Academic fulltext search engine Egothor has recently became starting point of several thesis aimed on searching. Until now, there was no solution available to provide robust set of web content processing tools. This master thesis is aiming on design and implementation of distributed search system working primary with internet sources. We analyze first generation components for processing of web content and summarize their primary features. We use those features to propose architecture of distributed web search engine. We aim mainly to phases of data fetching, processing and indexing. We also describe final implementation of such system and propose few ideas for future extensions.

Jamais les films n’ont été aussi facilement accessibles aux spectateurs qui peuvent profiter de leur potentiel presque sans limite à susciter des émotions. Savoir à l’avance les émotions qu’un film est susceptible d’induire à ses spectateurs pourrait donc aider à améliorer la précision des systèmes de distribution de contenus, d’indexation ou même de synthèse des vidéos. Cependant, le transfert de cette expertise aux ordinateurs est une tâche complexe, en partie due à la nature subjective des émotions. Cette thèse est donc dédiée à la détection automatique des émotions induites par les films, basée sur les propriétés intrinsèques du signal audiovisuel. Pour s’atteler à cette tâche, une base de données de vidéos annotées selon les émotions induites aux spectateurs est nécessaire. Cependant, les bases de données existantes ne sont pas publiques à cause de problèmes de droit d’auteur ou sont de taille restreinte. Pour répondre à ce besoin spécifique, cette thèse présente le développement de la base de données LIRIS-ACCEDE. Cette base a trois avantages principaux: (1) elle utilise des films sous licence Creative Commons et peut donc être partagée sans enfreindre le droit d’auteur, (2) elle est composée de 9800 extraits vidéos de bonne qualité qui proviennent de 160 films et courts métrages, et (3) les 9800 extraits ont été classés selon les axes de “valence” et “arousal” induits grâce un protocole de comparaisons par paires mis en place sur un site de crowdsourcing. L’accord inter-annotateurs élevé reflète la cohérence des annotations malgré la forte différence culturelle parmi les annotateurs. Trois autres expériences sont également présentées dans cette thèse. Premièrement, des scores émotionnels ont été collectés pour un sous-ensemble de vidéos de la base LIRIS-ACCEDE dans le but de faire une validation croisée des classements obtenus via crowdsourcing. Les scores émotionnels ont aussi rendu possible l’apprentissage d’un processus gaussien par régression, modélisant le bruit lié aux annotations, afin de convertir tous les rangs liés aux vidéos de la base LIRIS-ACCEDE en scores émotionnels définis dans l’espace 2D valence-arousal. Deuxièmement, des annotations continues pour 30 films ont été collectées dans le but de créer des modèles algorithmiques temporellement fiables. Enfin, une dernière expérience a été réalisée dans le but de mesurer de façon continue des données physiologiques sur des participants regardant les 30 films utilisés lors de l’expérience précédente. La corrélation entre les annotations physiologiques et les scores continus renforce la validité des résultats de ces expériences. Equipée d’une base de données, cette thèse présente un modèle algorithmique afin d’estimer les émotions induites par les films. Le système utilise à son avantage les récentes avancées dans le domaine de l’apprentissage profond et prend en compte la relation entre des scènes consécutives. Le système est composé de deux réseaux de neurones convolutionnels ajustés. L’un est dédié à la modalité visuelle et utilise en entrée des versions recadrées des principales frames des segments vidéos, alors que l’autre est dédié à la modalité audio grâce à l’utilisation de spectrogrammes audio. Les activations de la dernière couche entièrement connectée de chaque réseau sont concaténées pour nourrir un réseau de neurones récurrent utilisant des neurones spécifiques appelés “Long-Short-Term- Memory” qui permettent l’apprentissage des dépendances temporelles entre des segments vidéo successifs. La performance obtenue par le modèle est comparée à celle d’un modèle basique similaire à l’état de l’art et montre des résultats très prometteurs mais qui reflètent la complexité de telles tâches. En effet, la prédiction automatique des émotions induites par les films est donc toujours une tâche très difficile qui est loin d’être complètement résolue. / Never before have movies been as easily accessible to viewers, who can enjoy anywhere the almost unlimited potential of movies for inducing emotions. Thus, knowing in advance the emotions that a movie is likely to elicit to its viewers could help to improve the accuracy of content delivery, video indexing or even summarization. However, transferring this expertise to computers is a complex task due in part to the subjective nature of emotions. The present thesis work is dedicated to the automatic prediction of emotions induced by movies based on the intrinsic properties of the audiovisual signal. To computationally deal with this problem, a video dataset annotated along the emotions induced to viewers is needed. However, existing datasets are not public due to copyright issues or are of a very limited size and content diversity. To answer to this specific need, this thesis addresses the development of the LIRIS-ACCEDE dataset. The advantages of this dataset are threefold: (1) it is based on movies under Creative Commons licenses and thus can be shared without infringing copyright, (2) it is composed of 9,800 good quality video excerpts with a large content diversity extracted from 160 feature films and short films, and (3) the 9,800 excerpts have been ranked through a pair-wise video comparison protocol along the induced valence and arousal axes using crowdsourcing. The high inter-annotator agreement reflects that annotations are fully consistent, despite the large diversity of raters’ cultural backgrounds. Three other experiments are also introduced in this thesis. First, affective ratings were collected for a subset of the LIRIS-ACCEDE dataset in order to cross-validate the crowdsourced annotations. The affective ratings made also possible the learning of Gaussian Processes for Regression, modeling the noisiness from measurements, to map the whole ranked LIRIS-ACCEDE dataset into the 2D valence-arousal affective space. Second, continuous ratings for 30 movies were collected in order develop temporally relevant computational models. Finally, a last experiment was performed in order to collect continuous physiological measurements for the 30 movies used in the second experiment. The correlation between both modalities strengthens the validity of the results of the experiments. Armed with a dataset, this thesis presents a computational model to infer the emotions induced by movies. The framework builds on the recent advances in deep learning and takes into account the relationship between consecutive scenes. It is composed of two fine-tuned Convolutional Neural Networks. One is dedicated to the visual modality and uses as input crops of key frames extracted from video segments, while the second one is dedicated to the audio modality through the use of audio spectrograms. The activations of the last fully connected layer of both networks are conv catenated to feed a Long Short-Term Memory Recurrent Neural Network to learn the dependencies between the consecutive video segments. The performance obtained by the model is compared to the performance of a baseline similar to previous work and shows very promising results but reflects the complexity of such tasks. Indeed, the automatic prediction of emotions induced by movies is still a very challenging task which is far from being solved.

