Métodos baseados em grafos consistem em uma poderosa forma de representação e abstração de dados que proporcionam, dentre outras vantagens, representar relações topológicas, visualizar estruturas, representar grupos de dados com formatos distintos, bem como, fornecer medidas alternativas para caracterizar os dados. Esse tipo de abordagem tem sido cada vez mais considerada para solucionar problemas de aprendizado de máquina, principalmente no aprendizado não supervisionado, como agrupamento de dados, e mais recentemente, no aprendizado semissupervisionado. No aprendizado supervisionado, por outro lado, o uso de algoritmos baseados em grafos ainda tem sido pouco explorado na literatura. Este trabalho apresenta um algoritmo não paramétrico baseado em grafos para problemas de classificação com distribuição estacionária, bem como sua extensão para problemas que apresentam distribuição não estacionária. O algoritmo desenvolvido baseia-se em dois conceitos, a saber, 1) em uma estrutura chamada grafo K-associado ótimo, que representa o conjunto de treinamento como um grafo esparso e dividido em componentes; e 2) na medida de pureza de cada componente, que utiliza a estrutura do grafo para determinar o nível de mistura local dos dados em relação às suas classes. O trabalho também considera problemas de classificação que apresentam alteração na distribuição de novos dados. Este problema caracteriza a mudança de conceito e degrada o desempenho do classificador. De modo que, para manter bom desempenho, é necessário que o classificador continue aprendendo durante a fase de aplicação, por exemplo, por meio de aprendizado incremental. Resultados experimentais sugerem que ambas as abordagens apresentam vantagens na classificação de dados em relação aos algoritmos testados / Graph-based methods consist in a powerful form for data representation and abstraction which provides, among others advantages, representing topological relations, visualizing structures, representing groups of data with distinct formats, as well as, supplying alternative measures to characterize data. Such approach has been each time more considered to solve machine learning related problems, mainly concerning unsupervised learning, like clustering, and recently, semi-supervised learning. However, graph-based solutions for supervised learning tasks still remain underexplored in literature. This work presents a non-parametric graph-based algorithm suitable for classification problems with stationary distribution, as well as its extension to cope with problems of non-stationary distributed data. The developed algorithm relies on the following concepts, 1) a graph structure called optimal K-associated graph, which represents the training set as a sparse graph separated into components; and 2) the purity measure for each component, which uses the graph structure to determine local data mixture level in relation to their classes. This work also considers classification problems that exhibit modification on distribution of data flow. This problem qualifies concept drift and worsens any static classifier performance. Hence, in order to maintain accuracy performance, it is necessary for the classifier to keep learning during application phase, for example, by implementing incremental learning. Experimental results, concerning both algorithms, suggest that they had presented advantages over the tested algorithms on data classification tasks

The goal of a Brain-Computer Interface (BCI) consists of the development of a unidirectional interface between a human and a computer to allow control of a device only via brain signals. While the BCI systems of almost all other groups require the user to be trained over several weeks or even months, the group of Prof. Dr. Klaus-Robert Müller in Berlin and Potsdam, which I belong to, was one of the first research groups in this field which used machine learning techniques on a large scale. The adaptivity of the processing system to the individual brain patterns of the subject confers huge advantages for the user. Thus BCI research is considered a hot topic in machine learning and computer science. It requires interdisciplinary cooperation between disparate fields such as neuroscience, since only by combining machine learning and signal processing techniques based on neurophysiological knowledge will the largest progress be made.<br><br> In this work I particularly deal with my part of this project, which lies mainly in the area of computer science. I have considered the following three main points:<br><br> <b>Establishing a performance measure based on information theory:</b> I have critically illuminated the assumptions of Shannon's information transfer rate for application in a BCI context. By establishing suitable coding strategies I was able to show that this theoretical measure approximates quite well to what is practically achieveable.