Early detection is crucial for effectively treating glaucoma, a leading cause of irreversible blindness. Diagnosing glaucoma can be challenging due to its subtle early symptoms. This study aims to enhance glaucoma prediction by fine-tuning pre-trained convolutional neural networks. Several networks were re-trained and tested on publicly available retinal image datasets. Additionally, the models were evaluated on fundus images from patients at Region Västernorrland (RVN). The methodology involved exploring how to effectively process and prepare patient data for prediction purposes. The results showed that a majority voting ensemble of the fine-tuned models produced the highest performance, achieving an accuracy of approximately 0.94, with a specificity and sensitivity of 0.97 and 0.90 respectively. The ensemble also identified 0.90 glaucomatous images from RVN correctly. In terms of specificity and sensitivity, all models outperformed the results of ophthalmologist specialists described in a previous study. The findings suggest the effectiveness of transfer learning in enhancing the diagnostic accuracy of glaucoma. It also underscores the importance of proper storage and preparation of medical data for developing predicitive machine learning models. / Glaukom, mer känt som grön starr, är en av de vanligast förekommande ögonsjukdomarna som orsakar blindhet. Det är viktigt att diagnostisera glaukom tidigt i sjukdomsförloppet för att genom behandling, sakta ner eller stoppa ytterligare synförlust. Att diagnostisera glaukom kan vara utmanande, eftersom det vanligtvis inte visar några tidiga symtom. Artificiell intelligens (AI), eller mer specifikt maskininlärning (ML), kan hjälpa läkare att ställa rätt diagnos om det används som ett beslutsstöd. Faltande neurala nätverk (convolutional neural network, CNN) kan lära sig att känna igen mönster i bilder, för att därigenom klassificera bilder till olika kategorier. Ett sätt att diagnostisera glaukom är att studera näthinnan och synnerven i ögats bakre del, som kallas ögonbotten. I denna studie finjusterades redan tränade CNN:s för att prediktera glaukom utifrån ögonbottenbilder. Detta uppnåddes genom att träna om modellerna på publikt tillgängliga ögonbottenbilder. Målet var att jämföra nätverkens noggrannhet på en delmängd av bilderna, samt att evaluera dem på ögonbottenbilder från sjukhus i Region Västernorrland (RVN). För att uppnå detta ingick det även i metodiken att utforska begränsningarna och möjligheterna med hur patientdata får användas, samt att undersöka hur datat bör lagras och tillrättaläggas för att möjliggöra utvecklingen av prediktionsmodeller. Syftet med studien var att öka noggrannheten vid diagnostisering av glaukom. Resultaten visade att en ensemble baserad på majoritetsröstning av alla modeller gav den bästa noggrannheten, ungefär 0.94. Sensitiviteten och specificiteten var 0.90, respektive 0.97. Vidare klassificerades 90% av ögonbottenbilderna från RVN korrekt. Resultaten tyder på att maskininlärning är effektivt för att förbättra den diagnostiska noggrannheten för glaukom. Det understryker också vikten av strategisk lagring och förberedelse av medicinska data för att utveckla prediktiva maskininlärningsmodeller i framtiden.

High-resolution digital elevation models (HRDEMs) enable precise mapping of hydrographic features. However, the absence of drainage crossings underpassing roads or bridges hinders accurate delineation of stream networks. Traditional methods such as on-screen digitization and field surveys for locating these crossings are time-consuming and expensive for extensive areas. This study investigates the effectiveness of deep learning models for automated drainage crossing detection using HRDEMs. The study also explores the performance of advanced classification algorithm such as EfficientNetV2 model using various co-registered HRDRM-derived geomorphological features, such as positive openness, geometric curvature, and topographic position index (TPI) variants, for drainage crossings classification. The results reveal that individual layers, particularly HRDEM and TPI21, achieve the best performance, while combining all five layers doesn't improve accuracy. Hence, effective feature screening is crucial, as eliminating less informative features enhances the F1 score. For drainage crossing detection, this study develops and trains deep learning models, Faster R-CNN and YOLOv5 object detectors, using HRDEM tiles and ground truth labels. These models achieve an average F1-score of 0.78 in Nebraska watershed and demonstrate successful transferability to other watersheds. This spatial object detection approach offers a promising avenue for automated, large-scale drainage crossing detection, facilitating the integration of these features into HRDEMs and improving the accuracy of hydrographic network delineation.

Intelligent transportation systems (ITS) are essential tools for traffic planning, analysis, and forecasting that can utilize the huge amount of traffic data available nowadays. In this work, we aggregated detailed traffic flow sensor data, Waze reports, OpenStreetMap (OSM) features, and weather data, from California Bay Area for 6 months. Using that data, we studied three novel ITS applications using convolutional neural networks (CNNs) and recurrent neural networks (RNNs). The first experiment is an analysis of the relation between roadway shapes and accident occurrence, where results show that the speed limit and number of lanes are significant predictors for major accidents on highways. The second experiment presents a novel method for forecasting congestion severity using crowdsourced data only (Waze, OSM, and weather), without the need for traffic sensor data. The third experiment studies the improvement of traffic flow forecasting using accidents, number of lanes, weather, and time-related features, where results show significant performance improvements when the additional features where used.