Background: Generative Adversarial Networks (Goodfellow et al., 2014) (GANs)are the current state of the art machine learning data generating systems. Designed with two neural networks in the initial architecture proposal, generator and discriminator. These neural networks compete in a zero-sum game technique, to generate data having realistic properties inseparable to that of original datasets. GANs have interesting applications in various domains like Image synthesis, 3D object generation in gaming industry, fake music generation(Dong et al.), text to image synthesis and many more. Despite having a widespread application domains, GANs are popular for image data synthesis. Various architectures have been developed for image synthesis evolving from fuzzy images of digits to photorealistic images. Objectives: In this research work, we study various literature on different GAN architectures. To understand significant works done essentially to improve the GAN architectures. The primary objective of this research work is synthesis of plant images using Style GAN (Karras, Laine and Aila, 2018) variant of GAN using style transfer. The research also focuses on identifying various machine learning performance evaluation metrics that can be used to measure Style GAN model for the generated image datasets. Methods: A mixed method approach is used in this research. We review various literature work on GANs and elaborate in detail how each GAN networks are designed and how they evolved over the base architecture. We then study the style GAN (Karras, Laine and Aila, 2018a) design details. We then study related literature works on GAN model performance evaluation and measure the quality of generated image datasets. We conduct an experiment to implement the Style based GAN on leaf dataset(Kumar et al., 2012) to generate leaf images that are similar to the ground truth. We describe in detail various steps in the experiment like data collection, preprocessing, training and configuration. Also, we evaluate the performance of Style GAN training model on the leaf dataset. Results: We present the results of literature review and the conducted experiment to address the research questions. We review and elaborate various GAN architecture and their key contributions. We also review numerous qualitative and quantitative evaluation metrics to measure the performance of a GAN architecture. We then present the generated synthetic data samples from the Style based GAN learning model at various training GPU hours and the latest synthetic data sample after training for around ~8 GPU days on leafsnap dataset (Kumar et al., 2012). The results we present have a decent quality to expand the dataset for most of the tested samples. We then visualize the model performance by tensorboard graphs and an overall computational graph for the learning model. We calculate the Fréchet Inception Distance score for our leaf Style GAN and is observed to be 26.4268 (the lower the better). Conclusion: We conclude the research work with an overall review of sections in the paper. The generated fake samples are much similar to the input ground truth and appear to be convincingly realistic for a human visual judgement. However, the calculated FID score to measure the performance of the leaf StyleGAN accumulates a large value compared to that of Style GANs original celebrity HD faces image data set. We attempted to analyze the reasons for this large score.

