July 19, 2018 In this thesis we present an investigation of multi-task and transfer learning using the recently introduced task of semantic tagging. First we employ a number of natural language processing tasks as auxiliaries for semantic tag- ging. Secondly, going in the other direction, we employ seman- tic tagging as an auxiliary task for three di erent NLP tasks: Part-of-Speech Tagging, Universal Dependency parsing, and Natural Language Inference. We compare full neural network sharing, partial neural network sharing, and what we term the learning what to share setting where neg- ative transfer between tasks is less likely. Fi- nally, we investigate multi-lingual learning framed as a special case of multi-task learning. Our ndings show considerable improvements for most experiments, demonstrating a variety of cases where multi-task and transfer learning methods are bene cial. 1 References 2

Dans cette dissertation, nous proposons des méthodes d’apprentissage automa-tique aptes à bénéficier de la récente explosion des volumes de données digitales.Premièrement nous considérons l’amélioration de l’efficacité des méthodes derécupération d’image. Nous proposons une approche d’apprentissage de métriques locales coordonnées (Coordinated Local Metric Learning, CLML) qui apprends des métriques locales de Mahalanobis, puis les intègre dans une représentation globale où la distance l2 peut être utilisée. Ceci permet de visualiser les données avec une unique représentation 2D, et l’utilisation de méthodes de récupération efficaces basées sur la distance l2. Notre approche peut être interprétée comme l’apprentissage d’une projection linéaire de descripteurs donnés par une méthode a noyaux de grande dimension définie explictement. Cette interprétation permet d’appliquer des outils existants pour l’apprentissage de métriques de Mahalanobis à l’apprentissage de métriques locales coordonnées. Nos expériences montrent que la CLML amé-liore les résultats en matière de récupération de visage obtenues par les approches classiques d’apprentissage de métriques locales et globales.Deuxièmement, nous présentons une approche exploitant les modèles de ré-seaux neuronaux convolutionnels (CNN) pour la reconnaissance faciale dans lespectre visible. L’objectif est l’amélioration de la reconnaissance faciale hétérogène, c’est à dire la reconnaissance faciale à partir d’images infra-rouges avec des images d’entraînement dans le spectre visible. Nous explorerons différentes stratégies d’apprentissage de métriques locales à partir des couches intermédiaires d’un CNN, afin de faire le rapprochement entre des images de sources différentes. Dans nos expériences, la profondeur de la couche optimale pour une tâche donnée est positivement corrélée avec le changement entre le domaine source (données d’entraînement du CNN) et le domaine cible. Les résultats montrent que nous pouvons utiliser des CNN entraînés sur des images du spectre visible pour obtenir des résultats meilleurs que l’état de l’art pour la reconnaissance faciale hétérogène (images et dessins quasi-infrarouges).Troisièmement, nous présentons les "tissus de neurones convolutionnels" (Convolutional Neural Fabrics) permettant l’exploration de l’espace discret et exponentiellement large des architectures possibles de réseaux neuronaux, de manière efficiente et systématique. Au lieu de chercher à sélectionner une seule architecture optimale, nous proposons d’utiliser un "tissu" d’architectures combinant un nombre exponentiel d’architectures en une seule. Le tissu est une représentation 3D connectant les sorties de CNNs à différentes couches, échelles et canaux avec un motif de connectivité locale, homogène et creux. Les seuls hyper-paramètres du tissu (le nombre de canaux et de couches) ne sont pas critiques pour la performance. La nature acyclique du tissu nous permet d’utiliser la rétro-propagation du gradient durant la phase d’apprentissage. De manière automatique, nous pouvons donc configurer le tissu de manière à implémenter l’ensemble de toutes les architectures possibles (un nombre exponentiel) et, plus généralement, des ensembles (combinaisons) de ces modèles. La complexité de calcul et de taille mémoire du tissu évoluent de manière linéaire alors qu’il permet d’exploiter un nombre exponentiel d’architectures en parallèle, en partageant les paramètres entre architectures. Nous présentons des résultats à l’état de l’art pour la classification d’images sur le jeu de données MNIST et CIFAR10, et pour la segmentation sémantique sur le jeu de données Part Labels. / In this dissertation, we propose methods and data driven machine learning solutions which address and benefit from the recent overwhelming growth of digital media content.First, we consider the problem of improving the efficiency of image retrieval. We propose a coordinated local metric learning (CLML) approach which learns local Mahalanobis metrics, and integrates them in a global representation where the l2 distance can be used. This allows for data visualization in a single view, and use of efficient ` 2 -based retrieval methods. Our approach can be interpreted as learning a linear projection on top of an explicit high-dimensional embedding of a kernel. This interpretation allows for the use of existing frameworks for Mahalanobis metric learning for learning local metrics in a coordinated manner. Our experiments show that CLML improves over previous global and local metric learning approaches for the task of face retrieval.Second, we present an approach to leverage the success of CNN models forvisible spectrum face recognition to improve heterogeneous face recognition, e.g., recognition of near-infrared images from visible spectrum training images. We explore different metric learning strategies over features from the intermediate layers of the networks, to reduce the discrepancies between the different modalities. In our experiments we found that the depth of the optimal features for a given modality, is positively correlated with the domain shift between the source domain (CNN training data) and the target domain. Experimental results show the that we can use CNNs trained on visible spectrum images to obtain results that improve over the state-of-the art for heterogeneous face recognition with near-infrared images and sketches.Third, we present convolutional neural fabrics for exploring the discrete andexponentially large CNN architecture space in an efficient and systematic manner. Instead of aiming to select a single optimal architecture, we propose a “fabric” that embeds an exponentially large number of architectures. The fabric consists of a 3D trellis that connects response maps at different layers, scales, and channels with a sparse homogeneous local connectivity pattern. The only hyperparameters of the fabric (the number of channels and layers) are not critical for performance. The acyclic nature of the fabric allows us to use backpropagation for learning. Learning can thus efficiently configure the fabric to implement each one of exponentially many architectures and, more generally, ensembles of all of them. While scaling linearly in terms of computation and memory requirements, the fabric leverages exponentially many chain-structured architectures in parallel by massively sharing weights between them. We present benchmark results competitive with the state of the art for image classification on MNIST and CIFAR10, and for semantic segmentation on the Part Labels dataset

