We study the problem of embedding high-dimensional visual data into low-dimensional vector representations. This is an important component in many computer vision applications involving nearest neighbor retrieval, as embedding techniques not only perform dimensionality reduction, but can also capture task-specific semantic similarities. In this thesis, we use deep neural networks to learn vector embeddings, and develop a gradient-based optimization framework that is capable of optimizing ranking-based retrieval performance metrics, such as the widely used Average Precision (AP) and Normalized Discounted Cumulative Gain (NDCG). Our framework is applied in three applications. First, we study Supervised Hashing, which is concerned with learning compact binary vector embeddings for fast retrieval, and propose two novel solutions. The first solution optimizes Mutual Information as a surrogate ranking objective, while the other directly optimizes AP and NDCG, based on the discovery of their closed-form expressions for discrete Hamming distances. These optimization problems are NP-hard, therefore we derive their continuous relaxations to enable gradient-based optimization with neural networks. Our solutions establish the state-of-the-art on several image retrieval benchmarks. Next, we learn deep neural networks to extract Local Feature Descriptors from image patches. Local features are used universally in low-level computer vision tasks that involve sparse feature matching, such as image registration and 3D reconstruction, and their matching is a nearest neighbor retrieval problem. We leverage our AP optimization technique to learn both binary and real-valued descriptors for local image patches. Compared to competing approaches, our solution eliminates complex heuristics, and performs more accurately in the tasks of patch verification, patch retrieval, and image matching. Lastly, we tackle Deep Metric Learning, the general problem of learning real-valued vector embeddings using deep neural networks. We propose a learning to rank solution through optimizing a novel quantization-based approximation of AP. For downstream tasks such as retrieval and clustering, we demonstrate promising results on standard benchmarks, especially in the few-shot learning scenario, where the number of labeled examples per class is limited.

En mars 1947, le gouvernement australien dirigé par Joseph Chifley présente devant le Parlement une liste des priorités en matière de politique étrangère et de sécurité pour le Commonwealth d’Australie : « en premier, un soutien plein et entier aux Nations Unies, en second le maintien des liens avec le Commonwealth, en troisième les accords de sécurité avec les Etats-Unis et d’autres Etats [et] en quatrième la participation à la politique régionale ». En 1994, selon le Livre Blanc de la Défense, la politique étrangère de l’Australie a considérablement évolué. Elle repose désormais en priorité sur l’Asie Pacifique : « la sécurité future de l’Australie – comme [sa] prospérité économique – est liée inextricablement à la sécurité et à la prospérité de l’Asie Pacifique ». / N March 1947 the Australian government led by Joseph Chifley present to Parliament a list of priorities in foreign policy and security for the Commonwealth of Australia: "First, a full support to the United Nations, in the second maintaining links with the Commonwealth, in the third security agreements with the United States and other countries [and] in the fourth participation in regional policy. "In 1994, the White Book of Defence, the Australian foreign policy has evolved considerably. She is now based primarily on the Asia Pacific, "the future security of Australia - as [its] economic prosperity - is inextricably linked to security and prosperity of Asia Pacific."

Um dos grandes desafios em medidas de espectrometria gama utilizando fontes que não podem ser consideradas pontuais é a determinação da eficiência de detecção do sistema, uma vez que a geometria da fonte impõe complexidades na tarefa como o tratamento do ângulo sólido e da auto-absorção. Alguns métodos para o cálculo dessa eficiência já são bem estabelecidos e podem ser encontrados na literatura. Neste trabalho é proposto um método com características distintas aos já existentes cujas principais vantagens são, não depender de conhecimento prévio da geometria de detecção e da composição química da amostra a ser medida, e permitir o cálculo da eficiência de uma fonte extensa a partir de medidas com fontes padrão em formato pontual. O método proposto foi testado comparando os valores das eficiências obtidas por ele com os valores obtidos usando uma fonte padrão extensa. Para testar o método proposto em diferentes condições foi utilizado solo como matriz para a amostra em duas diferentes espessuras, 1cm e 3cm. Testou-se também o comportamento do método quando a matriz tem número atômico efetivo mais alto (aqui utilizou-se solo misturado à limalha de ferro) e mais baixo (adição de pó de grafite ao solo). Os resultados indicaram que o método é adequado para amostras de solo com espessura mais fina; já para fontes mais espessas, dois dos três resultados indicaram que a eficiência foi subestimada pelo método proposto, sugerindo que uma geometria mais adequada deva ser escolhida para a amostragem das medidas / One of the greatest challenges in gamma spectroscopy measurements using extense sources is the determination of the system\'s detection efficiency, since the source\'s geometry leads to added complexity to the treatment of parameters as geometry and self-absorption. Some well-established methods for the efficiency determination can be found in the literature. In the present work a method with distinct characteristics is proposed with the main advantages that it doesn\'t require previous knowledge of the detection geometry or of the sample composition; moreover, it allows the determination of the detection efficiency for a large sample based only on measurements using a point-like standard source. The proposed method was tested comparing the efficiency values obtained using the method with the values obtained using a real large source with the same composition and geometry. In order to verify the performance of the method under different conditions, soil matrixes of two different thicknesses (1cm and 3cm) were used; also, to asses the method under different sample compositions, 3cm-thick soil matrixes with increased effective atomic number (where iron filings was added) and with reduced effective atomic mass (where graphite powder was added) were used. The results indicate that the proposed method is suitable for thinner soil samples; as for thicker samples, two of the three results showed a subestimation of the detection efficiency value determined, suggesting that a more suitable sampling geometry should be analyzed.

