Différentes disciplines des sciences humaines telles la philologie ou la paléographie font face à des tâches complexes et fastidieuses pour l'examen des sources de données. La proposition d'approches computationnelles en humanités permet d'adresser les problématiques rencontrées telles que la lecture, l'analyse et l'archivage de façon systématique. Les modèles conceptuels élaborés reposent sur des algorithmes et ces derniers donnent lieu à des implémentations informatiques qui automatisent ces tâches fastidieuses. La première partie de la thèse vise, d'une part, à établir la structuration thématique d'un corpus, en construisant des espaces sémantiques de grande dimension. D'autre part, elle vise au suivi dynamique des thématiques qui constitue un réel défi scientifique, notamment en raison du passage à l'échelle. La seconde partie de la thèse traite de manière holistique la page d'un document numérisé sans aucune intervention préalable. Le but est d'apprendre automatiquement des représentations du trait de l'écriture ou du tracé d'un certain script par rapport au tracé d'un autre script. Il faut dans ce cadre tenir compte de l'environnement où se trouve le tracé : image, artefact, bruits dus à la détérioration de la qualité du papier, etc. Notre approche propose un empilement de réseaux de neurones auto-encodeurs afin de fournir une représentation alternative des données reçues en entrée. / Different disciplines in the humanities, such as philology or palaeography, face complex and time-consuming tasks whenever it comes to examining the data sources. The introduction of computational approaches in humanities makes it possible to address issues such as semantic analysis and systematic archiving. The conceptual models developed are based on algorithms that are later hard coded in order to automate these tedious tasks. In the first part of the thesis we propose a novel method to build a semantic space based on topics modeling. In the second part and in order to classify historical documents according to their script. We propose a novel representation learning method based on stacking convolutional auto-encoder. The goal is to automatically learn plot representations of the script or the written language.

Financial institutions today generate a large amount of data, data that can contain interesting information to investigate to further the economic growth of said institution. There exists an interest in analyzing these points of information, especially if they are anomalous from the normal day-to-day work. However, to find these outliers is not an easy task and not possible to do manually due to the massive amounts of data being generated daily. Previous work to solve this has explored the usage of machine learning to find outliers in these financial datasets. Previous studies have shown that the pre-processing of data usually stands for a big part in information loss. This work aims to study if there is a proper balance in how the pre-processing is carried out to retain the highest amount of information while simultaneously not letting the data remain too complex for the machine learning models. The dataset used consisted of Foreign exchange transactions supplied by the host company and was pre-processed through the use of Principal Component Analysis (PCA). The main purpose of this work is to test if an ensemble of Long Short-Term Memory Recurrent Neural Networks (LSTM), configured as autoencoders, can be used to detect outliers in the data and if the ensemble is more accurate than a single LSTM autoencoder. Previous studies have shown that Ensemble autoencoders can prove more accurate than a single autoencoder, especially when SkipCells have been implemented (a configuration that skips over LSTM cells to make the model perform with more variation). A datapoint will be considered an outlier if the LSTM model has trouble properly recreating it, i.e. a pattern that is hard to classify, making it available for further investigations done manually. The results show that the ensembled LSTM model proved to be more accurate than that of a single LSTM model in regards to reconstructing the dataset, and by our definition of an outlier, more accurate in outlier detection. The results from the pre-processing experiments reveal different methods of obtaining an optimal number of components for your data. One of those is by studying retained variance and accuracy of PCA transformation compared to model performance for a certain number of components. One of the conclusions from the work is that ensembled LSTM networks can prove very powerful, but that alternatives to pre-processing should be explored such as categorical embedding instead of PCA. / Finansinstitut genererar idag en stor mängd data, data som kan innehålla intressant information värd att undersöka för att främja den ekonomiska tillväxten för nämnda institution. Det finns ett intresse för att analysera dessa informationspunkter, särskilt om de är avvikande från det