No description available.

As technology improves, scientists are able to generate high-throughput data faster and cheaper. Consequently, the field of biological sciences is progressively becoming more reliant on data science tools like machine learning methods for analysis and sorting of big data. The Complex Object Parametric Analyzer and Sorter (COPAS) is a large particle flow cytometer that can perform high-throughput fluorescence screens on small animals, like Caenorhabditis elegans. The outputs of the COPAS are extinction coefficient (EXT), Time of Flight (TOF, arbitrary length unit) and fluorescence. However, the COPAS outputs include unwanted objects like bubbles or bacteria and some animals pass the flow cell in a non-straight manner producing abnormal profiles leading to inaccurate developmental staging. In this thesis, I have created an R package, named COPASProfiler, that generates experiment-specific supervised machine learning (ML) classification models which can detect and remove abnormal profiles enabling standardized fluorescence quantification and analysis. I used COPASProfiler to develop a pipeline to automate fluorescence analysis of high-throughput COPAS data sets. Using R shiny, I created a web program with a graphical user interface that allows users to view, annotate, quantify fluorescence, and classify COPAS-generated datasets. The COPASProfiler is available on GitHub and can be installed using one single R command. Lastly, the COPASProfiler comes with multiple tutorials and examples, and was designed to accommodate users with minimal programming experience. COPASProfiler should enable robust high-throughput fluorescence studies of regulatory elements (e.g., enhancers, promoters, and 3’UTRs) and long-term epigenetic silencing in C. elegans.

Yes / Enterprises and individual users heavily rely on the abilities of antiviruses and other security mechanisms. However, the methodologies used by such software are not enough to detect and prevent most of the malicious activities and also consume a huge amount of resources of the host machine for their regular oper- ations. In this paper, we propose a combination of machine learning techniques applied on a rich set of features extracted from a large dataset of benign and malicious les through a bespoke feature extraction tool. We extracted a rich set of features from each le and applied support vector machine, decision tree, and boosting on decision tree to get the highest possible detection rate. We also introduce a cloud-based scalable architecture hosted on Amazon web services to cater the needs of detection methodology. We tested our methodology against di erent scenarios and generated high achieving results with lowest energy con- sumption of the host machine.

As of 2016, diagnosis of ADHD in the US is controversial. Diagnosis of ADHD is based on subjective observations, and treatment is usually done through stimulants, which can have negative side-effects in the long term. Evidence shows that the probability of diagnosing a child with ADHD not only depends on the observations of parents, teachers, and behavioral scientists, but also on state-level special education policies. In light of these facts, unbiased, quantitative methods are needed for the diagnosis of ADHD. This problem has been tackled since the 1990s, and has resulted in methods that have not made it past the research stage and methods for which claimed performance could not be reproduced. This work proposes a combination of machine learning algorithms and signal processing techniques applied to EEG data in order to classify subjects with and without ADHD with high accuracy and confidence. More specifically, the K-nearest Neighbor algorithm and Gaussian-Mixture-Model-based Universal Background Models (GMM-UBM), along with autoregressive (AR) model features, are investigated and evaluated for the classification problem at hand. In this effort, classical KNN and GMM-UBM were also modified in order to account for uncertainty in diagnoses. Some of the major findings reported in this work include classification performance as high, if not higher, than those of the highest performing algorithms found in the literature. One of the major findings reported here is that activities that require attention help the discrimination of ADHD and Non-ADHD subjects. Mixing in EEG data from periods of rest or during eyes closed leads to loss of classification performance, to the point of approximating guessing when only resting EEG data is used. / Master of Science / As of 2016, diagnosis of ADHD in the US is controversial. Diagnosis of ADHD is based on subjective observations, and treatment is usually done through stimulants, which can have negative side-effects in the long term. Evidence shows that the probability of diagnosing a child with ADHD not only depends on the observations of parents, teachers, and behavioral scientists, but also on state-level special education policies. In light of these facts, unbiased, quantitative methods are needed for the diagnosis of ADHD. This problem has been tackled since the 1990s, and has resulted in methods that have not made it past the research stage and methods for which claimed performance could not be reproduced. This work proposes a combination of machine learning algorithms and signal processing techniques applied to EEG data in order to classify subjects with and without ADHD with high accuracy and confidence. Signal processing techniques are used to extract autoregressive (AR) coefficients, which contain information about brain activities and are used as “features”. Then, the features, extracted from datasets containing ADHD and Non-ADHD subjects, are used to create or train models that can classify subjects as either ADHD or Non-ADHD. Lastly, the models are tested using datasets that are different from the ones used in the previous stage, and performance is analyzed based on how many of the predicted labels (ADHD or Non-ADHD) match the expected labels. Some of the major findings reported in this work include classification performance as high, if not higher, than those of the highest performing algorithms found in the literature. One of the major findings reported here is that activities that require attention help the discrimination of ADHD and Non-ADHD subjects. Mixing in EEG data from periods of rest or during eyes closed leads to loss of classification performance, to the point of approximating guessing when only resting EEG data is used.

