A5045463433- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Urban Transport and Accessibility
- Machine Learning and Data Classification
- Health disparities and outcomes
- AI in cancer detection
- Digital Imaging for Blood Diseases
- Advanced Image and Video Retrieval Techniques
- COVID-19 epidemiological studies
- Adversarial Robustness in Machine Learning
- Urban Green Space and Health
- Advanced Electron Microscopy Techniques and Applications
- Cell Image Analysis Techniques
- COVID-19 diagnosis using AI
- Retinal Imaging and Analysis
- Medical Image Segmentation Techniques
- Air Quality and Health Impacts
- Electron and X-Ray Spectroscopy Techniques
- Generative Adversarial Networks and Image Synthesis
- Data-Driven Disease Surveillance
- Noise Effects and Management
- Image Processing Techniques and Applications
- Collaboration in agile enterprises
- Impact of Light on Environment and Health
- Image Retrieval and Classification Techniques
University of Utah
2016-2020
Effective convolutional neural networks are trained on large sets of labeled data. However, creating datasets is a very costly and time-consuming task. Semi-supervised learning uses unlabeled data to train model with higher accuracy when there limited set available. In this paper, we consider the problem semi-supervised networks. Techniques such as randomized augmentation, dropout random max-pooling provide better generalization stability for classifiers that using gradient descent. Multiple...
Many biomedical image analysis applications require segmentation. Convolutional neural networks (CNN) have become a promising approach to segment images; however, the accuracy of these methods is highly dependent on training data. We focus segmentation in context where there variation between source and target datasets ground truth for dataset very limited or non-existent. use an adversarial based train CNNs achieve good domain. DRIVE STARE eye vasculture show that our can significantly...
The spread of COVID-19 is not evenly distributed. Neighborhood environments may structure risks and resources that produce disparities. built allow greater flow people into an area or impede social distancing practices increase residents’ risk for contracting the virus. We leveraged Google Street View (GSV) images computer vision to detect environment features (presence a crosswalk, non-single family home, single-lane roads, dilapidated building visible wires). utilized Poisson regression...
The built environment is a structural determinant of health and has been shown to influence expenditures, behaviors, outcomes. Traditional methods assessing characteristics are time-consuming difficult combine or compare. Google Street View (GSV) images represent large, publicly available data source that can be used create indicators the physical with machine learning techniques. aim this study use GSV measure association features health-related behaviors outcomes at census tract level.We...
In this paper we consider the problem of semi-supervised learning with deep Convolutional Neural Networks (ConvNets). Semi-supervised is motivated on observation that unlabeled data cheap and can be used to improve accuracy classifiers. propose an unsupervised regularization term explicitly forces classifier's prediction for multiple classes mutually-exclusive effectively guides decision boundary lie low density space between manifolds corresponding different data. Our proposed approach...
Objectives Built environments can affect health, but data in many geographic areas are limited. We used a big source to create national indicators of neighborhood quality and assess their associations with health. Methods leveraged computer vision Google Street View images accessed from December 15, 2017, through July 17, 2018, detect features the built environment (presence crosswalk, non–single-family home, single-lane roads, visible utility wires) for 2916 US counties. multivariate linear...
Previous studies have demonstrated that there is a high possibility the presence of certain built environment characteristics can influence health outcomes, especially those related to obesity and physical activity. We examined associations between select neighborhood indicators (crosswalks, non-single family home buildings, single-lane roads, visible wires), including obesity, diabetes, cardiovascular disease, premature mortality, at state level. utilized 31,247,167 images collected from...
We introduce a novel unsupervised loss function for learning semantic segmentation with deep convolutional neural nets (ConvNet) when densely labeled training images are not available. More specifically, the proposed penalizes L1-norm of gradient label probability vector image , i.e. total variation, produced by ConvNet. This can be seen as regularization term that promotes piecewise smoothness ConvNet during learning. The is combined supervised in semi-supervised setting to learn ConvNets...
While convolutional neural networks (CNN) produce state-of-the-art results in many applications including biomedical image analysis, they are not robust to variability the data that is well represented by training set. An important source of images appearance objects such as contrast and texture due different imaging settings. We introduce neighborhood similarity layer (NSL) which can be used a CNN improve robustness changes data. The proposed NSL transforms its input feature map at given...
Deep learning and, specifically, convoltional neural networks (CNN) represent a class of powerful models that facilitate the understanding many problems in computer vision. When combined with reasonable amount data, CNNs can outperform traditional for tasks, including image classification. In this work, we utilize these tools imagery data collected through Google Street View images to perform virtual audits neighborhood characteristics. We further investigate different architectures chronic...
Segmenting images with low-quality, low signal to noise ratio has been a challenging task in computer vision. It shown that statistical prior information about the shape of object be segmented can used significantly mitigate this problem. However estimating probability densities shapes space difficult. This problem becomes more difficult when there is limited amount training data or testing contain missing data. Most model based segmentation approaches tend minimize an energy functional...
In this paper we consider the problem of semi-supervised learning with deep Convolutional Neural Networks (ConvNets). Semi-supervised is motivated on observation that unlabeled data cheap and can be used to improve accuracy classifiers. propose an unsupervised regularization term explicitly forces classifier's prediction for multiple classes mutually-exclusive effectively guides decision boundary lie low density space between manifolds corresponding different data. Our proposed approach...
We present a neighborhood similarity layer (NSL) which induces appearance invariance in network when used conjunction with convolutional layers. are motivated by the observation that, even though networks have low generalization error, their capability does not extend to samples represented training data. For instance, while novel appearances of learned concepts pose no problem for human visual system, feedforward generally successful such situations. Motivated Gestalt principle grouping...