RESUMO A qualidade da superfície de trabalhos cerâmicos que requer ajuste durante a sua adaptação ao remanescente dentário desperta a atenção durante os procedimentos odontológicos restauradores estéticos. Observa-se a necessidade freqüente de realizar desgastes superficiais nas restaurações cerâmicas, sendo que a preservação de uma superfície lisa é importante para as propriedades destes materiais, bem como para o seu comportamento clínico. Este trabalho propõe-se a avaliar a técnica com maior capacidade de polimento da superfície cerâmica asperizada, através da medição com ponta apalpadora de um rugosímetro, e utilizando o MEV para observações visuais com relação às técnicas empregadas. Para tal 80 corpos-de-prova foram confeccionados utilizando-se as porcelanas Omega 900 e Vitadur Alpha, constituindo quatro grupos experimentais: A – glaze (controle); B – polimento com borrachas abrasivas da DFL mais pasta para polimento da KOTA; C – polimento com borrachas abrasivas mais rodas de feltro mais pasta para polimento Porcelize (Cosmedent); D – polimento com borrachas abrasivas mais pasta porcelize mais discos Flexibuff mais pasta diamantada para brilho Enamelize (Cosmedent). Os dados receberam tratamento estatístico por meio da Análise de variância (ANOVA) e teste de Tukey, o qual indicou haver diferença estatisticamente significante entre os tipos de polimento para os quais as porcelanas foram submetidas, e na interação porcelana / polimento. Os resultados sugerem que todos os tratamentos polidores diminuem a rugosidade superficial das porcelanas em relação ao grupo controle e que dependendo da porcelana utilizada para cada técnica pode-se ter uma variação na rugosidade superficial. Quanto à observação pelo microscópio eletrônico de varredura houve uma correlação parcial entre este e os resultados obtidos pelo rugosímetro, pois aqui a porcelana Omega 900 mostrou-se mais polida no grupo controle, enquanto que na análise pelo rugosímetro apontou para a técnica polidora do Grupo D como sendo aquela que produziu uma superfície mais lisa para esta porcelana. Por outro lado, mostrou que o melhor polimento alcançado pela porcelana Vitadur Alpha foi através do polimento do Grupo C, o que está de acordo com os resultados obtidos pelo rugosímetro. / SUMMARY Aesthetic dentistry uses materials that can copy some lost caractheristics of the natural dentition, as color, mechanical resistance and smoothness.The better material for this are the ceramics. On the other hand, it is commum to have to grind them with diamond instruments for adjustment wich removes the glazed surfaces. This procedure can cause abrasion of the opposite teeth, plaque retention, injury of surrounding tissues and low aesthetics qualities. This study evaluated the roughness average in two dental ceramics, Omega 900 e Vitadur Alpha, that were grinding with diamond instruments after the glaze technique, and received three different polishing systems. The control group remained glazed until the end of the research. The groups were: Group A – control; Group B – grinding by fine and ultrafine points + polishing with DFL rubber points + KOTA diamond paste; Group C –grinding by fine and ultrafine points + polishing with rubber points + felt discs + Porcelize diamond paste (Cosmedent); Group D - grinding by fine and ultrafine points + polishing with rubber points + Porcelize + Flexibuff discs + Enamelize polishing paste (Cosmedent). To measure the roughness average in the samples it was used a profilometer Surftest SV-600 with the software Surfpack and the images were observed in the scanning electron microscopy (SEM) to comparation. The results received estatistical analysis using ANOVA and Tukey test (p˜0.05), and a significant tatistical difference was observed between the polishing systems , and in the porcelain/polishing interaction, which showed that all the polishing systems decreased the surface roughness of the porcelains after been grinding with diamond instruments, which was a better result than the control group. In the SEM analysis it was clear the presence of a partial relationship between the profilometer and that one, because here the Omega porcelain showed a more smoothness surface with the glazed treatment compaired with the others, differently from the profilometer analysis