Les entreprises, les administrations, et parfois les particuliers, doivent faire face à de nombreuses fraudes sur les documents qu’ils reçoivent de l’extérieur ou qu’ils traitent en interne. Les factures, les notes de frais, les justificatifs... tout document servant de preuve peut être falsifié dans le but de gagner plus d’argent ou de ne pas en perdre. En France, on estime les pertes dues aux fraudes à plusieurs milliards d’euros par an. Étant donné que le flux de documents échangés, numériques ou papiers, est très important, il serait extrêmement coûteux en temps et en argent de les faire tous vérifier par des experts de la détection des fraudes. C’est pourquoi nous proposons dans notre thèse un système de détection automatique des faux documents. Si la plupart des travaux en détection automatique des faux documents se concentrent sur des indices graphiques, nous cherchons quant à nous à vérifier les informations textuelles du document afin de détecter des incohérences ou des invraisemblances. Pour cela, nous avons tout d’abord constitué un corpus de tickets de caisse que nous avons numérisés et dont nous avons extrait le texte. Après avoir corrigé les sorties de l’OCR et fait falsifier une partie des documents, nous en avons extrait les informations et nous les avons modélisées dans une ontologie, afin de garder les liens sémantiques entre elles. Les informations ainsi extraites, et augmentées de leurs possibles désambiguïsations, peuvent être vérifiées les unes par rapport aux autres au sein du document et à travers la base de connaissances constituée. Les liens sémantiques de l’ontologie permettent également de chercher l’information dans d’autres sources de connaissances, et notamment sur Internet. / Companies, administrations, and sometimes individuals, have to face many frauds on documents they receive from outside or process internally. Invoices, expense reports, receipts...any document used as proof can be falsified in order to earn more money or not to lose it. In France, losses due to fraud are estimated at several billion euros per year. Since the flow of documents exchanged, whether digital or paper, is very important, it would be extremely costly and time-consuming to have them all checked by fraud detection experts. That’s why we propose in our thesis a system for automatic detection of false documents. While most of the work in automatic document detection focuses on graphic clues, we seek to verify the textual information in the document in order to detect inconsistencies or implausibilities.To do this, we first compiled a corpus of documents that we digitized. After correcting the characters recognition outputs and falsifying part of the documents, we extracted the information and modelled them in an ontology, in order to keep the semantic links between them. The information thus extracted, and increased by its possible disambiguation, can be verified against each other within the document and through the knowledge base established. The semantic links of ontology also make it possible to search for information in other sources of knowledge, particularly on the Internet.

<p>The problem of estimating an adequate level of detail for an object for a specific view is one of the important problems in computer 3d-graphics and is especially important in real-time applications. The well-known continuous level-of-detail technique, Real-time Optimally Adapting Meshes (ROAM), has been employed with success for almost 10 years but has at present, due to rapid development of graphics hardware, been found to be inadequate. Compared to many other level-of-detail techniques it cannot benefit from the higher triangle throughput available on graphics cards of today.</p><p>This thesis will describe the implementation of the new version of ROAM (informally known as ROAM 2.0) for the purpose of massive planetary data visualization. It will show how the problems of the old technique can be bridged to be able to adapt to newer graphics card while still benefiting from the advantages of ROAM. The resulting implementation that is presented here is specialized on spherical objects and handles both texture and geometry data of arbitrary large sizes in an efficient way.</p>

