The objective of this thesis is to design a system that can be used to retrieve the music songs by acoustic input. The system listens to the melody or the partial song singing by the Karaoke users, and then prompts them the whole song paragraphs. Note segmentation is completed by both the magnitude of the song and the k-Nearest Neighbor technique. In order to speed up our system, the pitch period estimation algorithm is rewritten by a theory in communications. Besides, a large popular music database is built to make this system more practical.

Age related macular degeneration (AMD) is a public health concern in an aging society. The retinal pigment epithelium (RPE) layer of the eye is a principal site of pathogenesis for AMD. Morphological characteristics of the cells in the RPE layer can be used to discriminate age and disease status of individuals. In this thesis three genotypes of mice of various ages are used to study the predictive abilities of these characteristics. The disease state is represented by two mutant genotypes and the healthy state by the wild-type. Classification analysis is applied to the RPE morphology from the different spatial regions of the RPE layer. Variable reduction is accomplished by principal component analysis (PCA) and classification analysis by the k-nearest neighbor (k-NN) algorithm. In this way the differential ability of the spatial regions to predict age and disease status by cellular variables is explored.

Submitted by Fernando Souza (fernandoafsou@gmail.com) on 2017-08-15T15:15:35Z No. of bitstreams: 1 arquivototal.pdf: 3344501 bytes, checksum: 9deb01db1972288d73b0c48155123f90 (MD5) / Made available in DSpace on 2017-08-15T15:15:35Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 3344501 bytes, checksum: 9deb01db1972288d73b0c48155123f90 (MD5) Previous issue date: 2016-02-19 / Currently, unmanned aerial vehicles (UAV) are increasingly used to aid the various tasks around the world. The popularization of this equipment associated with the advancement of technology, particularly the miniaturization of processors, extend their functionalitys. In agricultural applications, these devices allow monitoring of production by capturing aerial images, for which are processed and identified areas of interest through specific software. The research proposes a low-cost solution capable of processing aerial images obtained by non-metric digital cameras coupled to UAV to identify gaps in plantations or estimate levels of environmental degradation, which can be deployed in small computers and low power consumption. Embedded systems coupled in UAV allow perform processing in real time, which contributes to a preventive diagnosis, reduces the response time and can avoid damages in the crop. The algorithm used is based on watershed, while the second algorithm uses classification techniques based on the 1-Nearest Neighbor (1-NN). Are used the embedded systems DE2i-150 and Intel Edison, both x86 architecture, and Raspberry Pi 2 of ARM architecture. Moreover, the technique 1-NN showed higher tolerance to lighting problems, however, require more processing power compared to the algorithm based on watershed. The results show that the proposed system is an efficient and relatively low-cost solution compared to traditional means of monitoring and can be coupled in a UAV to perform the processing during the flight. / Atualmente, veículos aéreos não tripulados (VANT) são cada vez mais utilizados no auxílio a diversas tarefas em todo o mundo. A popularização destes equipamentos associada ao avanço da tecnologia, sobretudo a miniaturização de processadores, ampliam suas funcionalidades. Em aplicações agrícolas, estes equipamentos permitem o monitoramento da produção por meio da captação de imagens aéreas, a partir dos quais são processadas e identificadas áreas de interesse por meio de softwares específicos. A pesquisa propõe uma solução de baixo custo capaz de processar imagens aéreas obtidas por câmeras digitais não métricas acopladas a VANT para identificar falhas em plantações ou estimar níveis de degradação ambiental, os quais possam ser implantados em computadores de pequeno porte e baixo consumo, conhecido como sistemas embarcados. Plataformas embarcadas acopladas a VANT permitem realizar o processamento em tempo real, que contribui para um diagnóstico preventivo, reduz o tempo de resposta e pode evitar prejuízos na lavoura. O algoritmo inicialmente avaliado é baseado em watershed, enquanto que o segundo algoritmo proposto faz uso de técnicas de classificação baseada no 1-vizinho mais próximo (1-NN). Utilizam-se os sistemas embarcados DE2i-150 e Intel Edison, ambos de arquitetura x86, e a plataforma Raspberry Pi 2 de arquitetura ARM. Em relação ao processamento das imagens são alcançados níveis de acurácia em torno de 90%, com uso do algoritmo baseado em 1-NN. Além disso, a técnica 1-NN apresentou maior tolerância aos problemas de luminosidade, em contrapartida, demandam maior poder de processamento quando comparados com o algoritmo baseado em watershed. Os resultados mostram que o sistema proposto é uma solução eficiente e de custo relativamente baixo em comparação com os meios tradicionais de monitoramento e pode ser acoplada em um VANT para realizar o processamento durante o voo.