normala dagliga arbetet. Att upptäcka dessa avvikelser är dock inte en lätt uppgift och ej möjligt att göra manuellt på grund av de stora mängderna data som genereras dagligen. Tidigare arbete för att lösa detta har undersökt användningen av maskininlärning för att upptäcka avvikelser i finansiell data. Tidigare studier har visat på att förbehandlingen av datan vanligtvis står för en stor del i förlust av emphinformation från datan. Detta arbete syftar till att studera om det finns en korrekt balans i hur förbehandlingen utförs för att behålla den högsta mängden information samtidigt som datan inte förblir för komplex för maskininlärnings-modellerna. Det emphdataset som användes bestod av valutatransaktioner som tillhandahölls av värdföretaget och förbehandlades genom användning av Principal Component Analysis (PCA). Huvudsyftet med detta arbete är att undersöka om en ensemble av Long Short-Term Memory Recurrent Neural Networks (LSTM), konfigurerad som autoenkodare, kan användas för att upptäcka avvikelser i data och om ensemblen är mer precis i sina predikteringar än en ensam LSTM-autoenkodare. Tidigare studier har visat att en ensembel avautoenkodare kan visa sig vara mer precisa än en singel autokodare, särskilt när SkipCells har implementerats (en konfiguration som hoppar över vissa av LSTM-cellerna för att göra modellerna mer varierade). En datapunkt kommer att betraktas som en avvikelse om LSTM-modellen har problem med att återskapa den väl, dvs ett mönster som nätverket har svårt att återskapa, vilket gör datapunkten tillgänglig för vidare undersökningar. Resultaten visar att en ensemble av LSTM-modeller predikterade mer precist än en singel LSTM-modell när det gäller att återskapa datasetet, och då enligt vår definition av avvikelser, mer precis avvikelse detektering. Resultaten från förbehandlingen visar olika metoder för att uppnå ett optimalt antal komponenter för dina data genom att studera bibehållen varians och precision för PCA-transformation jämfört med modellprestanda. En av slutsatserna från arbetet är att en ensembel av LSTM-nätverk kan visa sig vara mycket kraftfulla, men att alternativ till förbehandling bör undersökas, såsom categorical embedding istället för PCA.

In the hope to increase the detection rate of faults in combined heat and power plant boilers thus lowering unplanned maintenance three machine learning models are constructed and evaluated. The algorithms; k-Nearest Neighbor, One-Class Support Vector Machine, and Auto-encoder have a proven track record in research for anomaly detection, but are relatively unexplored for industrial applications such as this one due to the difficulty in collecting non-artificial labeled data in the field.The baseline versions of the k-Nearest Neighbor and Auto-encoder performed very similarly. Nevertheless, the Auto-encoder was slightly better and reached an area under the precision-recall curve (AUPRC) of 0.966 and 0.615 on the trainingand test period, respectively. However, no sufficiently good results were reached with the One-Class Support Vector Machine. The Auto-encoder was made more sophisticated to see how much performance could be increased. It was found that the AUPRC could be increased to 0.987 and 0.801 on the trainingand test period, respectively. Additionally, the model was able to detect and generate one alarm for each incident period that occurred under the test period.The conclusion is that ML can successfully be utilized to detect faults at an earlier stage and potentially circumvent otherwise costly unplanned maintenance. Nevertheless, there is still a lot of room for improvements in the model and the collection of the data. / I hopp om att öka identifieringsgraden av störningar i kraftvärmepannor och därigenom minska oplanerat underhåll konstrueras och evalueras tre maskininlärningsmodeller.Algoritmerna; k-Nearest Neighbor, One-Class Support Vector Machine, och Autoencoder har bevisad framgång inom forskning av anomalidetektion, men är relativt outforskade för industriella applikationer som denna på grund av svårigheten att samla in icke-artificiell uppmärkt data inom området.Grundversionerna av k-Nearest Neighbor och Auto-encoder presterade nästan likvärdigt. Dock var Auto-encoder-modellen lite bättre och nådde ett AUPRC-värde av 0.966 respektive 0.615 på träningsoch testperioden. Inget tillräckligt bra resultat nåddes med One-Class Support Vector Machine. Auto-encoder-modellen gjordes mer sofistikerad för att se hur mycket prestandan kunde ökas. Det visade sig att AUPRC-värdet kunde ökas till 0.987 respektive 0.801 under träningsoch testperioden. Dessutom lyckades modellen identifiera och generera ett larm vardera för alla incidenter under testperioden. Slutsatsen är att ML framgångsrikt kan användas för att identifiera störningar iett tidigare skede och därigenom potentiellt kringgå i annat fall dyra oplanerade underhåll. Emellertid finns det fortfarande mycket utrymme för förbättringar av modellen samt inom insamlingen av data.