<br> <b>Transfer and development of suitable signal processing and machine learning techniques:</b> One substantial component of my work was to develop several machine learning and signal processing algorithms to improve the efficiency of a BCI. Based on the neurophysiological knowledge that several independent EEG features can be observed for some mental states, I have developed a method for combining different and maybe independent features which improved performance. In some cases the performance of the combination algorithm outperforms the best single performance by more than 50 %. Furthermore, I have theoretically and practically addressed via the development of suitable algorithms the question of the optimal number of classes which should be used for a BCI. It transpired that with BCI performances reported so far, three or four different mental states are optimal. For another extension I have combined ideas from signal processing with those of machine learning since a high gain can be achieved if the temporal filtering, i.e., the choice of frequency bands, is automatically adapted to each subject individually.<br> <b>Implementation of the Berlin brain computer interface and realization of suitable experiments:</b> Finally a further substantial component of my work was to realize an online BCI system which includes the developed methods, but is also flexible enough to allow the simple realization of new algorithms and ideas. So far, bitrates of up to 40 bits per minute have been achieved with this system by absolutely untrained users which, compared to results of other groups, is highly successful. / Ein Brain-Computer Interface (BCI) ist eine unidirektionale Schnittstelle zwischen Mensch und Computer, bei der ein Mensch in der Lage ist, ein Gerät einzig und allein Kraft seiner Gehirnsignale zu steuern. In den BCI Systemen fast aller Forschergruppen wird der Mensch in Experimenten über Wochen oder sogar Monaten trainiert, geeignete Signale zu produzieren, die vordefinierten allgemeinen Gehirnmustern entsprechen. Die BCI Gruppe in Berlin und Potsdam, der ich angehöre, war in diesem Feld eine der ersten, die erkannt hat, dass eine Anpassung des Verarbeitungssystems an den Menschen mit Hilfe der Techniken des Maschinellen Lernens große Vorteile mit sich bringt. In unserer Gruppe und mittlerweile auch in vielen anderen Gruppen wird BCI somit als aktuelles Forschungsthema im Maschinellen Lernen und folglich in der Informatik mit interdisziplinärer Natur in Neurowissenschaften und anderen Feldern verstanden, da durch die geeignete Kombination von Techniken des Maschinellen Lernens und der Signalverarbeitung basierend auf neurophysiologischem Wissen der größte Erfolg erzielt werden konnte.<br><br> In dieser Arbeit gehe ich auf meinem Anteil an diesem Projekt ein, der vor allem im Informatikbereich der BCI Forschung liegt. Im Detail beschäftige ich mich mit den folgenden drei Punkten:<br><br> <b>Diskussion eines informationstheoretischen Maßes für die Güte eines BCI's:</b> Ich habe kritisch die Annahmen von Shannon's Informationsübertragungsrate für die Anwendung im BCI Kontext beleuchtet. Durch Ermittlung von geeigneten Kodierungsstrategien konnte ich zeigen, dass dieses theoretische Maß den praktisch erreichbaren Wert ziemlich gut annähert.<br> <b>Transfer und Entwicklung von geeigneten Techniken aus dem Bereich der Signalverarbeitung und des Maschinellen Lernens:</b> Eine substantielle Komponente meiner Arbeit war die Entwicklung von Techniken des Machinellen Lernens und der Signalverarbeitung, um die Effizienz eines BCI's zu erhöhen. Basierend auf dem neurophysiologischem Wissen, dass verschiedene unabhängige Merkmale in Gehirnsignalen für verschiedene mentale Zustände beobachtbar sind, habe ich eine Methode zur Kombination von verschiedenen und unter Umständen unabhängigen Merkmalen entwickelt, die sehr erfolgreich die Fähigkeiten eines BCI's verbessert. Besonders in einigen Fällen übertraf die Leistung des entwickelten Kombinationsalgorithmus die beste Leistung auf den einzelnen Merkmalen mit mehr als 50 %. Weiterhin habe ich theoretisch und praktisch durch Einführung geeigneter Algorithmen die Frage untersucht, wie viele Klassen man für ein BCI nutzen kann und sollte. Auch hier wurde ein relevantes Resultat erzielt, nämlich dass für BCI Güten, die bis heute berichtet sind, die Benutzung von 3 oder 4 verschiedenen mentalen Zuständen in der Regel optimal im Sinne von erreichbarer Leistung sind. Für eine andere Erweiterung wurden Ideen aus der Signalverarbeitung mit denen des Maschinellen Lernens kombiniert, da ein hoher Erfolg erzielt werden kann, wenn der temporale Filter, d.h. die Wahl des benutzten Frequenzbandes, automatisch und individuell für jeden Menschen angepasst wird.<br> <b>Implementation des Berlin Brain-Computer Interfaces und Realisierung von geeigneten Experimenten:</b> Eine weitere wichtige Komponente meiner Arbeit war eine Realisierung eines online BCI Systems, welches die entwickelten Methoden umfasst, aber auch so flexibel ist, dass neue Algorithmen und Ideen einfach zu verwirklichen sind. Bis jetzt wurden mit diesem System Bitraten von bis zu 40 Bits pro Minute von absolut untrainierten Personen in ihren ersten BCI Experimenten erzielt. Dieses Resultat übertrifft die bisher berichteten Ergebnisse aller anderer BCI Gruppen deutlich. <br> <hr> Bemerkung:<br> Der Autor wurde mit dem <i>Michelson-Preis</i> 2005/2006 für die beste Promotion des Jahrgangs der Mathematisch-Naturwissenschaftlichen Fakultät der Universität Potsdam ausgezeichnet.