Nos últimos anos, tem havido um crescente interesse na detecção precoce das doenças que afetam as culturas agrícolas a fim de evitar grandes perdas econômicas devido à contaminação de novas plantas. As principais doenças cítricas, cancro cítrico e greening, são uma séria ameaça à produção de citros em todo o mundo, incluindo regiões do Brasil e dos Estados Unidos. A disseminação rápida das doenças leva à redução do número de pomares cultivados, resultando em danos econômicos aos produtores e às indústrias relacionadas. O desenvolvimento de métodos para o diagnóstico precoce pode resultar em uma importante ferramenta para o controle e gestão dos citros. Algumas deficiências nutricionais como a de ferro e zinco apresentam sintomas visuais semelhantes com o greening, enquanto que o cancro cítrico pode ser confundido com a verrugose ou leprose dos citros, podendo levar ao diagnóstico incorreto. Atualmente, somente testes bioquímicos são capazes de detectar especificamente o cancro cítrico e o greening, e consequentemente diferenciá-los das demais doenças e deficiências de nutrientes. No presente trabalho, a técnica de espectroscopia por imagens de fluorescência em conjunto com os métodos de aprendizado e classificação, SVM (do inglês, Support Vector Machine) e ANN (do inglês, Artificial Neural Network), foram utilizadas a fim de identificar e discriminar as principais doenças que afetam a citricultura nos estados de São Paulo/Brasil e da Flórida/EUA. As amostras em estudo são cancro cítrico, verrugose, greening e deficiência de zinco. O objetivo principal é discriminar as doenças com sintomas visuais semelhantes, no caso, cancro cítrico de verrugose e greening de deficiência de zinco para as amostras do Brasil, e greening de deficiência de zinco para as amostras dos Estados Unidos. Os resultados mostram que é possível utilizar a técnica de espectroscopia por imagens de fluorescência em conjunto com os métodos de classificação na discriminação das doenças que apresentam sintomas visuais semelhantes. Ambos os classificadores apresentaram uma elevada precisão na classificação tanto das amostras do Brasil como dos Estados Unidos, destacando assim eficácia da técnica sob condições diferentes. / In recent years, there has been an increasing interest in early detection of diseases that affect agricultural crops to avoid great economic losses due to contamination of new plants. The main citrus diseases, citrus canker and HLB, are a serious threat to citrus production worldwide, including regions in Brazil and the United States. The rapid spread of the diseases leads to the reduction of cultivated orchards resulting in economic losses to producers and industries. The development of methods for early diagnosis can result in an important tool for the control and management of citrus. Some nutritional deficiencies such as iron and zinc have similar visual symptoms to HLB, while the citrus canker can be confused with citrus scab and citrus leprosies, which may lead to incorrect diagnosis. Currently, only biochemical tests are able to detect, specifically, citrus canker and HLB, and thus distinguish them from other diseases and nutrient deficiencies. In this work, the fluorescence imaging spectroscopy technique with the learning and classification methods, SVM (Support Vector Machine) and ANN (Artificial Neural Network), were used to identify and discriminate the main diseases that affect citrus production in the states of São Paulo/Brazil and Florida/USA. The samples studied are citrus canker, citrus scab, HLB and zinc deficiency. The objective is to discriminate the diseases with similar visual symptoms, such as citrus canker from citrus scab and HLB from zinc deficiency for samples from Brazil and HLB from zinc deficiency for samples from the United States. The results show that it is possible to use the fluorescence imaging spectroscopy technique together with the classification methods for the discrimination of the diseases that have similar visual symptoms. Both classifiers showed high accuracy in the classification of the samples from Brazil and the United States, highlighting the efficiency of the technique under different conditions.