[pt] A exploração segura de reservatórios de petróleo necessita uma boa modelagem numérica dos objetos geológicos da sub superfície, que inclui entre outras etapas: interpretação sísmica e geração de malha. Esta tese apresenta um estudo nessas duas áreas. O primeiro estudo é uma contribuição para interpretação de dados sísmicos, que se baseia na detecção automática de falhas sísmicas usando redes neurais profundas. Em particular, usamos Redes Neurais Convolucionais (RNCs) diretamente sobre mapas de amplitude sísmica, com a particularidade de usar dados sintéticos para treinar a rede com o objetivo final de classificar dados reais. Num segundo estudo, propomos um novo algoritmo para geração de malhas bidimensionais de quadrilaterais para estudos geomecânicos, baseado numa abordagem inovadora do método de quadtree: definimos novos padrões de subdivisão para adaptar a malha de maneira eficiente a qualquer geometria de entrada. As malhas obtidas podem ser usadas para simulações com o Método de Elementos Finitos (MEF). / [en] Safe oil exploration requires good numerical modeling of the subsurface geobodies, which includes among other steps: seismic interpretation and mesh generation. This thesis presents a study in these two areas. The first study is a contribution to data interpretation, examining the possibilities of automatic seismic fault detection using deep learning methods. In particular, we use Convolutional Neural Networks (CNNs) on seismic amplitude maps, with the particularity to use synthetic data for training with the goal to classify real data. In the second study, we propose a new two-dimensional all-quadrilateral meshing algorithm for geomechanical domains, based on an innovative quadtree approach: we define new subdivision patterns to efficiently adapt the mesh to any input geometry. The resulting mesh is suited for Finite Element Method (FEM) simulations.

Using Convolutional Neural Networks for classification of images and for localization and detection of objects in images is becoming increasingly popular. Within radiology a huge amount of image data is produced and meta data containing information of what the images depict is currently added manually by a radiologist. To aid in streamlining physician’s workflow this study has investigated the possibility to use Convolutional Neural Networks (CNNs) that are pre-trained on natural images to automatically detect the presence and location of multiple organs and body-parts in medical CT images. The results show promise for multiclass classification with an average precision 89.41% and average recall 86.40%. This also confirms that a CNN that is pre-trained on natural images can be succesfully transferred to solve a different task. It was also found that adding additional data to the dataset does not necessarily result in increased precision and recall or decreased error rate. It is rather the type of data and used preprocessing techniques that matter.