During production of certain biopharamaceutical drugs, cells are grown in a liquid mediainside bioreactors with the goal of producing a specic biomaterial that can be rened intoa drug. This project investigates whether the use of Neural Networks (NN) can decreasethe prediction error, in terms of Mean Squared Error (MSE), for 2 metabolic processes incells compared to current methods. The rst experiment tests predictions of cell-SpecicConsumption Rate (SCR) of 5 dierent metabolites and the second experiment testspredictions of cell-Specic Production Rate (SPR) of titer. Fully connected feed-forwardneural networks were trained and cross-validation was used to obtain MSE betweenpredictions and measured values. The SCR predictions made by the NN was better thanthe original model predictions for all 5 metabolites. The predictions of SPR from the NNcannot with certainty be said to be better than the original model, with a p-value of 0.13.These results indicate that using NNs when modeling cell metabolism in bioreactors candecrease its prediction error, leading to better control of the bioreactor environment andmore ecient production.

<div>In this work, we introduce a framework for constructing light-weight neural network approximations to the solution operators for partial differential equations (PDEs). Using a data-driven offline training procedure, the resulting operator network models are able to effectively reduce the computational demands of traditional numerical methods into a single forward-pass of a neural network. Importantly, the network models can be calibrated to specific distributions of input data in order to reflect properties of real-world data encountered in practice. The networks thus provide specialized solvers tailored to specific use-cases, and while being more restrictive in scope when compared to more generally-applicable numerical methods (e.g. procedures valid for entire function spaces), the operator networks are capable of producing approximations significantly faster as a result of their specialization.</div><div><br></div><div>In addition, the network architectures are designed to place pointwise posterior distributions over the observed solutions; this setup facilitates simultaneous training and uncertainty quantification for the network solutions, allowing the models to provide pointwise uncertainties along with their predictions. An analysis of the predictive uncertainties is presented with experimental evidence establishing the validity of the uncertainty quantification schema for a collection of linear and nonlinear PDE systems. The reliability of the uncertainty estimates is also validated in the context of both in-distribution and out-of-distribution test data.</div><div><br></div><div>The proposed neural network training procedure is assessed using a novel convolutional encoder-decoder model, ConvPDE-UQ, in addition to an existing fully-connected approach, DeepONet. The convolutional framework is shown to provide accurate approximations to PDE solutions on varying domains, but is restricted by assumptions of uniform observation data and homogeneous boundary conditions. The fully-connected DeepONet framework provides a method for handling unstructured observation data and is also shown to provide accurate approximations for PDE systems with inhomogeneous boundary conditions; however, the resulting networks are constrained to a fixed domain due to the unstructured nature of the observation data which they accommodate. These two approaches thus provide complementary frameworks for constructing PDE-based operator networks which facilitate the real-time approximation of solutions to PDE systems for a broad range of target applications.</div>

