 2710502 bytes, checksum: 2b3967310df95d3b8ec1f8cbff51c07b (MD5) license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5)
Approved for entry into archive by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2015-01-07T11:20:20Z (GMT) No. of bitstreams: 2 rafael_ribeiro_souza.pdf: 2710502 bytes, checksum: 2b3967310df95d3b8ec1f8cbff51c07b (MD5) license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5)
Approved for entry into archive by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2015-01-07T11:20:38Z (GMT) No. of bitstreams: 2 rafael_ribeiro_souza.pdf: 2710502 bytes, checksum: 2b3967310df95d3b8ec1f8cbff51c07b (MD5) license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5)
Made available in DSpace on 2015-01-07T11:20:39Z (GMT). No. of bitstreams: 2 rafael_ribeiro_souza.pdf: 2710502 bytes, checksum: 2b3967310df95d3b8ec1f8cbff51c07b (MD5) license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5) Previous issue date: 2013
Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior (CAPES)
Os objetivos deste estudo foram: avaliar o efeito do arranjo espacial de plantio na forma do fuste de ?rvores de eucalipto; avaliar se as redes neurais artificiais estimam com precis?o o afilamento, os di?metros ao a qualquer altura especificada e as alturas a qualquer di?metro especificado em fustes de eucalipto; estudar a forma dos fustes em arvores de eucalipto em diferentes arranjos espaciais e idades, e de maneira espec?fica, determinar as formas geom?tricas aproximadas, que ocorrem ao longo destes fustes e as suas propor??es em rela??o ? altura total. Foram utilizados dados de um experimento implantado em dezembro de 2002, no delineamento em blocos (tr?s blocos), sendo testados os arranjos espaciais de 3,0 x 0,5; 3,0 x 1,0; 3,0 x 1,5; 3,0 x 2,0 e 3,0 x 3,0 m. Foram utilizados tamb?m, dados de ?rvores provenientes de um sistema agroflorestal-SAF, plantio com 6,03 hectares, implantado em dezembro de 1993, no arranjo espacial de 10,0 x 4,0 m. Para analisar a forma m?dia dos fustes foram utilizados modelos de Kozak, Sch?epfer, Garay, Demaerschalk e Ormerod. Os modelos selecionados foram de Garay e de Sh?epfer. A escolha foi feita com base nas estat?sticas, coeficiente de correla??o, erro-padr?o residual, Bias e da an?lise gr?fica dos res?duos. Testes de identidade de modelos foram aplicados nos modelos selecionados com a finalidade de verificar a igualdade entre a forma dos fustes. Foram utilizadas redes neurais artificiais feed-forward, do tipo Multilayer Perceptrons, treinadas por meio do algoritmo error-backpropagation. Para os ajustes das redes foi utilizada uma aplica??o computacional em linguagem Java, e para as fun??es destinadas ao treinamento e aplica??o foi utilizada a biblioteca Weka. As estat?sticas de acur?cia utilizadas para avaliar as melhores redes foram a raiz quadrada do erro m?dio e as correla??es entre os valores observados e os valores estimados. Utilizando a equa??o geral das curvas, buscou-se determinar a forma aproximada do s?lido geom?trico m?dio dos fustes e descrever as formas geom?tricas aproximadas que os fustes assumem ao longo de seu comprimento, determinar os seus pontos aproximados de inflex?o e a propor??o de cada forma em rela??o ? altura total. O modelo de Garay ? o mais indicado para descrever o taper de eucaliptos nos arranjos espaciais avaliados. Foi aceita a hip?tese de nulidade no teste de identidade de modelos n?o linear, indicando a igualdade entre as equa??es nos arranjos de 3,0 x 0,5 e 3,0 x 1,0 m. Para uma dist?ncia de 3,0 metros entre fileiras, quanto maior a dist?ncia entre plantas, mais c?nica ? a forma do fuste. A rede neural artificial estimou com precis?o o afilamento dos fustes, distinguindo as varia??es na forma dos fustes em virtude dos diferentes arranjos espaciais. A rede neural artificial estimou com precis?o os di?metros a qualquer altura especificada. As estimativas das alturas a qualquer di?metro especificado, obtidas pela RNA e pelo modelo de taper de Garay, apresentaram valores de erros percentuais acentuados na base dos fustes, em todos os arranjos espaciais. Os fustes nos arranjos de 3,0 x 0,5 e 3,0 x 3,0 m, apresentaram a forma m?dia de um parabol?ide, j? os fustes no arranjo de 10,0 x 4,0 m, a forma m?dia de um tronco de cone. Os fustes nos arranjos espaciais 3,0 x 0,5 e 3,0 x 3,0 m, assumem as formas de um tronco de neil?ide, um tronco de cone e um parabol?ide, nas propor??es de 10,96; 43,81 e 45,14 %, e de 14,58; 37,76 e 47,66 %, respectivamente. Os fustes referentes ao arranjo espacial 10,0 x 4,0 m, assumem as formas de um tronco de neil?ide e de um tronco de cone, nas propor??es de 20,78 e 79,30 %, respectivamente, em rela??o ? altura total.