Sistemas baseados em processadores de uso geral caracterizam-se pela flexibilidade a mudanças de projeto, porém com desempenho computacional abaixo daqueles baseados em circuitos dedicados otimizados. A implementação de algoritmos em dispositivos reconfiguráveis, conhecidos como Field Programmable Gate Arrays - FPGAs, oferece uma solução de compromisso entre a flexibilidade dos processadores e o desempenho dos circuitos dedicados, pois as FPGAs permitem que seus recursos de hardware sejam configurados por software, com uma granularidade menor que a do processador de uso geral e flexibilidade maior que a dos circuitos dedicados. As versões atuais de FPGAs apresentam um tempo de reconfiguração suficientemente pequeno para viabilizar sua reconfiguração dinâmica, i.e., mesmo com o dispositivo executando um algoritmo, a forma como seus recursos são dispostos pode ser alterada, oferecendo a possibilidade de particionar temporalmente um algoritmo. Novas linhas de FPGAs já são fabricadas com opção de reconfiguração dinâmica parcial, i.e., é possível reconfigurar áreas selecionadas de uma FPGA enquanto o restante continua em operação. No entanto, para que esta nova tecnologia se torne largamente difundida é necessário o desenvolvimento de uma metodologia própria, que ofereça soluções eficazes aos novos desdobramentos do projeto digital. Em particular, uma das principais dificuldades apresentadas por esta abordagem refere-se à maneira de particionar o algoritmo, de forma a minimizar o tempo necessário para completar sua tarefa. Este manuscrito oferece uma metodologia de projeto para dispositivos dinamicamente reconfiguráveis, com ênfase no problema do particionamento temporal de circuitos, tendo como aplicação alvo uma família de algoritmos, utilizados principalmente em Bioinformática, representada pelo classificador binário conhecido como Support Vector Machine. Algumas técnicas de particionamento para FPGA Dinamicamente Reconfigurável, especificamente aplicáveis ao particionamento de FSM, foram desenvolvidas para garantir que um projeto dominado por fluxo de controle seja mapeado numa única FPGA, sem alterar sua funcionalidade. / Systems based on general-purpose processors are characterized by a flexibility to design changes, although with a computational performance below those based on optimized dedicated circuits. The implementation of algorithms in reconfigurable devices, known as Field Programmable Gate Arrays, FPGAs, offers a solution with a trade-off between the processor\'s flexibility and the dedicated circuit\'s performance. With FPGAs it is possible to have their hardware resources configured by software, with a smaller granularity than that of the general-purpose processor and greater flexibility than that of dedicated circuits. Current versions of FPGAs present a reconfiguration time sufficiently small as to make feasible dynamic reconfiguration, i.e., even with the device executing an algorithm, the way its resources are displayed can be modified, offering the possibility of temporal partitioning of an algorithm. New lines of FPGAs are already being manufactured with the option of partial dynamic reconfiguration, i.e. it is possible to reconfigure selected areas of an FPGA anytime, while the remainder area continue in operation. However, in order for this new technology to become widely adopted the development of a proper methodology is necessary, which offers efficient solutions to the new stages of the digital project. In particular, one of the main difficulties presented by this approach is related to the way of partitioning the algorithm, in order to minimize the time necessary to complete its task. This manuscript offers a project methodology for dynamically reconfigurable devices, with an emphasis on the problem of the temporal partitioning of circuits, having as a target application a family of algorithms, used mainly in Bioinformatics, represented by the binary classifier known as Support Machine Vector. Some techniques of functional partitioning for Dynamically Reconfigurable FPGA, specifically applicable to partitioning of FSMs, were developed to guarantee that a control flow dominated design be mapped in only one FPGA, without modifying its functionality.