In recent years, deep neural network models have shown to outperform many state of the art algorithms. The reason for this is, unsupervised pretraining with multi-layered deep neural networks have shown to learn better features, which further improves many supervised tasks. These models not only automate the feature extraction process but also provide with robust features for various machine learning tasks. But the unsupervised pretraining and feature extraction using multi-layered networks are restricted only to the input features and not to the output. The performance of many supervised learning algorithms (or models) depends on how well the output dependencies are handled by these algorithms [Dembczy´nski et al., 2012]. Adapting the standard neural networks to handle these output dependencies for any specific type of problem has been an active area of research [Zhang and Zhou, 2006, Ribeiro et al., 2012]. On the other hand, inference into multimodal data is considered as a difficult problem in machine learning and recently ‘deep multimodal neural networks’ have shown significant results [Ngiam et al., 2011, Srivastava and Salakhutdinov, 2012]. Several problems like classification with complete or missing modality data, generating the missing modality etc., are shown to perform very well with these models. In this work, we consider three nontrivial supervised learning tasks (i) multi-class classification (MCC), (ii) multi-label classification (MLC) and (iii) label ranking (LR), mentioned in the order of increasing complexity of the output. While multi-class classification deals with predicting one class for every instance, multi-label classification deals with predicting more than one classes for every instance and label ranking deals with assigning a rank to each label for every instance. All the work in this field is associated around formulating new error functions that can force network to identify the output dependencies. Aim of our work is to adapt neural network to implicitly handle the feature extraction (dependencies) for output in the network structure, removing the need of hand crafted error functions. We show that the multimodal deep architectures can be adapted for these type of problems (or data) by considering labels as one of the modalities. This also brings unsupervised pretraining to the output along with the input. We show that these models can not only outperform standard deep neural networks, but also outperform standard adaptations of neural networks for individual domains under various metrics over several data sets considered by us. We can observe that the performance of our models over other models improves even more as the complexity of the output/ problem increases.