A Mineração de Opiniões é um processo que tem por objetivo extrair as opiniões e suas polaridades de sentimentos expressas em textos em língua natural. Essa área de pesquisa tem ganhado destaque devido ao volume de opiniões que os usuários compartilham na Internet, como revisões em sites de e-commerce, rede sociais e tweets. A Mineração de Opiniões baseada em Aspectos é uma alternativa promissora para analisar a polaridade do sentimento em um maior nível de detalhes. Os métodos tradicionais para extração de aspectos e classificação de sentimentos exigem a participação de especialistas de domínio para criar léxicos ou definir regras de extração para diferentes idiomas e domínios. Além disso, tais métodos usualmente exploram algoritmos de aprendizado supervisionado, porém exigem um grande conjunto de dados rotulados para induzir um modelo de classificação. Os desafios desta tese de doutorado estão relacionados a como diminuir a necessidade de grande esforço humano tanto para rotular dados, quanto para tratar a dependência de domínio para as tarefas de extração de aspectos e classificação de sentimentos dos aspectos para Mineração de Opiniões. Para reduzir a necessidade de grande quantidade de exemplos rotulados foi proposta uma abordagem semissupervisionada, denominada por Aspect-based Sentiment Propagation on Heterogeneous Networks (ASPHN) em que são propostas representações de textos nas quais os atributos linguísticos, os aspectos candidatos e os rótulos de sentimentos são modelados por meio de redes heterogêneas. Para redução dos esforços para construir recursos específicos de domínio foi proposta uma abordagem baseada em aprendizado por transferência entre domínios denominada Cross-Domain Aspect Label Propagation through Heterogeneous Networks (CD-ALPHN) que utiliza dados rotulados de outros domínios para suportar tarefas de aprendizado em domínios sem dados rotulados. Nessa abordagem são propostos uma representação em uma rede heterogênea e um método de propagação de rótulos. Os vértices da rede são os aspectos rotulados do domínio de origem, os atributos linguísticos e os candidatos a aspectos do domínio alvo. Além disso, foram analisados métodos de extração de aspectos e propostas algumas variações para considerar cenários nãosupervisionados e independentes de domínio. As soluções propostas nesta tese de doutorado foram avaliadas e comparadas as do estado-da-arte utilizando coleções de revisões de diferentes produtos e serviços. Os resultados obtidos nas avaliações experimentais são competitivos e demonstram que as soluções propostas são promissoras. / Opinion Mining is a process that aims to extract opinions and their sentiment polarities expressed in natural language texts. This area of research has been in the highlight because of the volume of opinions that users share on the available visualization means on the Internet (reviews on e-commerce sites, social networks, tweets, others). Aspect-based Opinion Mining is a promising alternative for analyzing the sentiment polarity on a high level of detail. The traditional methods for aspect extraction and sentiment classification require the participation of domain experts to create lexicons or define extraction rules for different languages and domains. In addition, such methods usually exploit supervised machine learning algorithms, but require a large set of labeled data to induce a classification model. The challenges of this doctoral thesis are related on to how to reduce the need for great human effort both: (i) to label data; and (ii) to treat domain dependency for the tasks of aspect extraction and aspect sentiment classification for Opinion Mining. In order to reduce the need for a large number of labeled examples, a semi-supervised approach was proposed, called Aspect-based Sentiment Propagation on Heterogeneous Networks (ASPHN). In this approach, text representations are proposed in which linguistic attributes, candidate aspects and sentiment labels are modeled by heterogeneous networks. Also, a cross-domain learning approach called Cross-Domain Aspect Label Propagation through Heterogeneous Networks (CD-ALPHN) is proposed in order to reduce efforts to build domain-specific resources, This approach uses labeled data from other domains to support learning tasks in domains without labeled data. A representation in a heterogeneous network and a label propagation method are proposed in this cross-domain learning approach. The vertices of the network are the labeled aspects of the source domain, the linguistic attributes, and the candidate aspects of the target domain. In addition, aspect extraction methods were analyzed and some variations were proposed to consider unsupervised and domain independent scenarios. The solutions proposed in this doctoral thesis were evaluated and compared to the state-of-the-art solutions using collections of different product and service reviews. The results obtained in the experimental evaluations are competitive and demonstrate that the proposed solutions are promising.

Deep learning has been rapidly growing in recent years obtaining excellent results for many computer vision applications, such as image classification and object detection. One aspect for the increased popularity of deep learning is that it mitigates the need for hand-crafted features. This thesis work investigates deep learning as a methodology to solve the problem of autonomous collision avoidance for a small robotic car. To accomplish this, transfer learning is used with the VGG16 deep network pre-trained on ImageNet dataset. A dataset has been collected and then used to fine-tune and validate the network offline. The deep network has been used with the robotic car in a real-time manner. The robotic car sends images to an external computer, which is used for running the network. The predictions from the network is sent back to the robotic car which takes actions based on those predictions. The results show that deep learning has great potential in solving the collision avoidance problem.

The 2019 novel coronavirus has been proven to present several unique features on chest X-rays and CT-scans that distinguish it from imaging of other pulmonary diseases such as bacterial pneumonia and viral pneumonia unrelated to COVID-19. However, the key characteristics of a COVID-19 infection have been proven challenging to detect with the human eye. The aim of this project is to explore if it is possible to distinguish a patient with COVID-19 from a patient who is not suffering from the disease from posteroanterior chest X-ray images using synthetic minority over-sampling and transfer learning. Furthermore, the report will also present the mechanics of COVID-19, the used dataset and models and the validity of the results.