[pt] Esta dissertação faz uma comparação de três métodos de classificação em cascata de imagens multitemporais. Os classificadores se baseiam nas seguintes técnicas: (1) Máquina de Suporte Vetorial (SVM), (2) Modelos Ocultos de Markov (HMM) e (3) Cadeias de Markov Nebulosas(FMC). Para verificar a robustez dos modelos de classificação, introduziram-se nos dados de entrada outliers, avaliando-se assim, a robustez dos classificadores. Adicionalmente, avaliou-se o desempenho dos métodos quando a proporção de ocorrências de cada transição de classe no conjunto de treinamento difere da proporção no conjunto de teste. Determinou-se também qual o benefício do uso de conhecimento a priori sobre as transições possíveis. A análise experimental foi realizada sobre dois conjuntos de imagens de diferentes características, um par de imagens IKONOS do Rio de Janeiro, Brasil e um par de imagens LANDSAT7 de Alcinópolis, Mato Grosso do Sul. O estudo revelou que acurácia global das três abordagens tem um comportamento similar nos diferentes experimentos. Mostrou também que todas as três abordagens multitemporais apresentam desempenho superior aos seus homólogos monotemporais. / [en] This dissertation compares three cascade multitemporal image classification methods based on: (1) Support Vector Machines (SVM), (2) Hidden Markov Models (HMM) and (3) Fuzzy Markov Chains (FMC). The robustness of the classification models is verified, by introducing outliers in the data set. Additionally, performance of each method is evaluated when the number of occurrences of each class transition is different in the training and in the testing set. The gain of exploiting a prior knowledge regarding the admissible transitions in each target site is also investigated. The experimental analysis is conducted over two data sets with different characteristics; specifically a pair of IKONOS images of Rio de Janeiro and a pair of LANDSAT7 images of Alcinópolis, Mato Grosso do Sul. This study has concluded that the overall accuracy of the three approaches are similar through all experiments. The superiority ofthe multitemporal approaches over the monotemporal counterparts was confirmed.

Dans un contexte présentiel, un acte de communication comprend des expressions orales et émotionnelles. A partir de l’observation, du diagnostic et de l’identification de l’état émotionnel d’un individu, son interlocuteur pourra entreprendre des actions qui influenceraient la qualité de la communication. A cet égard, nous pensons améliorer la manière dont les individus perçoivent leurs échanges en proposant d’enrichir la CEMO (communication écrite médiatisée par ordinateur) par des émotions ressenties par les collaborateurs. Pour ce faire, nous proposons d’intégrer un système de reconnaissance, en temps réel, des émotions (joie, peur, surprise, colère, dégoût, tristesse, neutralité) à la plate-forme pédagogique “Moodle”, à partir de l’analyse des expressions faciales de l’apprenant à distance lors des activités collaboratives. La reconnaissance des expressions faciales se fait en trois étapes. Tout d’abord, le visage et ses composants (sourcils, nez, bouche, yeux) sont détectés à partir de la configuration de points caractéristiques. Deuxièmement, une combinaison de descripteurs hétérogènes est utilisée pour extraire les traits caractéristiques du visage. Finalement, un classifieur est appliqué pour classer ces caractéristiques en six émotions prédéfinies ainsi que l’état neutre. Les performances du système proposé seront évaluées sur des bases publiques d’expressions faciales posées et spontanées telles que Cohn-Kanade (CK), Karolinska Directed Emotional Faces (KDEF) et Facial Expressions and Emotion Database (FEED). / In face-to-face settings, an act of communication includes verbal and emotional expressions. From observation, diagnosis and identification of the individual's emotional state, the interlocutor will undertake actions that would influence the quality of the communication. In this regard, we suggest to improve the way that the individuals perceive their exchanges by proposing to enrich the textual computer-mediated communication by emotions felt by the collaborators. To do this, we propose to integrate a real time emotions recognition system in a platform “Moodle”, to extract them from the analysis of facial expressions of the distant learner in collaborative activities. There are three steps to recognize facial expressions. First, the face and its components (eyebrows, nose, mouth, eyes) are detected from the configuration of facial landmarks. Second, a combination of heterogeneous descriptors is used to extract the facial features. Finally, a classifier is applied to classify these features into six predefined emotions as well as the neutral state. The performance of the proposed system will be assessed on a public basis of posed and spontaneous facial expressions such as Cohn-Kanade (CK), Karolinska Directed Emotional Faces (KDEF) and Facial Expressions and Emotion Database (FEED).