This thesis evaluates the performance of different machine learning approaches to log classification based on a dataset derived from simulating intrusive behavior towards an enterprise web application. The first experiment consists of performing attacks towards the web app in correlation with the logs to create a labeled dataset. The second experiment consists of one unsupervised model based on a variational autoencoder and four super- vised models based on both conventional feature-engineering techniques with deep neural networks and embedding-based feature techniques followed by long-short-term memory architectures and convolutional neural networks. With this dataset, the embedding-based approaches performed much better than the conventional one. The autoencoder did not perform well compared to the supervised models. To conclude, embedding-based ap- proaches show promise even on datasets with different characteristics compared to natural language.

Techniques for imposing a structure on the latent space of neural networks have seen much development in recent years. Clustering techniques used for classification have been used to great success, and with this work we hope to bridge the gap between contrastive losses and Generative models. We introduce an embedding loss derived from Triplet loss to show that attributes and information can be clustered in specific dimensions in the latent space of Variational Auto-Encoders. This allows control over the embedded attributes via manipulation of these latent space dimensions. This work also serves to take steps towards the usage of any data augmentation when applying Triplet loss to Variational Auto-Encoders. In this work three different Variational Auto-Encoders are trained on three different datasets to embed information in three different ways using this novel method. Our results show the method working to varying degrees depending on the implementation and the information embedded. Two experiments using image data and one using waveform audio shows that the method is modality invariant. / Tekniker för att införa en struktur i det latenta utrymmet i neurala nätverk har sett mycket utveckling under de senaste åren. Kluster metoder som används för klassificering har använts till stor framgång, och med detta arbete hoppas vi kunna brygga gapet mellan kontrastiva förlustfunktioner och generativa modeller. Vi introducerar en förlustfunktion för inbäddning härledd från triplet loss för att visa att attribut och information kan klustras i specifika dimensioner i det latenta utrymmet hos variationsautokodare. Detta tillåter kontroll över de inbäddade attributen via manipulering av dessa dimensioner i latenta utrymmet. Detta arbete tjänar också till att ta steg mot användningen av olika data augmentationer när triplet loss tillämpas på generativa modeller. Tre olika Variationsautokodare tränas på tre olika dataset för att bädda in information på tre olika sätt med denna nya metod. Våra resultat visar att metoden fungerar i varierande grad beroende på hur den tillämpas och vilken information som inbäddas. Två experiment använder bild-data och ett använder sig av ljud, vilket visar på att metoden är modalitetsinvariant.

Backdoor (Trojan) attacks are a major threat to the security of deep neural network (DNN) models. They are created by an attacker who adds a certain pattern to a portion of given training dataset, causing the DNN model to misclassify any inputs that contain the pattern. These infected classifiers are called Trojan models and the added pattern is referred to as the trigger. In image domain, a trigger can be a patch of pixel values added to the images and in text domain, it can be a set of words. In this thesis, we propose Trojan-Miner (T-Miner), a defense scheme against such backdoor attacks on text classification deep learning models. The goal of T-Miner is to detect whether a given classifier is a Trojan model or not. To create T-Miner , our approach is based on a sequence-to-sequence text generation model. T-Miner uses feedback from the suspicious (test) classifier to perturb input sentences such that their resulting class label is changed. These perturbations can be different for each of the inputs. T-Miner thus extracts the perturbations to determine whether they include any backdoor trigger and correspondingly flag the suspicious classifier as a Trojan model. We evaluate T-Miner on three text classification datasets: Yelp Restaurant Reviews, Twitter Hate Speech, and Rotten Tomatoes Movie Reviews. To illustrate the effectiveness of T-Miner, we evaluate it on attack models over text classifiers. Hence, we build a set of clean classifiers with no trigger in their training datasets and also using several trigger phrases, we create a set of Trojan models. Then, we compute how many of these models are correctly marked by T-Miner. We show that our system is able to detect trojan