Dans cet article, nous présentons une nouvelle approche de détection multi-classes basée sur un parcours hiérarchique de classifieurs appris simultanément. Pour plus de robustesse et de rapidité, nous proposons d’utiliser un arbre de classes d’objets. Notre modèle de détection est appris en combinant les contraintes de tri et de classification dans un seul problème d’optimisation. Notre formulation convexe permet d’utiliser un algorithme de recherche pour accélérer le temps d’exécution. Nous avons mené des évaluations de notre algorithme sur les benchmarks PASCAL VOC (2007 et 2010). Comparé à l’approche un-contre-tous, notre méthode améliore les performances pour 20 classes et gagne 10x en vitesse. / Recent years have witnessed a competition in autonomous navigation for vehicles boosted by the advances in computer vision. The on-board cameras are capable of understanding the semantic content of the environment. A core component of this system is to localize and classify objects in urban scenes. There is a need to have multi-class object detection systems. Designing such an efficient system is a challenging and active research area. The algorithms can be found for applications in autonomous driving, object searches in images or video surveillance. The scale of object classes varies depending on the tasks. The datasets for object detection started with containing one class only e.g. the popular INRIA Person dataset. Nowadays, we witness an expansion of the datasets consisting of more training data or number of object classes. This thesis proposes a solution to efficiently learn a multi-class object detector. The task of such a system is to localize all instances of target object classes in an input image. We distinguish between three major efficiency criteria. First, the detection performance measures the accuracy of detection. Second, we strive low execution times during run-time. Third, we address the scalability of our novel detection framework. The two previous criteria should scale suitably with the number of input classes and the training algorithm has to take a reasonable amount of time when learning with these larger datasets. Although single-class object detection has seen a considerable improvement over the years, it still remains a challenge to create algorithms that work well with any number of classes. Most works on this subject extent these single-class detectors to work accordingly with multiple classes but remain hardly flexible to new object descriptors. Moreover, they do not consider all these three criteria at the same time. Others use a more traditional approach by iteratively executing a single-class detector for each target class which scales linearly in training time and run-time. To tackle the challenges, we present a novel framework where for an input patch during detection the closest class is ranked highest. Background labels are rejected as negative samples. The detection goal is to find the highest scoring class. To this end, we derive a convex problem formulation that combines ranking and classification constraints. The accuracy of the system is improved by hierarchically arranging the classes into a tree of classifiers. The leaf nodes represent the individual classes and the intermediate nodes called super-classes group recursively these classes together. The super-classes benefit from the shared knowledge of their descending classes. All these classifiers are learned in a joint optimization problem along with the previouslymentioned constraints. The increased number of classifiers are prohibitive to rapid execution times. The formulation of the detection goal naturally allows to use an adapted tree traversal algorithm to progressively search for the best class but reject early in the detection process the background samples and consequently reduce the system’s run-time. Our system balances between detection performance and speed-up. We further experimented with feature reduction to decrease the overhead of applying the high-level classifiers in the tree. The framework is transparent to the used object descriptor where we implemented the histogram of orientated gradients and deformable part model both introduced in [Felzenszwalb et al., 2010a]. The capabilities of our system are demonstrated on two challenging datasets containing different object categories not necessarily semantically related. We evaluate both the detection performance with different number of classes and the scalability with respect to run-time. Our experiments show that this framework fulfills the requirements of a multi-class object detector and highlights the advantages of structuring class-level knowledge.

Orientador: Siome Klein Goldenstein / Tese (doutorado) - Universidade Estadual de Campinas, Instituto da Computação / Made available in DSpace on 2018-08-12T17:37:15Z (GMT). No. of bitstreams: 1 Rocha_AndersondeRezende_D.pdf: 10303487 bytes, checksum: 243dccfe5255c828ce7ead27c27eb1cd (MD5) Previous issue date: 2009 / Resumo: Neste trabalho de doutorado, propomos a utilizaçãoo de classificadores e técnicas de aprendizado de maquina para extrair informações relevantes de um conjunto de dados (e.g., imagens) para solução de alguns problemas em Processamento de Imagens e Visão Computacional. Os problemas de nosso interesse são: categorização de imagens em duas ou mais classes, detecçãao de mensagens escondidas, distinção entre imagens digitalmente adulteradas e imagens naturais, autenticação, multi-classificação, entre outros. Inicialmente, apresentamos uma revisão comparativa e crítica do estado da arte em análise forense de imagens e detecção de mensagens escondidas em imagens. Nosso objetivo é mostrar as potencialidades das técnicas existentes e, mais importante, apontar suas limitações. Com esse estudo, mostramos que boa parte dos problemas nessa área apontam para dois pontos em comum: a seleção de características e as técnicas de aprendizado