The master thesis examines the war reporting of two Czech dailies (Mladá fronta DNES and Právo), specifically their coverage of the Czech Army mission and deployment in Afghanistan. In the theoretical part the quality of reporting, topic and factors that influence the journalist work are considered. A major theme is the relationship of media and political elites and some of its manifestations - the CNN effect connected with the media and public push on the departure of US military from Vietnam, and peace- journalism a concept of reporting on wars by exploiting more themes like reconstruction and peaceful solutions rather than war and combat. The quantitative analysis aims to verify these phenomenon in the Czech media. This is done mainly by analysing the reports sources and looking for the presence of three frames derived from the theoretical literature on this topic: heroic framing applied on Czech soldiers, national framing reflecting the national interests, and humanitarian and development aid framing.

The purpose of this Master's thesis is to analyse the image of Mexico in the media of the United States in the first half of 2010. First, the author gives an overview of selected media theories and describes the specifics of the U.S. media market. On the sample of the three media (El Paso Times, New York Times and Fox News) she analyses how were the U.S. media referring about Mexico and its citizens. She is validating hypotheses, which she based on characteristics of each of analysed media. She comes to a conclusion that reporting about Mexico varies among the selected media, which reflects specifics of each of them. In conclusion, she is applying some of the media theories presented in the first chapter. She finds that CNN effect and framing occur. Additionally, she argues that Baudrillard's simulacrum also appears.

Les réseaux de neurones artificiels sont nés, il y a presque cinquante ans, de la volonté de modéliser les capacités de mémorisation et de traitement du cerveau biologique. Aujourd'hui encore, les nombreux modèles obtenus brillent par leur simplicité de mise en œuvre, leur puissance de traitement, leur polyvalence, mais aussi par la complexité des méthodes de programmation disponibles. En réalité, très peu d'entre-elles sont capables d'aboutir analytiquement à un réseau de neurones correctement configuré. Bien au contraire, la plupart se " contentent " d'ajuster, petit à petit, une ébauche de réseau de neurones, jusqu'à ce qu'il fonctionne avec suffisamment d'exemples de la tâche à accomplir. Au travers de ces méthodes, dites " d'apprentissages ", les réseaux de neurones sont devenus des boîtes noires, que seuls quelques experts sont effectivement capables de programmer. Chaque traitement demande en effet de choisir convenablement une configuration initiale, la nature des exemples, leur nombre, l'ordre d'utilisation, ... Pourtant, la tâche finalement apprise n'en reste pas moins le résultat d'une stratégie algorithmique implémentée par le réseau de neurones. Une stratégie qui peut donc être identifiée par le biais de l'analyse, et surtout réutilisée lors de la conception d'un réseau de neurones réalisant une tâche similaire, court-circuitant ainsi les nombreux aléas liés à ces méthodes d'apprentissage. Les bénéfices de l'analyse sont encore plus évidents dans le cas de réseaux de neurones à sortie binaire. En effet, le caractère discret des signaux traités simplifie grandement l'identification des mécanismes mis en jeu, ainsi que leur contribution au traitement global. De ce type d'analyse systématique naît un formalisme original, qui décrit la stratégie implémentée par les réseaux de neurones à sortie binaire de façon particulièrement efficace. Schématiquement, ce formalisme tient lieu d'" état intermédiaire " entre la forme boîte noire d'un réseau de neurones et sa description mathématique brute. En étant plus proche des modèles de réseaux de neurones que ne l'est cette dernière, il permet de retrouver, par synthèse analytique, un réseau de neurones effectuant la même opération que celui de départ, mais de façon optimisée selon un ou plusieurs critères : nombre de neurones, nombre de connexions, dynamique de calcul, etc. Cette approche analyse-formalisation-synthèse constitue la contribution de ces travaux de thèse.

This thesis focuses on novel video retrieval scenarios. More particularly, we aim at the Known-item Search scenario wherein users search for a short video segment known either visually or by a textual description. The scenario assumes that there is no ideal query example available. Our former known- item search tool relying on color feature signatures is extended with major enhancements. Namely, we introduce a multi-modal sketching tool, the exploration of video content with semantic descriptors derived from deep convolutional networks, new browsing/visualization methods and two orthogonal approaches for textual search. The proposed approaches are embodied in our video retrieval tool Enhanced Sketch-based Video Browser (ESBVB). To evaluate ESBVB performance, we participated in international competitions comparing our tool with the state-of-the-art approaches. Repeatedly, our tool outperformed the other methods. Furthermore, we show in our user study that even novice users are able to effectively employ ESBVB capabilities to search and browse known video clips. Powered by TCPDF (www.tcpdf.org)

