Understanding how users interact with a company's service is essential for data-driven businesses that want to better cater to their users and improve their offering. By using a generative machine learning approach it is possible to model user behaviour and generate new data to simulate or recognize and explain typical usage patterns. In this work we introduce an approach for modelling users' interaction behaviour at scale in a client-service model. We propose a novel representation of multivariate time-series data as time pictures that express temporal correlations through spatial organization. This representation shares two key properties that convolutional networks have been built to exploit and allows us to develop an approach based on deep generative models that use convolutional networks as backbone. In introducing this approach of feature learning for time-series data, we expand the application of convolutional neural networks in the multivariate time-series domain, and specifically user interaction data. We adopt a variational approach inspired by the β-VAE framework in order to learn hidden factors that define different user behaviour patterns. We explore different values for the regularization parameter β and show that it is possible to construct a model that learns a latent representation of identifiable and different user behaviours. We show on real-world data that the model generates realistic samples, that capture the true population-level statistics of the interaction behaviour data, learns different user behaviours, and provides accurate imputations of missing data. / Förståelse för hur användare interagerar med ett företags tjänst är essentiell för data-drivna affärsverksamheter med ambitioner om att bättre tillgodose dess användare och att förbättra deras utbud. Generativ maskininlärning möjliggör modellering av användarbeteende och genererande av ny data i syfte att simulera eller identifiera och förklara typiska användarmönster. I detta arbete introducerar vi ett tillvägagångssätt för storskalig modellering av användarinteraktion i en klientservice-modell. Vi föreslår en ny representation av multivariat tidsseriedata i form av tidsbilder vilka representerar temporala korrelationer via spatial organisering. Denna representation delar två nyckelegenskaper som faltningsnätverk har utvecklats för att exploatera, vilket tillåter oss att utveckla ett tillvägagångssätt baserat på på djupa generativa modeller som bygger på faltningsnätverk. Genom att introducera detta tillvägagångssätt för tidsseriedata expanderar vi applicering av faltningsnätverk inom domänen för multivariat tidsserie, specifikt för användarinteraktionsdata. Vi använder ett tillvägagångssätt inspirerat av ramverket β-VAE i syfte att lära modellen gömda faktorer som definierar olika användarmönster. Vi utforskar olika värden för regulariseringsparametern β och visar att det är möjligt att konstruera en modell som lär sig en latent representation av identifierbara och multipla användarbeteenden. Vi visar med verklig data att modellen genererar realistiska exempel vilka i sin tur fångar statistiken på populationsnivå hos användarinteraktionsdatan, samt lär olika användarbeteenden och bidrar med precisa imputationer av saknad data.

Many diseases of genetic origin originate from non-synonymous single nucleotide polymorphisms (nsSNPs). These cause changes in the final protein product encoded by a gene. Through large scale sequencing and population studies, there is growing availability of information of which variations are tolerated and which are not. Variant effect predictors use a wide range of information about such variations to predict their effect, often focusing on evolutionary information. Here, a novel amino acid substitution variant effect predictor is developed. The predictor is a deep convolutional neural network incorporating evolutionary information, sequence information, as well as structural information, to predict both the pathogenicity as well as the severity of amino acid substitutions. The model achieves state-of-the-art performance on benchmark datasets.

Sand casting is used to manufacture large metal workpieces. The processing is executed by pouring molten metal into a sand mold. During the process, the mold is subjected to mechanical and thermal stress. It is of economic interest to inspect the molds for defects that can affect casting results, in the worst case leading to discarded products. This thesis investigates and proposes an automated solution for inspecting surface cracks in sand molds. A hybrid solution using image processing and convolutional neural networks has been developed. The first is to find and implement a crack detection method that can perform equally well or better than a human. The second objective is to investigate the amount of training data needed. Twenty-one machine learning models have been trained to evaluate the impact training data size along with transfer learning, fine-tuning, data augmentation, and image processing have on performance. As a result, it was found that the image processing part of the method is not effective in finding cracks in its current form. However, the convolutional neural network still achieves good performance. The method has been trained and tested on sand mold core images captured with a test workbench along with images of concrete walls and pavement acquired from the SDNET2018 data set. Sand mold images achieve 82% accuracy and 79% recall when training on 90 images while testing on 28 images separate from training. A maximal performance of 97.9% accuracy and 99.7% recall is achieved when training on 5400 SDNET2018 images and then testing on 608 images. When training on 100 SDNET2018 images and tested on the same 608 images, a performance of 86.0% and 96.7% recall is achieved. It is concluded that the proposed solution is feasible. Transfer learning and data augmentation are essential techniques to achieve good performance if a small amount of data is available, while fine-tuning may give a slight performance boost. Further work should be performed considering the impact of curved geometry on performance. Investigating alternative structures of the convolutional neural network and testing alternative hyperparameters may improve generalization performance. The image processing performance may be improved if the manufacturing process is more precisely defined, as parameters can be more optimally tuned.