Fuzzy ARTMAP (FAM) is currently considered to be one of the premier neural network architectures in solving classification problems. One of the limitations of Fuzzy ARTMAP that has been extensively reported in the literature is the category proliferation problem. That is Fuzzy ARTMAP has the tendency of increasing its network size, as it is confronted with more and more data, especially if the data is of noisy and/or overlapping nature. To remedy this problem a number of researchers have designed modifications to the training phase of Fuzzy ARTMAP that had the beneficial effect of reducing this phenomenon. In this thesis we propose a new approach to handle the category proliferation problem in Fuzzy ARTMAP by evolving trained FAM architectures. We refer to the resulting FAM architectures as GFAM. We demonstrate through extensive experimentation that an evolved FAM (GFAM) exhibits good (sometimes optimal) generalization, small size (sometimes optimal size), and requires reasonable computational effort to produce an optimal or sub-optimal network. Furthermore, comparisons of the GFAM with other approaches, proposed in the literature, which address the FAM category proliferation problem, illustrate that the GFAM has a number of advantages (i.e. produces smaller or equal size architectures, of better or as good generalization, with reduced computational complexity). Furthermore, in this dissertation we have extended the approach used with Fuzzy ARTMAP to other ART architectures, such as Ellipsoidal ARTMAP (EAM) and Gaussian ARTMAP (GAM) that also suffer from the ART category proliferation problem. Thus, we have designed and experimented with genetically engineered EAM and GAM architectures, named GEAM and GGAM. Comparisons of GEAM and GGAM with other ART architectures that were introduced in the ART literature, addressing the category proliferation problem, illustrate similar advantages observed by GFAM (i.e, GEAM and GGAM produce smaller size ART architectures, of better or improved generalization, with reduced computational complexity). Moverover, to optimally cover the input space of a problem, we proposed a genetically engineered ART architecture that combines the category structures of two different ART networks, FAM and EAM. We named this architecture UART (Universal ART). We analyzed the order of search in UART, that is the order according to which a FAM category or an EAM category is accessed in UART. This analysis allowed us to better understand UART's functionality. Experiments were also conducted to compare UART with other ART architectures, in a similar fashion as GFAM and GEAM were compared. Similar conclusions were drawn from this comparison, as in the comparison of GFAM and GEAM with other ART architectures. Finally, we analyzed the computational complexity of the genetically engineered ART architectures and we compared it with the computational complexity of other ART architectures, introduced into the literature. This analytical comparison verified our claim that the genetically engineered ART architectures produce better generalization and smaller sizes ART structures, at reduced computational complexity, compared to other ART approaches. In review, a methodology was introduced of how to combine the answers (categories) of ART architectures, using genetic algorithms. This methodology was successfully applied to FAM, EAM and FAM and EAM ART architectures, with success, resulting in ART neural networks which outperformed other ART architectures, previously introduced into the literature, and quite often produced ART architectures that attained optimal classification results, at reduced computational complexity.

Test Automation is becoming a more vital part of the software development cycle, as it aims to lower the cost of testing and allow for higher test frequency. However, automating manual tests can be difficult as they tend to require complex human interaction. In this thesis, we aim to solve this by using Imitation Learning as a tool for automating manual software tests. The software under test consists of a virtual simulation, connected to a physical input device in the form of a sight. The sight can rotate on two axes, yaw and pitch, which require human motion control. Based on this, we use a Behavioral Cloning approach with a k-NN regressor trained on human demonstrations. Evaluation of model resemblance to the human is done by comparing the state path taken by the model and human. The model task performance is measured with a score based on the time taken to stabilize the sight pointing at a given object in the virtual world. The results show that a simple k-NN regression model using high-level states and actions, and with limited data, can imitate the human motion well. The model tends to be slightly faster than the human on the task while keeping realistic motion. It also shows signs of human errors, such as overshooting the object at higher angular velocities. Based on the results, we conclude that using Imitation Learning for Test Automation can be practical for specific tasks, where capturing human factors are of importance. However, further exploration is needed to identify the full potential of Imitation Learning in Test Automation.

Résumé en français non disponible

In this paper, we propose a novel similarity measure for document data processing and apply it to text classification and clustering. For two documents, the proposed measure takes three cases into account: (a) The feature considered appears in both documents, (b) the feature considered appears in only one document, and (c) the feature considered appears in none of the documents. For the first case, we give a lower bound and decrease the similarity according to the difference between the feature values of the two documents. For the second case, we give a fixed value disregarding the magnitude of the feature value. For the last case, we ignore its effectiveness. We apply it to the similarity based single-label classifier k-NN and multi-label classifier ML-KNN, and adopt these properties to measure the similarity between a document and a specific set for document clustering, i.e., k-means like algorithm, to compare the effectiveness with other measures. Experimental results show that our proposed method can work more effectively than others.

La chimie de coordination de complexes cuivreux du type [Cu(PP)(NN)]+ (PP : ligand bisphosphine et NN : ligand phénanthroline) a été étudiée. La stabilité des complexes hétéroleptiques dépend de la nature du ligand PP. Des complexes hétéroleptiques stables et fortement luminescents ont été obtenus en utilisant le ligand POP (bis[2-diphénylphosphinophényl]éther). Lors de la préparation de cette première famille de complexes cuivreux, nous avons découvert fortuitement qu'il était aussi possible d'obtenir des complexes de type [Cu(PP)2][BF4]. Cette nouvelle famille de composés possède des propriétés de luminescence remarquable qui ont été exploitées pour l'élaboration de diodes électroluminescentes. Finalement, nous nous sommes également intéressés à l'élaboration de matériaux donneur-accepteur en associant les fullerènes avec nos complexes du Cu(I).