a serem utilizadas. Nesse estudo, também discutimos questões legais associadas a análise forense de imagens como, por exemplo, o uso de fotografias digitais por criminosos. Em seguida, introduzimos uma técnica para análise forense de imagens testada no contexto de detecção de mensagens escondidas e de classificação geral de imagens em categorias como indoors, outdoors, geradas em computador e obras de arte. Ao estudarmos esse problema de multi-classificação, surgem algumas questões: como resolver um problema multi-classe de modo a poder combinar, por exemplo, caracteríisticas de classificação de imagens baseadas em cor, textura, forma e silhueta, sem nos preocuparmos demasiadamente em como normalizar o vetor-comum de caracteristicas gerado? Como utilizar diversos classificadores diferentes, cada um, especializado e melhor configurado para um conjunto de caracteristicas ou classes em confusão? Nesse sentido, apresentamos, uma tecnica para fusão de classificadores e caracteristicas no cenário multi-classe através da combinação de classificadores binários. Nós validamos nossa abordagem numa aplicação real para classificação automática de frutas e legumes. Finalmente, nos deparamos com mais um problema interessante: como tornar a utilização de poderosos classificadores binarios no contexto multi-classe mais eficiente e eficaz? Assim, introduzimos uma tecnica para combinação de classificadores binarios (chamados classificadores base) para a resolução de problemas no contexto geral de multi-classificação. / Abstract: In this work, we propose the use of classifiers and machine learning techniques to extract useful information from data sets (e.g., images) to solve important problems in Image Processing and Computer Vision. We are particularly interested in: two and multi-class image categorization, hidden messages detection, discrimination among natural and forged images, authentication, and multiclassification. To start with, we present a comparative survey of the state-of-the-art in digital image forensics as well as hidden messages detection. Our objective is to show the importance of the existing solutions and discuss their limitations. In this study, we show that most of these techniques strive to solve two common problems in Machine Learning: the feature selection and the classification techniques to be used. Furthermore, we discuss the legal and ethical aspects of image forensics analysis, such as, the use of digital images by criminals. We introduce a technique for image forensics analysis in the context of hidden messages detection and image classification in categories such as indoors, outdoors, computer generated, and art works. From this multi-class classification, we found some important questions: how to solve a multi-class problem in order to combine, for instance, several different features such as color, texture, shape, and silhouette without worrying about the pre-processing and normalization of the combined feature vector? How to take advantage of different classifiers, each one custom tailored to a specific set of classes in confusion? To cope with most of these problems, we present a feature and classifier fusion technique based on combinations of binary classifiers. We validate our solution with a real application for automatic produce classification. Finally, we address another interesting problem: how to combine powerful binary classifiers in the multi-class scenario more effectively? How to boost their efficiency? In this context, we present a solution that boosts the efficiency and effectiveness of multi-class from binary techniques. / Doutorado / Engenharia de Computação / Doutor em Ciência da Computação

Submitted by Renata Lopes (renatasil82@gmail.com) on 2017-03-07T15:27:21Z No. of bitstreams: 1 mauricioarchanjonunescoelho.pdf: 2468130 bytes, checksum: 3f05daa8428e367942c4ad560b6375f2 (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2017-03-10T12:19:13Z (GMT) No. of bitstreams: 1 mauricioarchanjonunescoelho.pdf: 2468130 bytes, checksum: 3f05daa8428e367942c4ad560b6375f2 (MD5) / Made available in DSpace on 2017-03-10T12:19:13Z (GMT). No. of bitstreams: 1 mauricioarchanjonunescoelho.pdf: 2468130 bytes, checksum: 3f05daa8428e367942c4ad560b6375f2 (MD5) Previous issue date: 2010-06-18 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / O problema de planejamento de caminhos apresenta diversas subáreas, muitas das quais já extensamente abordadas na literatura. Uma dessas áreas em especial é a de determinação de caminhos, os algoritmos empregados para a solução deste problema dependem que os custos estipulados para os ambientes ou mapas sejam confiáveis. A dificuldade está justamente na definição dos custos referentes a cada tipo de área ou terreno nos mapas a serem examinados. Como se pode observar, o problema mencionado inclui a dificuldade em se determinar qual o custo de cada característica relevante presente no mapa, bem como os custos de suas possíveis combinações. A proposta deste trabalho é mostrar como é feita a predição desses custos em novos ambientes tendo como base a predição de dados estruturados definindo um aprendizado funcional entre domínios de entrada e saída, estruturados e arbitrários. O problema de aprendizado em questão é normalmente formulado como um problema de otimização convexa de máxima margem bastante similar a formulação de máquinas de vetores suporte multi-classe. Como técnica de solução realizou-se a implementação do algoritmo MMP (Maximum Margin Planning) (RATLIFF; BAGNELL; ZINKEVICH, 2006). Como contribuição, desenvolveu-se e implementou-se dois algoritmos alternativos, o primeiro denominado Perceptron Estruturado e o segundo Perceptron Estruturado com Margem, ambos os métodos de relaxação baseados na formulação do Perceptron. Os mesmos foram analisados e comparados. Posteriormente temos a exploração dos ambientes por um agente inteligente utilizando técnicas de aprendizado por