The goal of this work is to improve the robustness and generalization of deep learning models, using a similar approach to the unsupervised "innate learning" strategy in visual development. A series of research studies are presented to demonstrate how an unsupervised machine learning efficient coding approach can create filters similar to the receptive fields of the primary visual cortex (V1) in the brain, and these filters are capable of pretraining convolutional neural networks (CNNs) to enable faster training times and higher accuracy with less dependency on the source data. Independent component analysis (ICA) is used for unsupervised feature extraction as it has shown success in both applied machine learning and modeling biological neural receptive fields. This pretraining applies equally well to various forms of visual input, including natural color images, black and white, binocular, and video to drive the V1-like Gabor filters in the brain. For efficient processing of typical visual scenes, the filters that ICA produces are developed by encoding natural images. These filters are then used to initialize the kernels in the first layer of a CNN to train on the CIFAR-10 dataset to perform image classification. Results show that the ICA initialization for a custom made CNN produces models with a test accuracy up to 12% better than the standard model in the first 10 epochs, which for specific accuracy thresholds reduces the number of training epochs by approximately 40% (to reach 60% accuracy) and 50% (to reach 70% accuracy). Additionally, this pre-training results in marginally higher accuracy even after extensive training over 50 epochs. This proposed method of unsupervised machine learning to pre-train weights in deep learning improves both training time and accuracy, which is why it is observed in biological networks and is finding increased application in applied deep learning.

Yes / The quantification of corneal endothelial cell (CEC) morphology using manual and semi-automatic software enables an objective assessment of corneal endothelial pathology. However, the procedure is tedious, subjective, and not widely applied in clinical practice. We have developed the CellsDeepNet system to automatically segment and analyse the CEC morphology. The CellsDeepNet system uses Contrast-Limited Adaptive Histogram Equalization (CLAHE) to improve the contrast of the CEC images and reduce the effects of non-uniform image illumination, 2D Double-Density Dual-Tree Complex Wavelet Transform (2DDD-TCWT) to reduce noise, Butterworth Bandpass filter to enhance the CEC edges, and moving average filter to adjust for brightness level. An improved version of U-Net was used to detect the boundaries of the CECs, regardless of the CEC size. CEC morphology was measured as mean cell density (MCD, cell/mm2), mean cell area (MCA, µm2), mean cell perimeter (MCP, µm), polymegathism (coefficient of CEC size variation), and pleomorphism (percentage of hexagonality coefficient). The CellsDeepNet system correlated highly significantly with the manual estimations for MCD (r = 0.94), MCA (r = 0.99), MCP (r = 0.99), polymegathism (r = 0.92), and pleomorphism (r = 0.86), with p

Mammography screenings are performed regularly on women in order to detect early signs of breast cancer, which is the most common form of cancer. During an exam, X-ray images (called mammograms) are taken from two different angles and reviewed by a radiologist. If they find a suspicious lesion in one of the views, they confirm it by finding the corresponding region in the other view. Finding the corresponding region is a non-trivial task, due to the different image projections of the breast and different angles of compression needed during the exam. This thesis explores the possibility of using deep learning, a data-driven approach, to solve the corresponding regions problem. Specifically, a convolutional neural network (CNN) called U-net is developed and trained on scanned mammograms, and evaluated on both scanned and digital mammograms. A model based method called the arc model is developed for comparison. Results show that the best U-net produced better results than the arc model on all evaluated metrics, and succeeded in finding the corresponding area 83.9% of times, compared to 72.6%. Generalization to digital images was excellent, achieving an even higher score of 87.6%, compared to 83.5% for the arc model.

Identifying Low-head dams (LHD) and creating an inventory become a priority as fatalities continue to occur at these structures. Because obstruction inventories do not specifically identify LHDs, and they are not assigned a hazard classification, there is not an official inventory of LHD. However, there is a multi-agency taskforce that is creating an inventory of LHD. All efforts have been performed by manually identifying LHD on Google Earth Pro (GE Pro). The purpose of this paper is to assess whether a machine learning approach can accelerate the national inventory. We used a machine learning approach to implement a high-resolution remote sensing data and a Convolutional Neural Network (CNN) architecture. The model achieved 76% accuracy on identifying LHD (true positive) and 95% accuracy identifying NLHD (true negative) on the validation set. We deployed the trained model into the National Hydrologic Geospatial Fabric (Hydrofabric) flowlines on the Provo River watershed. The results showed a high number of false positives and low accuracy in identifying LHD due to the mismatch between Hydrofabric flowlines and actual waterways. We recommend improving the accuracy of the Hydrofabric waterway tracing algorithms to increase the percentage of correctly classified LHD.