Decarbonization of the power sector is of great importance for the transition to a sustainable and low-carbon world economy. Estimating carbon efficiency in the power sector is a key step to grasp the impact of demand-side usage changes and evaluate their potential environmental benefits. In order to quantify the environmental benefits of demand-side usage changes, Average Emission Factor (AEF) and Marginal Emission Factor (MEF) have been proposed in the electrical power sector. AEF is defined as the ratio of the total CO2 emitted in the system to the total electricity generated. It is an effective factor for reporting on CO2 emissions at system level and on an average basis, but the current AEF model lacks clarity on the factors actually affecting the estimation. MEF is defined as the incremental change in carbon emissions as a result of a change in demand. However, previous MEF assessments did not consider key technical limitations, such as ramp-rate constraint for generators and network constraints, and carbon trading mechanisms. This thesis improves the estimation for both AEF and MEF and key achievements can be summarized as: 1). A novel model of estimating AEF, with its application to GB, US and China’s electricity system. 2). Improvement on conventional MEF model by considering ramp-rate constraint in dispatch order. 3). Sensitivity studies on MEF using current fuel prices and future fuel prices. 4). A new model of estimating MEF considering both the utilization level of generators and the carbon costs when determining the dispatch order. 5). The effect of power network on MEF estimation, with a comparison of congested scenarios and non-congested scenario.

"In this thesis, based on the existing researches on power factor correction technology, I analyze, design and study the Boost type power factor correction technology, which is applied in the in-board two-stage battery charger. First I analyzed the basic working principle of the active power factor corrector. By comparing several different topologies of PFC converter main circuit and control methods, I specified the research object to be the average current control (ACM) boost power factor corrector. Then I calculated and designed the PFC circuit and the ACM controller applied in the first level charging of EVs. And I run the design in Simulink and study the important features like power factor, the input current waveform and the output DC voltage and the THD and odd harmonic magnitude."

O objetivo do presente trabalho foi avaliar o desempenho reprodutivo de leitoas que foram selecionadas com baixos ganhos de peso diários (GPD) aos 160 dias de vida. Leitoas da linhagem Camborough® (n = 3301) foram alocadas em quatro grupos de acordo com o GPD no momento da seleção: G1 (n = 799) leitoas com GPD ≥480 a ≤ 530g/dia, G2 (n = 978) leitoas com >530 a ≤ 580g/dia, G3 (n = 916) fêmeas com >580 a ≤ 630g/dia e G4 (n = 608) leitoas com >630 a ≤ 810g/dia. Não foram encontradas diferenças na idade de início de estimulação com o macho e manifestação do primeiro estro entre as classes (P> 0,05). Na cobertura todos os grupos diferiram entre si sendo o G4 o grupo mais precoce e com maior peso (P< 0,05). O GPD da seleção até a cobertura diferiu entre as classes, sendo que o G1 teve maiores ganhos de peso quando comparado ao G3 e G4 respectivamente (P< 0,05). As médias de nascidos totais, nascidos vivos e intervalo desmame-estro não diferiram entre os grupos (P> 0,05). As remoções entre a seleção e a cobertura pós-desmame foram maiores no G1 e G2 quando comparados ao G4 (P< 0,05). Quando agrupados G1 e G2 e, a partir disso, formadas duas classes de acordo com o peso na cobertura (≤ 130 kg e > 130 kg), o grupo > 130 kg obteve 0,93 leitão a mais no primeiro parto (P< 0,05). Em conclusão, leitoas selecionadas com baixos GPDs (inferiores a 630g/dia) podem ser incorporadas no plantel sem gerar prejuízos no desempenho reprodutivo até o primeiro parto, desde que alcancem 130 kg no momento da cobertura. / The aim of this study was to evaluate the reproductive performance of gilts selected with low average daily gain (ADG) at 160 days of age. Camborough® gilts (n = 3301) were allocated into four groups according to ADG at selection: G1 (n = 799) represented the group of gilts with ADG ≥480-≤ 530g/day, G2 (n = 978) corresponded to gilts with >530- ≤ 580g/day, G3 (n = 916) represented the gilts with > 580-≤ 630g/day and G4 (n = 608) gilts with > 630-≤ 810g/day. There were no differences in the age at the onset of boar exposure and first estrus expression among groups (P> 0.05). Age and weight at insemination were different (P< 0.05) among groups. Gilts from G4 were inseminated earlier and heavier. The ADG from selection until mating was different among the groups, and G1 had greater weight gains when compared to G3 and G4 (P< 0.05). Total number of piglets born, number of piglets born alive and weaning to oestrus interval were not different among classes (P> 0.05). The removals since the selection until post-weaning insemination were higher in G1 and G2 when compared to G4 (P< 0.05). When gilts from G1 and G2 were grouped and categorized into two classes according to the weight at insemination (≤ 130 kg and > 130 kg), the heavier group (> 130 kg) had an increasing of 0.93 total born piglets at first farrowing (P< 0.05). The results of this study suggest that gilts selected with low ADG (less than 630g/day) could be incorporated into the breeding herd with no reproductive performance losses until the first farrowing if they reach 130 kg at insemination.