reforço. Tornando todo o processo, desde a análise do ambiente e descoberta de custos, até sua exploração e planejamento do caminho, um completo processo de aprendizado. / The problem of path planning has several sub-areas, many of which are widely discussed in the literature. One of these areas in particular is the determination of paths, the algorithms used to solve this problem depend on the reliability of the estimated costs in the environments and maps. The difficulty is precisely the definition of costs for each type of area or land on the maps to be examined. As you can see, the problem mentioned includes the difficulty in determining what the cost of each relevant characteristic on the map, and the costs of their possible combinations. The purpose of this study is to show how the prediction of these costs is made into new environments based on the prediction of structured data by defining functional learning areas between input and output, structured and arbitrary. The problem of learning in question is usually formulated as a convex optimization problem of maximum margin very similar to the formulation of multiclass support vector machines. A solution technic was performed through implementation of the algorithm MMP (Maximum Margin Planning) (RATLIFF; BAGNELL; ZINKEVICH, 2006). As a contribution, two alternative algorithms were developed and implemented, the first named Structured Perceptron, and the second Structured Perceptron with Margin both methods of relaxation based formulation of the Perceptron. They were analyzed and compared. Posteriorly we have the exploitation of the environment by an intelligent agent using reinforcement learning techniques. This makes the whole process, from the environment analysis and discovery of cost to the exploitation and path planning, a complete learning process.

Métodos baseados em grafos consistem em uma poderosa forma de representação e abstração de dados que proporcionam, dentre outras vantagens, representar relações topológicas, visualizar estruturas, representar grupos de dados com formatos distintos, bem como, fornecer medidas alternativas para caracterizar os dados. Esse tipo de abordagem tem sido cada vez mais considerada para solucionar problemas de aprendizado de máquina, principalmente no aprendizado não supervisionado, como agrupamento de dados, e mais recentemente, no aprendizado semissupervisionado. No aprendizado supervisionado, por outro lado, o uso de algoritmos baseados em grafos ainda tem sido pouco explorado na literatura. Este trabalho apresenta um algoritmo não paramétrico baseado em grafos para problemas de classificação com distribuição estacionária, bem como sua extensão para problemas que apresentam distribuição não estacionária. O algoritmo desenvolvido baseia-se em dois conceitos, a saber, 1) em uma estrutura chamada grafo K-associado ótimo, que representa o conjunto de treinamento como um grafo esparso e dividido em componentes; e 2) na medida de pureza de cada componente, que utiliza a estrutura do grafo para determinar o nível de mistura local dos dados em relação às suas classes. O trabalho também considera problemas de classificação que apresentam alteração na distribuição de novos dados. Este problema caracteriza a mudança de conceito e degrada o desempenho do classificador. De modo que, para manter bom desempenho, é necessário que o classificador continue aprendendo durante a fase de aplicação, por exemplo, por meio de aprendizado incremental. Resultados experimentais sugerem que ambas as abordagens apresentam vantagens na classificação de dados em relação aos algoritmos testados / Graph-based methods consist in a powerful form for data representation and abstraction which provides, among others advantages, representing topological relations, visualizing structures, representing groups of data with distinct formats, as well as, supplying alternative measures to characterize data. Such approach has been each time more considered to solve machine learning related problems, mainly concerning unsupervised learning, like clustering, and recently, semi-supervised learning. However, graph-based solutions for supervised learning tasks still remain underexplored in literature. This work presents a non-parametric graph-based algorithm suitable for classification problems with stationary distribution, as well as its extension to cope with problems of non-stationary distributed data. The developed algorithm relies on the following concepts, 1) a graph structure called optimal K-associated graph, which represents the training set as a sparse graph separated into components; and 2) the purity measure for each component, which uses the graph structure to determine local data mixture level in relation to their classes. This work also considers classification problems that exhibit modification on distribution of data flow. This problem qualifies concept drift and worsens any static classifier performance. Hence, in order to maintain accuracy performance, it is necessary for the classifier to keep learning during application phase, for example, by implementing incremental learning. Experimental results, concerning both algorithms, suggest that they had presented advantages over the tested algorithms on data classification tasks

Master's thesis deals with multiclass image segmentation using convolutional neural networks. The theoretical part of the Master's thesis focuses on image segmentation. There are basics principles of neural networks and image segmentation with more types of approaches. In practical part the Unet architecture is choosen and is described for image segmentation more. U-net was applied for medicine dataset. There is processing procedure which is more described for image proccesing of three-dimmensional data. There are also methods for data preproccessing which were applied for image multiclass segmentation. Final part of current master's thesis evaluates results.