The goal of this master thesis was to propose a suitable strategy to detect and classify objects of interest in mammogram images. A part of this goal was to implement an experimentation framework, that will be used for data preparation, model training and comparison. Patch and full-image versions of the dataset were used in the analysis. Initialisation with weights that were pretrained on the images from other domain improved classifier performance. ResNet-34 had better AUC scores on the test set that ResNet-18. Semi-supervised training using entropy minimisation has no significant improvement over the supervised one. The thesis includes the visualisation of the network predictions and the analysis of the knowledge representation of the classier. The achieved results for a patch version of the dataset are comparable to the results of another article that utilised the same test set. For a full-image dataset the results of the training were suboptimal. 1

System log files are filled with logged events, status codes, and other messages. By analyzing the log files, the systems current state can be determined, and find out if something during its execution went wrong. Log file analysis has been studied for some time now, where recent studies have shown state-of-the-art performance using machine learning techniques. In this thesis, document classification solutions were tested on log files in order to classify regular system runs versus abnormal system runs. To solve this task, supervised and unsupervised learning methods were combined. Doc2Vec was used to extract document features, and Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) based architectures on the classification task. With the use of the machine learning models and preprocessing techniques the tested models yielded an f1-score and accuracy above 95% when classifying log files.

This study evaluates noise robustness of convolutional autoencoders and neural networks for classification of Low Probability of Intercept (LPI) radar modulation type. Specifically, a number of different neural network architectures are tested in four different synthetic noise environments. Tests in Gaussian noise show that performance is decreasing with decreasing Signal to Noise Ratio (SNR). Training a network on all SNRs in the dataset achieved a peak performance of 70.8 % at SNR=-6 dB with a denoising autoencoder and convolutional classifier setup. Tests indicate that the models have a difficult time generalizing to SNRs lower than what is provided in training data, performing roughly 10-20% worse than when those SNRs are included in the training data. If intermediate SNRs are removed from the training data the models can generalize and perform similarly to tests where, intermediate noise levels are included in the training data. When testing data is generated with different parameters to training data performance is underwhelming, with a peak performance of 22.0 % at SNR=-6 dB. The last tests done use telecom signals as additive noise instead of Gaussian noise. These tests are performed when the LPI and telecom signals appear at different frequencies. The models preform well on such cases with a peak performance of 80.3 % at an intermidiate noise level. This study also contribute with a different, and more realistic, way of generating data than what is prevalent in literature as well as a network that performs well without the need for signal preprocessing. Without preprocessing a peak performance of 64.9 % was achieved at SNR=-6 dB. It is customary to generate data such that each sample always includes the start of its signals period which increases performance by around 20 % across all tests. In a real application however it is not certain that the start of a received signal can be determined. / Detta arbete studerar brustålighet hos neurala nätverk för klassificering av \textit{låg sannolikhet för avlyssning} (LPI) radars modulationstyp. Specifikt testas ett antal arkitekturer baserade på faltningsnätverk och evalueras i fyra olika syntetiska brusmiljöer. Tester genomförda på data med Gaussiskt brus visar att klasificeringsfelet är ökande med ett minskande signal-till-brusförhållande. Om man låter nätverken träna på alla brusnivåer som ingår i datan uppnås en högsta pricksäkerhet om 70.8 % vid ett signal-till-brusförhållande på -6 dB. Vidare tester tyder på att nätverken presterar sämre på låga signal-till-brusförhållanden om de inte finns med i träningsdata och ger i allmänhet mellan 10-20 % sämre pricksäkerhet. Om de mellersta brusnivåerna inte finns med i träningsdata presterar nätverken lika bra som när de finns med i träningsdata. Om träningsdata och testdata genereras med olika parameterar presterar nätverken dåligt. För dessa tester uppnås en högsta pricksäkerhet om 22.0 % vid ett signal-till-brusförhållande på -6 dB. Den sista brusmiljön som testades på använder sig av telekom signaler som om de vore brus istället för Gaussiskt brus. I detta fall är LPI och telekom signalerna väl skiljda i frekvens och nätverken presterar lika bra som tester i Gaussiskt brus med högt signal-till-brusförhållande. Högsta pricksäkerhet som uppnåts på dessa tester är 80.3 % i mellanhög brusnivå. Detta arbete bidrar även med nätverk som presterar bra utan att data behöver signalbehandlas innnan den kan klassificeras samt genererar data på ett mer realistiskt vis än tidigare litteratur inom detta område. Utan att signalbehandla data uppnåddes en högsta pricksäkerhet om 64.9 % vid ett signal-till-brusförhållande på -6 dB. Den mer realistiska datan genereras så att dess startpunkt är slumpmässig. I litteraturen brukar startpunkten inkluderas och uppnår på så vis överlag pricksäkerheter som är ungefär 20 % högre än de tester som genomförs i detta arbete. I verkliga applikationer är det sällan man kan identifera en signals startpunkt med säkerhet.