Monte Carlo methods in derivative pricing are computationally expensive, in particular for evaluating models partial derivatives with regard to inputs. This research proposes the use of deep learning to approximate such valuation models for highly exotic derivatives, using automatic differentiation to evaluate input sensitivities. Deep learning models are trained to approximate Phoenix Autocall valuation using a proprietary model used by Svenska Handelsbanken AB. Models are trained on large datasets of low-accuracy (10^4 simulations) Monte Carlo data, successfully learning the true model with an average error of 0.1% on validation data generated by 10^8 simulations. A specific model parametrisation is proposed for 2-day valuation only, to be recalibrated interday using transfer learning. Automatic differentiation approximates sensitivity to (normalised) underlying asset prices with a mean relative error generally below 1.6%. Overall error when predicting sensitivity to implied volatililty is found to lie within 10%-40%. Near identical results are found by finite difference as automatic differentiation in both cases. Automatic differentiation is not successful at capturing sensitivity to interday contract change in value, though errors of 8%-25% are achieved by finite difference. Model recalibration by transfer learning proves to converge over 15 times faster and with up to 14% lower relative error than training using random initialisation. The results show that deep learning models can efficiently learn Monte Carlo valuation, and that these can be quickly recalibrated by transfer learning. The deep learning model gradient computed by automatic differentiation proves a good approximation of the true model sensitivities. Future research proposals include studying optimised recalibration schedules, using training data generated by single Monte Carlo price paths, and studying additional parameters and contracts.

In the scheme of Non Destructive Testing (NDT), defect detection is an important process. Traditional image processing techniques have successfully been used for defect recognition. Usage of machine learning techniques is still in the initial stages of development. Convolution Neural Networks (CNN) is widely used for object classification one such scenario is defect classification in weld tubes. With the advent of deep learning techniques such as transfer learning, we can transfer knowledge gained in one domain successfully into other. Pre-trained models successfully learn features from large scale datasets that can be used for in domains having sparse data and smaller datasets. The aim of this work is to help a manual inspector in recognition of defects on the weld tubes. With a given set of images, we proceed by forming unique pipeline architecture for automatic defect recognition. The research in this thesis focuses on extraction of welds using image segmentation techniques, creating a dataset of defects and using it to on pre-trained Convolution Neural Networks of VGG16, VGG19, Inception V3 and ResNet101. We evaluate the models on different metrics finding the best suited model for the created dataset. Further a prototype sliding window solution is used to find defects over the extracted weld region. We also present the limitations of this approach and suggest modifications that can be implemented in the future.

<div>Supervised classification methods based on deep learning have achieved great success in many domains and tasks that are previously unimaginable. Such approaches build on learning paradigms that require hundreds of examples in order to learn to classify objects or events. Thus, their immediate application to the domains with few or no observations is limited. This is because of the lack of ability to rapidly generalize to new categories from a few examples or from high-level descriptions of categories. This can be attributed to the significant gap between the way machines represent knowledge and the way humans represent categories in their minds and learn to recognize them. In this context, this research represents categories as semantic trees in a high-level attribute space and proposes an approach to utilize these representations to conduct N-Shot, Few-Shot, One-Shot, and Zero-Shot Learning (ZSL). This work refers to this paradigm as the problem of general classification (GCP) and proposes a unified framework for GCP referred to as the Few and Zero-Shot Technique (FAZT). FAZT framework is an end-to-end approach that uses trainable 3D convolutional neural networks and recurrent neural networks to simultaneously optimize for both the semantic and the classification tasks. Lastly, the problem of systematically obtaining semantic attributes by utilizing domain-specific ontologies is presented. The proposed framework is validated in the domains of hand gesture and action/activity recognition, however, this research can be applied to other domains such as video understanding, the study of human behavior, emotion recognition, etc. First, an attribute-based dataset for gestures is developed in a systematic manner by relying on literature in gestures and semantics, and crowdsourced platforms such as Amazon Mechanical Turk. To the best of our knowledge, this is the first ZSL dataset for hand gestures (ZSGL dataset). Next, our framework is evaluated in two experimental conditions: 1. Within-category (to test the attribute recognition power) and 2. Across-category (to test the ability to recognize an unknown category). In addition, we conducted experiments in zero-shot, one-shot, few-shot and continuous learning conditions in both open-set and closed-set scenarios. Results showed that our framework performs favorably on the ZSGL, Kinetics, UIUC Action, UCF101 and HMDB51 action datasets in all the experimental conditions.<br></div><div><br></div>