In this paper we explore various approaches to using deep neural networks to per- form cryptanalysis, with the ultimate goal of having a deep neural network deci- pher encrypted data. We use long short-term memory networks to try to decipher encrypted text and we use a convolutional neural network to perform classification tasks on encrypted MNIST images. We find that although the network is unable to decipher encrypted data, it is able to perform classification on encrypted data. We also find that the networks performance is depending on what key were used to en- crypt the data. These findings could be valuable for further research into the topic of cryptanalysis using deep neural networks.

"Pattern Extraction - Scenarios" is a documentation of experiences inside of the collective subconscious that is the internet and an attempt at understanding and highlighting mechanisms and effects of algorithmic governance and cultural automation.

This Master’s thesis deals with beat tracking systems, whose functionality is based on neural networks. It describes the structure of these systems and how the signal is processed in their individual blocks. Emphasis is then placed on recurrent and temporal convolutional networks, which by they nature can effectively detect tempo and beats in audio recordings. The selected methods, network architectures and their modifications are then implemented within a comprehensive detection system, which is further tested and evaluated through a cross-validation process on a genre-diverse data-set. The results show that the system, with proposed temporal convolutional network architecture, produces comparable results with foreign publications. For example, within the SMC dataset, it proved to be the most successful, on the contrary, in the case of other datasets it was slightly below the accuracy of state-of-the-art systems. In addition,the proposed network retains low computational complexity despite increased number of internal parameters.

<p>Doktorska disertacija razmatra problem algoritamske dijagnostike<br />metaboličkog sindroma na osnovu lako merljivih parametara: pol,<br />starosna dob, indeks telesne mase, odnos obima struka i visine,<br />sistolni i dijastolni krvni pritisak. U istraživanju su primenjene i<br />eksperimentalno ispitane tri različite metode mašinskog učenja:<br />stabla odluke, linearna regresija i veštačke neuronske mreže.<br />Pokazano je da veštačke neuronske mreže daju visok nivo<br />prediktivnih vrednosti dovoljan za primenu u praksi. Korišćenjem<br />dobijenog rezultata definisan je i implementiran inteligentni<br />softverski sistem za dijagnostiku metaboličkog sindroma.</p> / <p>The doctoral dissertation examines the problem of algorithmic diagnostics of<br />the metabolic syndrome based on easily measurable parameters: sex, age,<br />body mass index, waist and height ratio, systolic and diastolic blood<br />pressure. In the study, three different methods of machine learning were<br />applied and experimentally examined: decision trees, linear regression and<br />artificial neural networks. It has been shown that artificial neural networks<br />give a high level of predictive value sufficient to be applied in practice. Using<br />the obtained result, an intelligent software system for the diagnosis of<br />metabolic syndrome has been defined and implemented.</p>

Earth Observation is the gathering of information about planet Earth’s system via Remote Sensing technologies for monitoring land cover types and their changes. Through the years, image classification techniques have been widely studied and employed to extract useful information from Earth Observation data such as satellite imagery. One of the most attractive use cases is the monitoring of polar regions, that recently observed some dramatic changes due to global warming. Indeed drifting ice caps and icebergs represent threats to ship activities and navigation in polar areas, and the risk of collision with land-derived ice highlights the need to design a robust and automatic Sea-Ice classification for delivering up-to- date and accurate information. To achieve this goal, satellite data such as Sentinel-1 Synthetic Aperture Radar images from the European Union’s Copernicus program can be given in input to a Deep Learning classifier based on Convolutional Neural Networks capable of giving the content categorization of such images as output. For the task at hand, the availability of labeled data is generally scarce, there- fore the problem of learning with limited labeled data must be faced. There- fore, this work aims at leveraging the broader pool of unlabeled satellite data available to open up new classification solutions. This thesis proposes a Semi-Supervised Learning approach based on Generative Adversarial Networks. Such an architecture takes in input both labeled and unlabeled data and outputs the classification results exploiting the knowledge retrieved from both the data sources. Its classification performance is evaluated and it is later compared with the Supervised Learning approach and the Transfer Learning approach based on pre-trained networks. This work empirically proves that the Semi-Supervised Generative Adversarial Networks approach outperforms the Supervised Learning method, improving its Overall Accuracy by at least 5% in configurations with less than 100 training labeled samples available in the use cases under evaluation, achieving performance comparable to the Transfer Learning approach and even over- coming it under specific experimental configurations. Further analyses are then performed to highlight the effectiveness of the proposed solution. / Jordobservation är samlingen av information om jordklotets system via fjärravkänningstekniker för övervakning av landskapstyper och deras förändringar. Under årens lopp har bildklassificeringstekniker studerats och använts för att extrahera användbar information från jordobservationsdata som satellitbilder. Ett av de mest attraktiva användningsfallen är övervakningen av polära regioner, som nyligen observerade några dramatiska förändringar på grund av den globala uppvärmningen. Driftande istäcken och isberg representerar ett verkligt hot mot fartygsaktiviteter och navigering inom polära områden, och risken för kollision med land-baserad is belyser behovet av att utforma en robust och automatisk Hav-Is-klassificering för att leverera aktuell och korrekt information. För att uppnå detta mål kan satellitdata som Sentinel-1 Synthetic Aperture Radar-bilder från Europeiska unionens Copernicus-program ges som input till en Deep Learning-klassificerare baserad på Convolutional Neural Networks som kan ge innehållskategorisering av sådana bilder som output. För den aktuella uppgiften är tillgängligheten av märkt data i allmänhet otillräcklig, därför måste problemet med inlärning med begränsad mängd märkt data ställas inför rätta. Därav syftar detta arbete till att utnyttja den bredare samlingen av omärkt satellitdata som finns tillgänglig för att öppna nya klassificeringslösningar. Denna avhandling föreslår en Semi-Supervised Learning-strategi baserad på Generative Adversarial Networks. En sådan arkitektur tar som input både märkt och omärkt data, och matar ut klassificeringsresultat som utnyttjar den kunskap som hämtats från båda datakällorna. Dess klassificeringsprestanda ut- värderas och jämförs senare med tillvägagångssättet Supervised Learning och metoden Transfer Learning baserat på förtränade nätverk. Detta arbete bevisar empiriskt att Semi-Supervised Generative Adversarial Network överträffar metoden Supervised Learning och förbättrar dess totala noggrannhet med minst 5% i konfigurationer med mindre än 100 tränings- märkta prover tillgängliga i användningsfallen under utvärdering, vilket uppnår prestanda som både är jämförbar med Transfer Learning-metoden och överlägsen jämte den under specifika experimentella konfigurationer. Ytterligare analyser utförs sedan för att belysa effektiviteten hos den föreslagna lösningen.

<p>Problem nedostajućih podataka je često prisutan prilikom razvoja<br />prediktivnih modela. Umesto uklanjanja podataka koji sadrže<br />vrednosti koje nedostaju mogu se primeniti metode za njihovu<br />imputaciju. Disertacija predlaže metodologiju za pristup analizi<br />uspešnosti imputacija prilikom razvoja prediktivnih modela. Na<br />osnovu iznete metodologije prikazuju se rezultati primene algoritama<br />mašinskog učenja, kao metoda imputacije, prilikom razvoja određenih,<br />konkretnih prediktivnih modela.</p> / <p>The problem of missing data is often present when developing predictive<br />models. Instead of removing data containing missing values, methods for<br />imputation can be applied. The dissertation proposes a methodology for<br />analysis of imputation performance in the development of predictive models.<br />Based on the proposed methodology, results of the application of machine<br />learning algorithms, as an imputation method in the development of specific<br />models, are presented.</p>