La définition d'une métrique entre des séries temporelles est un élément important pour de nombreuses tâches en analyse ou en fouille de données, tel que le clustering, la classification ou la prédiction. Les séries temporelles présentent naturellement différentes caractéristiques, que nous appelons modalités, sur lesquelles elles peuvent être comparées, comme leurs valeurs, leurs formes ou leurs contenus fréquentielles. Ces caractéristiques peuvent être exprimées avec des délais variables et à différentes granularités ou localisations temporelles - exprimées globalement ou localement. Combiner plusieurs modalités à plusieurs échelles pour apprendre une métrique adaptée est un challenge clé pour de nombreuses applications réelles impliquant des données temporelles. Cette thèse propose une approche pour l'Apprentissage d'une Métrique Multi-modal et Multi-scale (M2TML) en vue d'une classification robuste par plus proches voisins. La solution est basée sur la projection des paires de séries temporelles dans un espace de dissimilarités, dans lequel un processus d'optimisation à vaste marge est opéré pour apprendre la métrique. La solution M2TML est proposée à la fois dans le contexte linéaire et non-linéaire, et est étudiée pour différents types de régularisation. Une variante parcimonieuse et interprétable de la solution montre le potentiel de la métrique temporelle apprise à pouvoir localiser finement les modalités discriminantes, ainsi que leurs échelles temporelles en vue de la tâche d'analyse considérée. L'approche est testée sur un vaste nombre de 30 bases de données publiques et challenging, couvrant des images, traces, données ECG, qui sont linéairement ou non-linéairement séparables. Les expériences montrent l'efficacité et le potentiel de la méthode M2TML pour la classification de séries temporelles par plus proches voisins. / The definition of a metric between time series is inherent to several data analysis and mining tasks, including clustering, classification or forecasting. Time series data present naturally several characteristics, called modalities, covering their amplitude, behavior or frequential spectrum, that may be expressed with varying delays and at different temporal granularity and localization - exhibited globally or locally. Combining several modalities at multiple temporal scales to learn a holistic metric is a key challenge for many real temporal data applications. This PhD proposes a Multi-modal and Multi-scale Temporal Metric Learning (M2TML) approach for robust time series nearest neighbors classification. The solution is based on the embedding of pairs of time series into a pairwise dissimilarity space, in which a large margin optimization process is performed to learn the metric. The M2TML solution is proposed for both linear and non linear contexts, and is studied for different regularizers. A sparse and interpretable variant of the solution shows the ability of the learned temporal metric to localize accurately discriminative modalities as well as their temporal scales.A wide range of 30 public and challenging datasets, encompassing images, traces and ECG data, that are linearly or non linearly separable, are used to show the efficiency and the potential of M2TML for time series nearest neighbors classification.