and clean models with 97% overall accuracy over 400 classifiers. Finally, we discuss the robustness of T-Miner in the case that the attacker knows T-Miner framework and wants to use this knowledge to weaken T-Miner performance. To this end, we propose four different scenarios for the attacker and report the performance of T-Miner under these new attack methods. / M.S. / Backdoor (Trojan) attacks are a major threat to the security of predictive models that make use of deep neural networks. The idea behind these attacks is as follows: an attacker adds a certain pattern to a portion of given training dataset and in the next step, trains a predictive model over this dataset. As a result, the predictive model misclassifies any inputs that contain the pattern. In image domain this pattern that is called trigger, can be a patch of pixel values added to the images and in text domain, it can be a set of words. In this thesis, we propose Trojan-Miner (T-Miner), a defense scheme against such backdoor attacks on text classification deep learning models. The goal of T-Miner is to detect whether a given classifier is a Trojan model or not. T-Miner is based on a sequence-to-sequence text generation model that is connected to the given predictive model and determine if the predictive model is being backdoor attacked. When T-Miner is connected to the predictive model, it generates a set of words, called perturbations, and analyses these perturbations to determine whether they include any backdoor trigger. Hence if any part of the trigger is present in the perturbations, the predictive model is flagged as a Trojan model. We evaluate T-Miner on three text classification datasets: Yelp Restaurant Reviews, Twitter Hate Speech, and Rotten Tomatoes Movie Reviews. To illustrate the effectiveness of T-Miner, we evaluate it on attack models over text classifiers. Hence, we build a set of clean classifiers with no trigger in their training datasets and also using several trigger phrases, we create a set of Trojan models. Then, we compute how many of these models are correctly marked by T-Miner. We show that our system is able to detect Trojan models with 97% overall accuracy over 400 predictive models.

L'évaluation de la démarche humaine est l'une des composantes essentielles dans les soins de santé. Les systèmes à base de marqueurs avec plusieurs caméras sont largement utilisés pour faire cette analyse. Cependant, ces systèmes nécessitent généralement des équipements spécifiques à prix élevé et/ou des moyens de calcul intensif. Afin de réduire le coût de ces dispositifs, nous nous concentrons sur un système d'analyse de la marche qui utilise une seule caméra de profondeur. Le principe de notre travail est similaire aux systèmes multi-caméras, mais l'ensemble de caméras est remplacé par un seul capteur de profondeur et des miroirs. Chaque miroir dans notre configuration joue le rôle d'une caméra qui capture la scène sous un point de vue différent. Puisque nous n'utilisons qu'une seule caméra, il est ainsi possible d'éviter l'étape de synchronisation et également de réduire le coût de l'appareillage. Notre thèse peut être divisée en deux sections: reconstruction 3D et analyse de la marche. Le résultat de la première section est utilisé comme entrée de la seconde. Notre système pour la reconstruction 3D est constitué d'une caméra de profondeur et deux miroirs. Deux types de capteurs de profondeur, qui se distinguent sur la base du mécanisme d'estimation de profondeur, ont été utilisés dans nos travaux. Avec la technique de lumière structurée (SL) intégrée dans le capteur Kinect 1, nous effectuons la reconstruction 3D à partir des principes de l'optique géométrique. Pour augmenter le niveau des détails du modèle reconstruit en 3D, la Kinect 2 qui estime la profondeur par temps de vol (ToF), est ensuite utilisée pour l'acquisition d'images. Cependant, en raison de réflections multiples sur les miroirs, il se produit une distorsion de la profondeur dans notre système. Nous proposons donc une approche simple pour réduire cette distorsion avant d'appliquer les techniques d'optique géométrique pour reconstruire un nuage de points de l'objet 3D. Pour l'analyse de la démarche, nous proposons diverses alternatives centrées sur la normalité de la marche et la mesure de sa symétrie. Cela devrait être utile lors de traitements cliniques pour évaluer, par exemple, la récupération du patient après une intervention chirurgicale. Ces méthodes se composent d'approches avec ou sans modèle qui ont des inconvénients et avantages différents. Dans cette thèse, nous présentons 3 méthodes qui traitent directement les nuages de points reconstruits dans la section précédente. La première utilise la corrélation croisée des demi-corps gauche et droit pour évaluer la symétrie de la démarche, tandis que les deux autres methodes utilisent des autoencodeurs issus de l'apprentissage profond