Ground-motion data processing is a necessity for most earthquake engineering related studies. Important engineering parameters such as the peak values of ground motion and the ordinates of the response spectra are determined from the strong ground-motion data recorded by accelerometers. However, the raw data needs to be processed since the recorded data always contains high- and low-frequency noise from different sources. Low-cut filters are the most popular ground-motion data processing scheme for removing long-period noise. Removing long-period noise from the raw accelogram is important since the displacement spectrum that provides primary information about deformation demands on structural systems is highly sensitive to the long-period noise. The objective of this study is to investigate the effect of low-cut filtering period on linear and nonlinear deformation demands. A large number of strong ground motions from Europe and the Middle East representing different site classes as well as different magnitude and distance ranges are used to conduct statistical analysis. The statistical results are used to investigate the influence of low-cut filter period on spectral displacements. The results of the study are believed to be useful for future generation ground-motion prediction equations on deformation demands that are of great importance in performance-based earthquake engineering.

<p>This project presents the research on software visualization techniques. We will introduce the concepts of software visualization, software metrics and our proposed visualization techniques: Static Visualization (glyphs object with static texture) and Dynamic Visualization (glyphs object with moving object). Our intent to study the existing visualization techniques for visualization of software</p><p>metrics and then proposed the new visualization approach that is more time efficient and easy to perceive by viewer. In this project, we focus on the practical aspects of visualization of multivariate dataset. This project also gives an implementation of proposed visualization techniques of software metrics. In this research based work, we have to compare practically the proposed visualization approaches. We will discuss the software development life cycle of our proposed visualization system, and we will also describe the complete software implementation of implemented software.</p>