La segmentation est l'un des principaux thèmes du domaine de l'analyse d'images. Segmenter une image consiste à trouver une partition constituée de régions, c'est-à-dire d'ensembles de pixels connexes homogènes selon un critère choisi. L'objectif de la segmentation consiste à obtenir des régions correspondant aux objets ou aux parties des objets qui sont présents dans l'image et dont la nature dépend de l'application visée. Même s'il peut être très fastidieux, un tel découpage de l'image peut être facilement obtenu par un être humain. Il n'en est pas de même quand il s'agit de créer un programme informatique dont l'objectif est de segmenter les images de manière entièrement automatique. La segmentation interactive est une approche semi-automatique où l'utilisateur guide la segmentation d'une image en donnant des indications. Les méthodes qui s'inscrivent dans cette approche se divisent en deux catégories en fonction de ce qui est recherché : les contours ou les régions. Les méthodes qui recherchent des contours permettent d'extraire un unique objet correspondant à une région sans trou. L'utilisateur vient guider la méthode en lui indiquant quelques points sur le contour de l'objet. L'algorithme se charge de relier chacun des points par une courbe qui respecte les caractéristiques de l'image (les pixels de part et d'autre de la courbe sont aussi dissemblables que possible), les indications données par l'utilisateur (la courbe passe par chacun des points désignés) et quelques propriétés intrinsèques (les courbes régulières sont favorisées). Les méthodes qui recherchent les régions groupent les pixels de l'image en des ensembles, de manière à maximiser la similarité en leur sein et la dissemblance entre les différents ensembles. Chaque ensemble correspond à une ou plusieurs composantes connexes et peut contenir des trous. L'utilisateur guide la méthode en traçant des traits de couleur qui désignent quelques pixels appartenant à chacun des ensembles. Si la majorité des méthodes ont été conçues pour extraire un objet principal du fond, les travaux menés durant la dernière décennie ont permis de proposer des méthodes dites multiclasses, capables de produire une partition de l'image en un nombre arbitraire d'ensembles. La contribution principale de ce travail de recherche est la conception d'une nouvelle méthode de segmentation interactive multiclasse par recherche des régions. Elle repose sur la modélisation du problème comme la minimisation d'une fonction de coût pouvant être représentée par un graphe de facteurs. Elle intègre une méthode de classification par apprentissage supervisé assurant l'adéquation entre la segmentation produite et les indications données par l'utilisateur, l'utilisation d'un nouveau terme de régularisation et la réalisation d'un prétraitement consistant à regrouper les pixels en petites régions cohérentes : les superpixels. L'utilisation d'une méthode de sur-segmentation produisant des superpixels est une étape clé de la méthode que nous proposons : elle réduit considérablement la complexité algorithmique et permet de traiter des images contenant plusieurs millions de pixels, tout en garantissant un temps interactif. La seconde contribution de ce travail est une évaluation des algorithmes permettant de grouper les pixels en superpixels, à partir d'un nouvel ensemble de données de référence que nous mettons à disposition et dont la particularité est de contenir des images de tailles différentes : de quelques milliers à plusieurs millions de pixels. Cette étude nous a également permis de concevoir et d'évaluer une nouvelle méthode de production de superpixels. / Image segmentation is one of the main research topics in image analysis. It is the task of researching a partition into regions, i.e., into sets of connected pixels, meeting a given uniformity criterion. The goal of image segmentation is to find regions corresponding to the objects or the object parts appearing in the image. The choice of what objects are relevant depends on the application context. Manually locating these objects is a tedious but quite simple task. Designing an automatic algorithm able to achieve the same result is, on the contrary, a difficult problem. Interactive segmentation methods are semi-automatic approaches where a user guide the search of a specific segmentation of an image by giving some indications. There are two kinds of methods : boundary-based and region-based interactive segmentation methods. Boundary-based methods extract a single object corresponding to a unique region without any holes. The user guides the method by selecting some boundary points of the object. The algorithm search for a curve linking all the points given by the user, following the boundary of the object and having some intrinsic properties (regular curves are encouraged). Region-based methods group the pixels of an image into sets, by maximizing the similarity of pixels inside each set and the dissimilarity between pixels belonging to different sets. Each set can be composed of one or several connected components and can contain holes. The user guides the method by drawing colored strokes, giving, for each set, some pixels belonging to it. If the majority of region-based methods extract a single object from the background, some algorithms, proposed during the last decade, are able to solve multi-class interactive segmentation problems, i.e., to extract more than two sets of pixels. The main contribution of this work is the design of a new multi-class interactive segmentation method. This algorithm is based on the minimization of a cost function that can be represented by a factor graph. It integrates a supervised learning classification method checking that the produced segmentation is consistent with the indications given by the user, a new regularization term, and a preprocessing step grouping pixels into small homogeneous regions called superpixels. The use of an over-segmentation method to produce these superpixels is a key step in the proposed interactive segmentation method : it significantly reduces the computational complexity and handles the segmentation of images containing several millions of pixels, by keeping the execution time small enough to ensure comfortable use of the method. The second contribution of our work is an evaluation of over-segmentation algorithms. We provide a new dataset, with images of different sizes with a majority of big images. This review has also allowed us to design a new over-segmentation algorithm and to evaluate it.