pour mesurer la normalité de la démarche. / The problem of assessing human gaits has received a great attention in the literature since gait analysis is one of key components in healthcare. Marker-based and multi-camera systems are widely employed to deal with this problem. However, such systems usually require specific equipments with high price and/or high computational cost. In order to reduce the cost of devices, we focus on a system of gait analysis which employs only one depth sensor. The principle of our work is similar to multi-camera systems, but the collection of cameras is replaced by one depth sensor and mirrors. Each mirror in our setup plays the role of a camera which captures the scene at a different viewpoint. Since we use only one camera, the step of synchronization can thus be avoided and the cost of devices is also reduced. Our studies can be separated into two categories: 3D reconstruction and gait analysis. The result of the former category is used as the input of the latter one. Our system for 3D reconstruction is built with a depth camera and two mirrors. Two types of depth sensor, which are distinguished based on the scheme of depth estimation, have been employed in our works. With the structured light (SL) technique integrated into the Kinect 1, we perform the 3D reconstruction based on geometrical optics. In order to increase the level of details of the 3D reconstructed model, the Kinect 2 with time-of-flight (ToF) depth measurement is used for image acquisition instead of the previous generation. However, due to multiple reflections on the mirrors, depth distortion occurs in our setup. We thus propose a simple approach for reducing such distortion before applying geometrical optics to reconstruct a point cloud of the 3D object. For the task of gait analysis, we propose various alternative approaches focusing on the problem of gait normality/symmetry measurement. They are expected to be useful for clinical treatments such as monitoring patient's recovery after surgery. These methods consist of model-free and model-based approaches that have different cons and pros. In this dissertation, we present 3 methods that directly process point clouds reconstructed from the previous work. The first one uses cross-correlation of left and right half-bodies to assess gait symmetry while the other ones employ deep auto-encoders to measure gait normality.

La maladie d'Alzheimer (MA) est une maladie cérébrale dégénérative qui entraîne une perte progressive de la mémoire, un déclin cognitif et une détérioration graduelle de la capacité d'une personne à faire face à la complexité et à l'exigence des tâches quotidiennes nécessaires pour vivre en autonomie dans notre société actuelle. Les traitements pharmacologiques actuels peuvent ralentir le processus de dégradation attribué à la maladie, mais ces traitements peuvent également provoquer certains effets secondaires indésirables. L'un des traitements non pharmacologiques qui peut soulager efficacement les symptômes est la thérapie assistée par l'animal (T.A.A.). Mais en raison de certaines limitations telles que le prix des animaux et des problèmes d'hygiène, des animaux virtuels sont utilisés dans ce domaine. Cependant, les animaux virtuels animés, la qualité d'image approximative et le mode d'interaction unidirectionnel des animaux qui attendent passivement les instructions de l’utilisateur, peuvent difficilement stimuler le retour émotionnel entre l'utilisateur et les animaux virtuels, ce qui affaiblit considérablement l'effet thérapeutique. Cette étude vise à explorer l'efficacité de l'utilisation d'animaux virtuels à la place d’animaux vivants et leur impact sur la réduction des émotions négatives chez le patient. Cet objectif a été gardé à l'esprit lors de la conception du projet Zoo Therapy, qui présente un environnement immersif d'animaux virtuels en 3D, où l'impact sur l'émotion du patient est mesuré en temps réel par électroencéphalographie (EEG). Les objets statiques et les animaux virtuels de Zoo Therapy sont tous présentés à l'aide de modèles 3D réels. Les mouvements des animaux, les sons et les systèmes de repérage spécialement développés prennent en charge le comportement interactif simulé des animaux virtuels. De plus, pour que l'expérience d'interaction de l'utilisateur soit plus réelle, Zoo Therapy propose un mécanisme de communication novateur qui met en œuvre une interaction bidirectionnelle homme-machine soutenue par 3 méthodes d'interaction : le menu sur les panneaux, les instructions vocales et le Neurofeedback. La manière la plus directe d'interagir avec l'environnement de réalité virtuelle (RV) est le menu sur les panneaux, c'est-à-dire une interaction en cliquant sur les boutons des panneaux par le contrôleur de RV. Cependant, il était difficile pour certains utilisateurs ayant la MA d'utiliser le contrôleur de RV. Pour accommoder ceux qui ne sont