Disserta??o (Mestrado) ? Programa de P?s-Gradua??o em Ci?ncia Florestal, Universidade Federal dos Vales do Jequitinhonha e Mucuri, 2013.
ABSTRACT The objectives of this study were: assess the effect of the planting?s spatial arrangement in the bole form of eucalypt trees; evaluate whether artificial neural networks estimate the tapering with accuracy, the diameters at any specified height and the heights at any specified diameter in eucalypt boles; study the boles? form in eucalypt trees in different spatial arrangements and ages, and in a specific manner, determine the approximate geometric shapes that occur along these boles and their proportions in relation to the total height. There were utilized data from an experiment established in December 2002, in blocks design (three blocks), being tested the spatial arrangements of 3,0 x 0,5; 3,0 x 1,0; 3,0 x 1,5; 3,0 x 2,0 and 3,0 x 3,0 m. There were also used, tree data from an agroforestry system-AFS, with 6,03 hectares of planting, established in December 1993, with spatial arrangement of 10,0 x 4,0 m. To analyze the average forms of the boles, there were used Kozak, Sch?epfer, Garay, Demaerschalk and Ormerod models. The selected models were Garay and Sh?epfer. The selection was made based on the statistics, correlation coefficient, residual standard error, Bias and graphical analysis of the residuals. Models identity tests were applied on the selected models in order to verify the equivalence between the shape of the boles. There were used feed-forward artificial neural networks, Multilayer Perceptron type, trained by the error-backpropagation algorithm. To adjust the networks it was utilized a computational application in Java language, and for the functions intended for training and application it was used the Weka library. The accuracy statistics used to evaluate the best networks were the root mean square error and the correlations between the observed and estimated values?. Using the general equation of the curves, it aimed to determine the approximate shape of the mean geometric solid for the boles and describe the approximate geometric shapes that the boles assume along their length, determine their approximate inflection points and the proportion of each form in relation to the total height. The Garay model is the most suitable to describe the taper of eucalypts in the evaluated spatial arrangements. It was accepted the null hypothesis in the identity test of nonlinear models, indicating equality between the equations in the arrangements of 3,0 x 0,5 and 3,0 x 1,0 m. For a distance of 3,0 m between the rows, the greater the distance between plants is, more conical shaped is the bole. The artificial neural network estimated with accuracy the tapering of the boles, distinguishing variations in the form of the boles due to the different spatial arrangements. The artificial neural network estimated with accuracy the diameters at any specified height. The estimates heights at any specified diameter, obtained by the ANN and taper model of Garay, presented values of percentage errors accented at the base of the boles, in all spatial arrangements. The boles in the arrangements of 3,0 x 0,5 and 3,0 x 3,0 m, presented the mean form of a paraboloid, while the boles in the arrangement of 10,0 x 4,0 m, the mean form of a conic trunk. The boles in the spatial arrangements 3,0 x 0,5 and 3,0 x 3,0 m, assume the forms of a neiloid trunk, a conic trunk and a paraboloid, in the proportions of 10,96; 43.81 and 45,14 %, and of 14,58; 37,76 and 47,66 %, respectively. The boles for the spatial arrangement of 10,0 x 4,0 m, assume the forms of a neiloid trunk and a conic trunk, in the proportions of 20,78 and 79,30 %, respectively, in relation to the total height.