Brain tumors represent a diverse spectrum of cancer types which can induce grave complications and lead to poor life expectancy. Amongst the various brain tumor types, gliomas are primary brain tumors that compose about 30% of adult brain tumors. They are graded according to the World Health Organization into Grades 1 to 4 (G1-G4), where G4 is the highest grade with the highest malignancy and poor prognosis. Early diagnosis and classification of brain tumor grade is very important since it can improve the treatment procedure and (potentially) prolong a patient's life, since life expectancy largely depends on the level of malignancy and the tumor's histological characteristics. While clinicians have diagnostic tools they use as a gold standard, such as biopsies these are either invasive or costly. A widely used example of a non-invasive technique is magnetic resonance imaging, due to its ability to produce images with different soft-tissue contrast and high spatial resolution thanks to multiple imaging sequences. However, the examination of such images can be overwhelming for radiologists due to the overall large amount of data. Deep learning approaches, on the other hand, have shown great potential in brain tumor diagnosis and can assist radiologists in the decision-making process. In this thesis, brain tumor grade classification in MR images is performed using deep learning. Two popular pre-trained CNN models (VGG-19, ResNet50) were employed using single MR modalities and combinations of them to classify gliomas into three grades. All models were trained using data augmentation on 2D images from the TCGA dataset, which consisted of 3D volumes from 142 anonymized patients. The models were evaluated based on accuracy, precision, recall, F1-score, AUC score, as well as the Wilcoxon Signed-Rank test to establish if one classifier was statistically significantly better than the other. Since deep learning models are typically 'black box' models and can be difficult to interpret by non-experts, Gradient-weighted Class Activation Mapping (Grad-CAM) was used in order to address model explainability. For single modalities, VGG-19 displayed the highest performance with a test accuracy of 77.86%, whilst for combinations of two and three modalities T1ce, FLAIR and T2, T1ce, FLAIR were the best performing ones for VGG-19 with a test accuracy of 74.48%, 75.78%, respectively. Statistical comparisons indicated that for single MR modalities and combinations of two MR modalities, there was not a statistically significant difference between the two classifiers, whilst for combination of three modalities, one model was better than the other. However, given the small size of the test population, these comparisons have low statistical power. The use of Grad-CAM for model explainability indicated that ResNet50 was able to localize the tumor region better than VGG-19.

In this extended abstract, we propose an Artificial Intelligence-based model dedicated to the representation of a multi-class traffic flow, i.e. a traffic flow in which different vehicle classes (at least cars and trucks) are explicitly represented, with the aim of using it for the development of freeway traffic control schemes based on ramp management. Specifically, the goal of this work is to develop a hybrid modelling technique in which a Machine Learning component and the multi-class version of METANET model are adopted to determine a better estimation and forecasting tool for freeway traffic. The resulting model is specifically devised in order to be included in a Model Predictive Control (MPC) scheme for the required traffic state prediction.