pas bien adaptés ou compatibles avec le contrôleur de RV, un système d'instructions vocales peut être utilisé comme interface. Ce système a été reçu positivement par les 5 participants qui l'ont essayé. Même si l'utilisateur choisit de ne pas interagir activement avec l'animal virtuel dans les deux méthodes ci-dessus, le système de Neurofeedback guidera l'animal pour qu'il interagisse activement avec l'utilisateur en fonction des émotions de ce dernier. Le système de Neurofeedback classique utilise un système de règles pour donner des instructions. Les limites de cette méthode sont la rigidité et l'impossibilité de prendre en compte la relation entre les différentes émotions du participant. Pour résoudre ces problèmes, ce mémoire présente une méthode basée sur l'apprentissage par renforcement (AR) qui donne des instructions à différentes personnes en fonction des différentes émotions. Dans l'expérience de simulation des données émotionnelles synthétiques de la MD, la méthode basée sur l’AR est plus sensible aux changements émotionnels que la méthode basée sur les règles et peut apprendre automatiquement des règles potentielles pour maximiser les émotions positives de l'utilisateur. En raison de l'épidémie de Covid-19, nous n'avons pas été en mesure de mener des expériences à grande échelle. Cependant, un projet de suivi a combiné la thérapie de RV Zoo avec la reconnaissance des gestes et a prouvé son efficacité en évaluant les valeurs d'émotion EEG des participants. / Alzheimer’s disease (AD) is a degenerative brain disease that causes progressive memory loss, cognitive decline, and gradually impairs one’s ability to cope with the complexity and requirement of the daily routine tasks necessary to live in autonomy in our current society. Actual pharmacological treatments can slow down the degradation process attributed to the disease, but such treatments may also cause some undesirable side effects. One of the non-pharmacological treatments that can effectively relieve symptoms is animal-assisted treatment (AAT). But due to some limitations such as animal cost and hygiene issues, virtual animals are used in this field. However, the animated virtual animals, the rough picture quality presentation, and the one-direction interaction mode of animals passively waiting for the user's instructions can hardly stimulate the emotional feedback background between the user and the virtual animals, which greatly weakens the therapeutic effect. This study aims to explore the effectiveness of using virtual animals in place of their living counterpart and their impact on the reduction of negative emotions in the patient. This approach has been implemented in the Zoo Therapy project, which presents an immersive 3D virtual reality animal environment, where the impact on the patient’s emotion is measured in real-time by using electroencephalography (EEG). The static objects and virtual animals in Zoo Therapy are all presented using real 3D models. The specially developed animal movements, sounds, and pathfinding systems support the simulated interactive behavior of virtual animals. In addition, for the user's interaction experience to be more real, the innovation of this approach is also in its communication mechanism as it implements a bidirectional human-computer interaction supported by 3 interaction methods: Menu panel, Speech instruction, and Neurofeedback. The most straightforward way to interact with the VR environment is through Menu panel, i.e., interaction by clicking buttons on panels by the VR controller. However, it was difficult for some AD users to use the VR controller. To accommodate those who are not well suited or compatible with VR controllers, a speech instruction system can be used as an interface, which was received positively by the 5 participants who tried it. Even if the user chooses not to actively interact with the virtual animal in the above two methods, the Neurofeedback system will guide the animal to actively interact with the user according to the user's emotions. The mainstream Neurofeedback system has been using artificial rules to give instructions. The limitation of this method is inflexibility and cannot take into account the relationship between the various emotions of the participant. To solve these problems, this thesis presents a reinforcement learning (RL)-based method that gives instructions to different people based on multiple emotions accordingly. In the synthetic AD emotional data simulation experiment, the RL-based method is more sensitive to emotional changes than the rule-based method and can automatically learn potential rules to maximize the user's positive emotions. Due to the Covid-19 epidemic, we were unable to conduct large-scale experiments. However, a follow-up project combined VR Zoo Therapy with gesture recognition and proved the effectiveness by evaluating participant's EEG emotion values.