Durante o desenvolvimento de um neurônio, genes são ativados e desativados, a anatomia se forma e as funcionalidades emergem. Estes três componentes influenciam continuamente uns aos outros. O estudo da forma, função e expressão gênica nos neurônios e no cérebro permanece um tema desafiador e com potencial a ser explorado. Neste contexto, uma importante questão ainda a ser respondida é como quantificar o inter-relacionamento entre forma, função e genes. Para isso, foram realizadas atividades envolvendo caracterização e comparação da forma neuronal, o estudo de processos dinâmicos ocorrendo em redes de estruturas ramificadas, e a comparação entre expressões gênicas. Os dados da base pública NeuroMorpho, que possui quase 6.000 neurônios segmentados, foram caracterizados utilizando-se métodos estatísticos e foram analisados pelo conceito de morfoespaço proposto por McGhee. Outra base pública explorada foi o Mouse Allen Brain Atlas, com imagens de expressão gênica de cérebros de camundongo. Foi proposta a utilização de um método baseado em diagramas de Voronoi para a comparação da distribuição espacial de densidades de expressão gênica entre genes, com o propósito de encontrar correlações entre distribuições. Também foram gerados dados sobre raízes de feijão para o estudo da influência de sua estrutura ramificada na dinâmica de propagação de doenças, seguindo o modelo SIR (Suscetível-Infectado-Recuperado). Integrando os desenvolvimentos anteriores, foi proposto um arcabouço para mensurar a influência da expressão gênica ao longo da escala biológica. Este arcabouço permite mensurar a influência da expressão gênica (escala molecular) na morfologia dos neurônios (escala celular), avançando à escala topológica formada pelas conexões sinápticas, e alcançando o nível funcional das dinâmicas sobre essa rede. Nesse contexto, deve-se ressaltar que a influência da expressão gênica é direta sobre a morfologia e indireta sobre a topologia e a dinâmica. As informações obtidas a partir do arcabouço são relevantes na investigação de como a expressão gênica influencia todo o processo, desde o neurônio individual até o funcionamento cerebral. O arcabouço proposto fornece uma metodologia sistemática, com um conjunto de ferramentas para essas análises.
During the development of a neuron, genes are turned on and off, the anatomy is shaped and the functionality emerges. These three components influence each other continuously. The study of form, function and gene expression in neurons and brain is still challenging and has many issues yet to be explored. In this context, an important question yet to be answered is how to quantify the inter-relationship between form, function and gene expression. In this way, we developed activities involving characterization and comparison of the neuronal form, the study of dynamical processes occurring in networks of branching structures, and the comparison between gene expressions. The data of the public database NeuroMorpho, which comprise almost 6,000 segmented neurons, were characterized using statistical methods and were analyzed by the concept of McGhee\'s morphospace. Another public database that was explored was the Mouse Allen Brain Atlas, with images of gene expression of mouse brains. We proposed to use a method based on Voronoi diagrams to compare the spatial distribution of the gene expression densities between genes, in order to find correlations in the distribution. We also generated data on bean roots to study the influence of their branched structures in the dynamics of disease spread, following the SIR model (Susceptible-Infected-Recovered). Integrating the previous developments, we proposed a framework to measure the gene expression influence through the biological scale. This framework allows the measurement of the gene expression (molecular scale) influence in the morphology of the neurons (cellular scale), advancing towards the topological scale formed by the synaptic connections, and reaching the functional level of the dynamics over this network. In this context, it is worth to note that the gene expression influence is direct on the morphology and indirect on the topology and dynamics. The obtained information through the framework is important on the investigation of how the gene expression influences the whole process, since the individual neuron to the cerebral functioning. The proposed framework yields a systematic methodology with a toolbox to carry out these analyses.

La sclérose en plaques (SEP) est une maladie neuro-inflammatoire chronique, première cause de handicap chez le jeune adulte. Actuellement, aucun traitement ne freine l'aggravation des symptômes liée aux formes progressives. Bien que connue, l'implication des macrophages et de la microglie dans la démyélinisation et l'atteinte axonale doit être plus finement caractérisée. Ce d'autant plus que la plasticité fonctionnelle de ces cellules suggère une réponse spécifique selon la pathologie, la localisation des lésions et le stade évolutif de la maladie. Ce travail de thèse a consisté en une caractérisation moléculaire des programmes d'activation macrophagique/ microgliale dans deux types d'altérations tissulaires du système nerveux central des patients SEP : les zones partiellement démyélinisées bordant les plaques de la moelle épinière et les lésions corticales. Cette étude a été réalisée sur des tissus post-mortem de patients atteints de formes progressives, formes dans lesquelles les lésions médullaires et corticales sont nombreuses et impliquées dans le handicap progressif et irréversible. Nous avons identifié des spécificités moléculaires caractérisant l'activation macrophagique/microgliale au cours de la SEP en comparant, par une approche in silico, les profils caractérisés à ceux observés dans des pathologies neuro-dégénératives à composantes inflammatoires, la maladie d'Alzheimer et de Parkinson notamment. Dans l'ensemble, ces résultats suggèrent que l'activation chronique des macrophages/cellules microgliales contribue à l'extension à bas bruit des lésions médullaires et corticales pendant la phase progressive de la SEP et proposent de nouvelles cibles thérapeutiques