Tsunami are a fascinating but potentially devastating natural phenomena that have occurred regularly throughout history along New Zealand's shorelines, and around the world. With increasing population and the construction of infrastructure in coastal zones, the effect of these large waves has become a major concern. Many natural phenomena are capable of creating tsunami. Of particular concern is the underwater landslide-induced tsunami, due to the potentially short warning before waves reach the shore. The aims of this research are to generate a quality benchmark dataset suitable for comprehensive comparisons with numerical model results and to increase our understanding of the physical processes involved in tsunami generation. The two-dimensional experimental configuration is based on a benchmark configuration described in the scientific literature, consisting of a semi-elliptical prism sliding down a submerged 15° slope. A unique feature of these experiments is the method developed to measure water surface variation continuously in both space and time. Water levels are obtained using an optical technique based on laser induced fluorescence, which is shown to be comparable in accuracy and resolution to traditional electrical point wave gauges. In the experiments, the landslide density and initial submergence are varied and detailed measurements of wave heights, lengths, propagation speeds, and shore run-up are made. Particle tracking velocimetry is used to record the landslide kinematics and sub-surface water velocities. Particular attention is paid to maintaining a high level of test repeatability throughout the experimental process. The experimental results show that a region of high pressure ahead of the landslide forces up the water over the front half of the landslide to form the leading wave crest, which propagates ahead of the landslide. The accelerating fluid above, and the turbulent wake behind, the moving landslide create a region of low pressure, which draws down the water surface above the rear half of the landslide to form the leading trough. Differences in the phase and group velocities of the components in the wave packet cause waves to be continually generated on the trailing end of the wave train. The downstream position that these waves form continually moves downstream with time and the wave packet is found to be highly dispersive. The interaction of the landslide pressure field with the free surface wave pressure field is important, as the location of the low pressure around the landslide relative to the wave field acts to reinforce or suppress the waves above. This has a substantial effect on the increase or decrease in wave potential energy. When the low pressure acts to draw down a wave trough, the wave potential energy increases. When the low pressure is below a wave crest, it acts to suppress the crest amplitude, leading to an overall decrease in wave potential energy. Measurements of the efficiency of energy transfer from the landslide to the wave field show that the ratio of maximum wave potential energy to maximum landslide kinetic energy is between 0.028 and 0.138, and tends to increase for shallower initial landslide submergences and heavier specific gravities. The ratio of maximum wave potential energy to maximum landslide potential energy ranges between 0.011 and 0.059 and tends to be greater for shallower initial submergences. For two experimental configurations the ratio of maximum wave potential energy to maximum fluid kinetic energy is estimated to be 0.435 and 0.588. The wave trough initially generated above the rear end of the landslide propagates in both onshore and offshore directions. The onshore-propagating trough causes a large initial draw-down at the shore. The magnitude of the maximum draw-down is related to the maximum amplitude of the offshore-propagating first wave trough. A wave crest generated by the landslide as it decelerates at the bottom of the slope causes the maximum wave run-up observed at the shore. A semi-analytical model, based on inviscid and irrotational theory, is used to investigate the wave generation process of a moving submerged object in a constant depth channel. The simplified geometry allows a variety of phenomena, observed during the experimental tests, to be investigated further in a more controlled setting. The variations in the growth, magnitude, and decay of energy as a function of time is due the interaction of the pressure distribution surrounding the moving slider with the wave field, in particular, the leading crest and trough. The largest energy transfer between slider kinetic energy and wave potential energy occurs when there is prolonged interaction between the slider's low pressure region and the leading wave trough. The generation of onshore propagating waves by a decelerating landslide is confirmed, and the magnitude of the maximum wave run-up is found to be dependent on the magnitude of the slider deceleration. The model also shows that slides with Froude number close to unity convert substantial amounts of energy into offshore propagating waves. The onshore propagating wave potential energy is not as sensitive to Froude number. A further result from the model simulations is that the specific shape of the slider has only a minor influence on the wave response, provided the slider's length and area are known. A boundary element model, based on inviscid and irrotational theory, is used to simulate the laboratory experiments. Model predictions of the wave field are generally accurate, particularly the magnitude and range of wave amplitudes within the wave packet, the arrival time of the wave group, the amplitude of the run-up and run-down at the shore, the time the maximum run-down occurs, and the form and magnitude of the wave potential energy time history. The ratios of maximum wave potential energy to maximum slider kinetic energy are predicted to within ± 29%. The model predictions of the crest arrival times are within 3.6% of the measured times. The inability of the inviscid and irrotational model to simulate the flow separation and wake motions lead to a 45% under prediction of the maximum fluid kinetic energy. Both the semi-analytical and BEM models highlight the need for the correct specification of initial slider accelerations in numerical simulations in order to accurately predict the wave energy.

Privacy preservation is important for machine learning and datamining, but measures designed to protect private information sometimes result in a trade off: reduced utility of the training samples. This thesis introduces a privacy preserving approach that can be applied to decision-tree learning, without concomitant loss of accuracy. It describes an approach to the preservation of privacy of collected data samples in cases when information of the sample database has been partially lost. This approach converts the original sample datasets into a group of unreal datasets, where an original sample cannot be reconstructed without the entire group of unreal datasets. This approach does not perform well for sample datasets with low frequency, or when there is low variance in the distribution of all samples. However, this problem can be solved through a modified implementation of the approach introduced later in this thesis, by using some extra storage.

We present a multifaceted investigation into the relevance of word order in machine translation. We introduce two tools, DTED and DERP, each using dependency structure to detect differences between the structures of machine-produced translations and human-produced references. DTED applies the principle of Tree Edit Distance to calculate edit operations required to convert one structure into another. Four variants of DTED have been produced, differing in the importance they place on words which match between the two sentences. DERP represents a more detailed procedure, making use of the dependency relations between words when evaluating the disparities between paths connecting matching nodes. In order to empirically evaluate DTED and DERP, and as a standalone contribution, we have produced WOJ-DB, a database of human judgments. Containing scores relating to translation adequacy and more specifically to word order quality, this is intended to support investigations into a wide range of translation phenomena. We report an internal evaluation of the information in WOJ-DB, then use it to evaluate variants of DTED and DERP, both to determine their relative merit and their strength relative to third-party baselines. We present our conclusions about the importance of structure to the tools and their relevance to word order specifically, then propose further related avenues of research suggested or enabled by our work.