O entendimento das relações entre as características do pasto e as respostas do comportamento ingestivo animal é fundamental para o aprimoramento de inovações no manejo de pastagens, tornando os sistemas pastoris mais eficientes e competitivos. Neste estudo comparamos dois métodos de pastoreio (Rotativo x Contínuo) e duas intensidades de pastejo (Moderada e Baixa) a fim de compreendermos como as características de pastos de azevém anual (Lolium multiflorum Lam.) influenciam o comportamento ingestivo e o ganho médio diário de ovinos. Observamos que o método de pastoreio afeta a quantidade de folhas e inflorescências, influenciando a composição química da forragem, além da taxa de bocados e o tempo de pastejo dos ovinos. Por outro lado, as intensidades de pastejo influenciaram os bocados por estação alimentar e o número de refeições. Além disso, variáveis como: estação alimentar por minuto, duração das refeições e tempo de pastejo foram positivamente correlacionadas com o ganho médio diário dos animais, assim, podem ser utilizadas como preditoras do ganho médio diário de ovinos. Encontramos evidências de que em pastoreio rotativo, ambas as intensidades de pastejo estudadas promoveram um rebaixamento em torno de 45% da altura pré-pastejo. A estrutura do pasto e a composição química da forragem foram adequadas para o pastejo de ovinos até o início do estádio reprodutivo do azevém anual. No entanto, ao final do estádio reprodutivo da pastagem, ambas as intensidades de pastejo restringiram os padrões de comportamento ingestivo dos ovinos nas maiores escalas espaço-temporais. Essa restrição é acentuada durante o rebaixamento do pasto, em decorrência da redução no percentual de lâminas foliares, alterando o comportamento ingestivo de ovinos. Logo, a resposta comportamental de ovinos depende de: (i) taxa de lotação, (ii) nível de desfolha do dossel, (iii) período do estádio reprodutivo do azevém anual e composição estrutural da forragem durante o rebaixamento. / The understanding of the relationships between sward characteristics and the animals' ingestive behavior responses is fundamental for the improvement of sward management innovations, making pastoral systems more efficient and competitive. In this study, we compared two stocking methods (Continuous and Rotational) and two grazing intensities (Moderate and Low) to understand how the sward characteristics of annual ryegrass (Lolium multiflorum Lam.) influence the ingestive behavior and the average daily gain of sheep. We observed that the stocking method affect the quantity of leaf blades and inflorescences influenced forage chemical composition, as well as the bite rate and grazing time of sheep. On the other hand, grazing intensities had influence the bites by feeding station and meals number. Moreover, variables as feeding station per minute, meals duration and grazing time were positively correlated with the average daily gain, thus, they can be used as predictors for the average daily gain of sheep. We found evidence that both grazing intensities studied promote a grazing down around 45% of pre-grazing sward height in rotational stocking. Sward structure and forage chemical composition, were adequate for sheep grazing until the beginning of the sward reproductive stage. However, from the stage 2 of grazing down at the end of the sward reproductive stage, both grazing intensities apparently restricted the sheep‟s ingestive behavior patterns in the larger spatial and temporal scales of grazing. Therefore sheep‟s behavioral response depends on: (i) stocking rate, (ii) canopy defoliation level, (iii) period of sward reproductive stage and forage structural composition during grazing down.