[pt] O petróleo e gás são importantes na economia mundial, utilizados como matéria-prima em vários produtos. Para a extração desses produtos é necessário realizar a caracterização dos reservatórios de hidrocarbonetos. A partir dessa caracterização são extraídos um volume com dados sísmicos da região de interesse. Esses dados são interpretados para identificação de várias características, como a classificação de fácies sísmicas, horizontes, falhas, e gás. A grande quantidade de dados do volume de sísmica, torna a interpretação manual cada vez mais desafiadora. Muitos pesquisadores da área de interpretação sísmica tem investido em métodos utilizando redes neurais. As redes neurais convolucionais (CNN) são muito utilizadas em problemas de visão computacional, e obtém ótimos resultados em muitos problemas com dados 2D. O presente trabalho tem como objetivo a aplicação de redes neurais convolucionais no mapeamento supervisionado de horizontes sísmicos. Avaliamos nossa proposta usando o bloco F3 com as anotações de fácies sísmicas. Os dados foram utilizados baseados em modelo de seção e patches. Na previsão de horizonte foram avaliadas as arquiteturas da ResUnet e DC-Unet. Como função de perda foram analisadas a Generalized Dice e a perda Focal Tversky. O método mostrou resultados promissores com a ResUnet e função de perda Focal Tversky, nos dados baseados em patches de 128x128, alcançando aproximadamente 56 por cento na métrica Dice. A implementação completa e as redes treinadas estão disponíveis em https://github.com/mayaragomys/seismic_horizons. / [en] Oil and gas are important in the world economy, used as raw materials in various products. For the extraction of these products, it is necessary to carry out the characterization of the hydrocarbon reservoirs. This characterization extracts a volume with seismic data from the region of interest. These data are interpreted to identify various features, such as the classification of seismic facies, horizons, faults, and gas. A large amount of seismic volume data makes manual interpretation increasingly challenging. Many researchers in the field of seismic interpretation have invested in methods using neural networks. Convolutional Neural Networks (CNN) are widely used in computer vision problems and get great results in many situations with 2D data. The present work aimed to apply convolutional neural networks in the supervised mapping of seismic horizons. We evaluated our proposal using the F3 block with seismic facies annotations. The data representation in the input layer are patches of sections. In the horizon forecast, we evaluate the architectures of ResUnet and DC-Unet. We use the Generalized Dice and the Focal Tversky loss functions for the loss function. The method delivered promising results with the ResUnet and Focal Tversky loss function on data based on 128x128 patches, reaching approximately 56 percent on the Dice metric. The full implementation and the trained networks are available at https://github.com/mayaragomys/seismic_horizons.

No description available.

This dissertation consists of three essays in health economics. The first chapter, "The Built Environment and Obesity in Philadelphia: The Use of Satellite Imagery and Transfer Learning," investigates the relationship between the built environment and health outcomes, specifically obesity prevalence in Philadelphia. The built environment can affect obesity prevalence through the physical activity environment and the food environment. The main innovation of this paper is to use a pre-trained convolutional neural network (CNN) to extract data representing the features of the built environment from high-resolution satellite imagery. Because of the lack of information on the food environment in satellite images, I combined a proxy variable for food access together with the feature variables to represent the characteristics of the built environment. I then employed the Elastic Net model to test the relationship between the feature variables of the built environment and obesity prevalence in Philadelphia. The results show that the built environment is highly associated with obesity prevalence. This study also provides some evidence that the features of the built environment that have been extracted from satellite imagery can reduce the role of food access in estimating obesity, as well as that adding these features can explain more variance of obesity. The second chapter, "Paid Maternity Leave and Child Health: Evidence from Urban China," uses the China Health and Nutrition Survey data to study whether the extension of paid maternity leave affects children's health outcomes in urban China. This paper uses the time variation of the implementation of a maternity leave policy across different provinces from 1987 to 1991 in China to estimate a two-way fixed-effects model. The results suggest that the expansion of paid maternity leave has no impact on children's health in urban China. The last chapter, titled "The Association between Paid Maternity Leave and Mothers' Health and Labor Outcomes in Urban China," studies whether the extension of paid maternity leave in 1987-1991 would affect the labor and health outcomes of mothers in urban China by using the China Health and Nutrition Survey data. Based on the variation in the implementation time of a paid maternity leave policy across different provinces, this paper employs a two-way fixed-effects model to estimate the policy impact on mothers' health and labor outcomes in China. The findings indicate that extending the duration of paid maternity leave is associated with an increased likelihood of mothers remaining employed after childbirth. However, the study also reveals a negative relationship between the extension of paid maternity leave and mothers' wage rates. / Economics