Data augmentation is a technique that can be performed in various ways to improve the training of discriminative models. The recent developments in deep generative models offer new ways of augmenting existing data sets. In this thesis, a framework for augmenting annotated data sets with deep generative models is proposed together with a method for quantitatively evaluating the quality of the generated data sets. Using this framework, two data sets for pupil localization was generated with different generative models, including both well-established models and a novel model proposed for this purpose. The unique model was shown both qualitatively and quantitatively to generate the best data sets. A set of smaller experiments on standard data sets also revealed cases where this generative model could improve the performance of an existing discriminative model. The results indicate that generative models can be used to augment or replace existing data sets when training discriminative models. / Dataaugmentering är en teknik som kan utföras på flera sätt för att förbättra träningen av diskriminativa modeller. De senaste framgångarna inom djupa generativa modeller har öppnat upp nya sätt att augmentera existerande dataset. I detta arbete har ett ramverk för augmentering av annoterade dataset med hjälp av djupa generativa modeller föreslagits. Utöver detta så har en metod för kvantitativ evaulering av kvaliteten hos genererade data set tagits fram. Med hjälp av detta ramverk har två dataset för pupillokalisering genererats med olika generativa modeller. Både väletablerade modeller och en ny modell utvecklad för detta syfte har testats. Den unika modellen visades både kvalitativt och kvantitativt att den genererade de bästa dataseten. Ett antal mindre experiment på standardiserade dataset visade exempel på fall där denna generativa modell kunde förbättra prestandan hos en existerande diskriminativ modell. Resultaten indikerar att generativa modeller kan användas för att augmentera eller ersätta existerande dataset vid träning av diskriminativa modeller.

Variational Auto-Encoders (VAEs) are one of the most famous deep generative models. After showing that standard VAEs may not preserve the topology, that is the shape of the data, between the input and the latent space, we tried to modify them so that the topology is preserved. This would help in particular for performing interpolations in the latent space. Our main contribution is two folds. Firstly, we propose successfully the InvMap-VAE which is a simple way to turn any dimensionality reduction technique, given its embedding, into a generative model within a VAE framework providing an inverse mapping, with all the advantages that this implies. Secondly, we propose the Witness Simplicial VAE as an extension of the Simplicial Auto-Encoder to the variational setup using a Witness Complex for computing a simplicial regularization. The Witness Simplicial VAE is independent of any dimensionality reduction technique and seems to better preserve the persistent Betti numbers of a data set than a standard VAE, although it would still need some further improvements. Finally, the two first chapters of this master thesis can also be used as an introduction to Topological Data Analysis, General Topology and Computational Topology (or Algorithmic Topology), for any machine learning student, engineer or researcher interested in these areas with no background in topology. / Variations autokodare (VAE) är en av de mest kända djupa generativa modellerna. Efter att ha visat att standard VAE inte nödvändigtvis bevarar topologiska egenskaper, det vill säga formen på datan, mellan inmatningsdatan och det latenta rummet, försökte vi modifiera den så att topologin är bevarad. Det här skulle i synnerhet underlätta när man genomför interpolering i det latenta rummet. Denna avhandling består av två centrala bidrag. I första hand så utvecklar vi InvMap-VAE, som är en enkel metod att omvandla vilken metod inom dimensionalitetsreducering, givet dess inbäddning, till en generativ modell inom VAE ramverket, vilket ger en invers avbildning och dess tillhörande fördelar. För det andra så presenterar vi Witness Simplicial VAE som en förlängning av en Simplicial Auto-Encoder till dess variationella variant genom att använda ett vittneskomplex för att beräkna en simpliciel regularisering. Witness Simplicial VAE är oberoende av dimensionalitets reducerings teknik och verkar bättre bevara Betti-nummer av ett dataset än en vanlig VAE, även om det finns utrymme för förbättring. Slutligen så kan de första två kapitlena av detta examensarbete också användas som en introduktion till Topologisk Data Analys, Allmän Topologi och Beräkningstopologi (eller Algoritmisk Topologi) till vilken maskininlärnings student, ingenjör eller forskare som är intresserad av dessa ämnesområden men saknar bakgrund i topologi.