Orientador: Mauro Dal Secco de Oliveira / Banca: Jane Maria Bertocco Ezequiel / Banca: Atushi Sugohara / Banca: André de Faria Pedroso / Banca: Flávia Fernanda Simili / Resumo: Este estudo objetivou avaliar o armazenamento da cana-de-açúcar hidrolisada com 0,5% de cal hidratada sobre o consumo das frações alimentares, desempenho animal, viabilidade econômica, comportamento ingestivo, características nutritivas da forragem e aspectos da digestão de novilhas confinadas. Foram realizados dois experimentos. No primeiro (Capítulo 2 e 3), foram utilizadas 24 novilhas da raça Nelore distribuídas inteiramente ao acaso nos tratamentos formados por dietas com cana-de-açúcar in natura (tempo zero), cana-de-açúcar hidrolisada e armazenada por 24, 48 ou 72 horas. No segundo (Capítulo 4), foram utilizadas quatro novilhas da raça Nelore delineadas num quadrado latino 4x4, onde foram avaliadas as dietas anteriormente descritas. No primeiro experimento, verificou-se variação cúbica (P < 0,05) em função do tempo de armazenamento da cana-de-açúcar da maioria das variáveis relacionadas ao consumo das frações alimentares e desempenho animal, apresentando melhores resultados para novilhas alimentadas com cana-de-açúcar in natura. O armazenamento da cana-de-açúcar até 72 horas proporcionou aumento linear (P < 0,05) da fração de carboidratos potencialmente degradáveis e diminuição dos custos de produção. Novilhas alimentadas com dietas a base de cana-de-açúcar in natura apresentaram maior tempo de ruminação (P < 0,05). No segundo experimento, verificou-se que o consumo e digestibilidade dos nutrientes, o balanço de nitrogênio, a eficiência de síntese de proteína microbiana e o nitrogênio amoniacal não foram influenciados (P > 0,05) pelas dietas. O armazenamento da cana-de-açúcar com adição de 0,5% de cal hidratada até 72 horas deprime o consumo e desempenho de novilhas jovens da raça Nelore / Abstract: This study evaluated the effect of storage of hydrolyzed sugarcane with .5% hydrated lime the dietary fractions intake, animal performance, economic viability, ingestive behavior, nutritional characteristics of forage and digestion aspects of confined heifers. Two experiments were conducted. In the first (Chapter 2 and 3), 24 Nellore heifers were used and randomly distributed in treatment formed by diets constituted by fresh sugarcane (zero time), hydrolyzed sugarcane and stored for 24, 48 or 72 hours. In the second (Chapter 4), four heifers were used constituting a 4x4 latin square, with similar diets previous. In the first experiment, there was variation cubic (P < .05) according the storage time of most variables related to dietary fractions intake and animal performance, with better results for heifers fed with fresh sugarcane. Storage of sugarcane until 72 hours promoted linear increased (P < .05) in the carbohydrate fraction potentially degradable and decreased production costs. Heifers fed diets with fresh sugarcane presented higher time ruminating (P < .05). In the second experiment, it was found that the intake and digestibility of nutrients, the nitrogen balance, the efficiency of microbial protein synthesis and the ammonia nitrogen were not affected (P > .05) by diets. The storage of hydrolyzed sugarcane with .5% of hydrated lime up to 72 hours depresses intake and performance of young Nellore heifers / Doutor

This thesis investigates the average consensus of multi-agent systems with linear dynamics whose interconnections are modelled by balanced digraphs with matrix- weights. The thesis first introduces the notion of balanced digraphs and mirror graphs for matrix weights. Then it proves that the matrix-weight-balanced con- sensus controller is indeed globally asymptotically stable. The Lyapunov stability analysis exploits the properties of the mirror graph of a balanced digraph. Further, the necessary and sufficient condition for the system to achieve average consensus is shown to be positive definiteness of the matrix weights of its balanced digraph. Simulations with robots in SIMULINK verify that positive definite matrix weights on balanced graphs are indeed necessary and sufficient for average consensus. Fi- nally formation control of a multi-robot system is shown to be an application of the matrix-weight-balanced consensus algorithm using real time simulation of Clearpath Ridgeback robots in Gazebo and ROS. / Graduate