Digital Commons Network^™

Private And Federated Deep Learning: System, Theory, And Applications For Social Good, Han Hu

Dissertations

During the past decade, drug abuse continues to accelerate towards becoming the most severe public health problem in the United States. The ability to detect drug­abuse risk behavior at a population scale, such as among the population of Twitter users, can help to monitor the trend of drug­abuse incidents. However, traditional methods do not effectively detect drug­abuse risk behavior in tweets, mainly due to the sparsity of such tweets and the noisy nature of tweets. In the first part of this dissertation work, the task of classifying tweets as containing drug­abuse risk behavior or not, is studied. Millions of public …

Electro-Chemo-Mechanics Of The Interfaces In 2d-3d Heterostructure Electrodes, Vidushi Sharma

Dissertations

Unique heterostructure electrodes comprising two-dimensional (2D) materials and bulk three dimensional (3D) high-performance active electrodes are recently synthesized and experimentally tested for their electrochemical performance in metal-ion batteries. Such electrodes exhibit long cycle life while they also retain high-capacity inherent to the active electrode. The role of 2D material is to provide a supportive mesh that allows buffer space for volume expansions upon ion intercalation in the active material and establishes a continuous electronic contact. Therefore, the binding strength between both materials is crucial for the success of such electrodes. Furthermore, battery cycles may bring about phase transformations in the …

Machine Learning And Computer Vision In Solar Physics, Haodi Jiang

Dissertations

In the recent decades, the difficult task of understanding and predicting violent solar eruptions and their terrestrial impacts has become a strategic national priority, as it affects the life of human beings, including communication, transportation, the power grid, national defense, space travel, and more. This dissertation explores new machine learning and computer vision techniques to tackle this difficult task. Specifically, the dissertation addresses four interrelated problems in solar physics: magnetic flux tracking, fibril tracing, Stokes inversion and vector magnetogram generation.

First, the dissertation presents a new deep learning method, named SolarUnet, to identify and track solar magnetic flux elements in …

Seizure Detection Using Deep Learning, Information Theoretic Measures And Factor Graphs, Bahareh Salafian

Electronic Thesis and Dissertation Repository

Epilepsy is a common neurological disorder that disrupts normal electrical activity in the brain causing severe impact on patients’ daily lives. Accurate seizure detection based on long-term time-series electroencephalogram (EEG) signals has gained vital importance for epileptic seizure diagnosis. However, visual analysis of these recordings is a time-consuming task for neurologists. Therefore, the purpose of this thesis is to propose an automatic hybrid model-based /data-driven algorithm that exploits inter-channel and temporal correlations. Hence, we use mutual information (MI) estimator to compute correlation between EEG channels as spatial features and employ a carefully designed 1D convolutional neural network (CNN) to extract …

Photoacoustic Imaging Of Colorectal Cancer And Ovarian Cancer, Xiandong Leng

McKelvey School of Engineering Theses & Dissertations

Photoacoustic (PA) imaging is an emerging hybrid imaging technology that uses a short-pulsed laser to excite tissue. The resulting photoacoustic waves are used to image the optical absorption distribution of the tissue, which is directly related to micro-vessel networks and thus to tumor angiogenesis, a key process in tumor growth and metastasis. In this thesis, the acoustic-resolution photoacoustic microscopy (AR-PAM) was first investigated on its role in human colorectal tissue imaging, and the optical-resolution photoacoustic microscopy (OR-PAM) was investigated on its role in human ovarian tissue imaging.Colorectal cancer is the second leading cause of cancer death in the United States. …

Deep Learning For Automatic Microscopy Image Analysis, Shenghua He

McKelvey School of Engineering Theses & Dissertations

Microscopy imaging techniques allow for the creation of detailed images of cells (or nuclei) and have been widely employed for cell studies in biological research and disease diagnosis in clinic practices.Microscopy image analysis (MIA), with tasks of cell detection, cell classification, and cell counting, etc., can assist with the quantitative analysis of cells and provide useful information for a cellular-level understanding of biological activities and pathology. Manual MIA is tedious, time-consuming, prone to subject errors, and are not feasible for the high-throughput cell analysis process. Thus, automatic MIA methods can facilitate all kinds of biological studies and clinical tasks. Conventional …

Deep Parameter Selection For Classic Computer Vision Applications, Michael Whitney

Theses and Dissertations

A trend in computer vision today is to retire older, so-called "classic'' methods in favor of ones based on deep neural networks. This has led to tremendous improvements in many areas, but for some problems deep neural solutions may not yet exist or be of practical application. For this and other reasons, classic methods are still widely used in a variety of applications. This paper explores the possibility of using deep neural networks to improve these older methods instead of replace them. In particular, it addresses the issue of parameter selection in these algorithms by using a neural network to …

Uncertainty-Aware Deep Learning For Prediction Of Remaining Useful Life Of Mechanical Systems, Samuel J. Cornelius

Theses and Dissertations

Remaining useful life (RUL) prediction is a problem that researchers in the prognostics and health management (PHM) community have been studying for decades. Both physics-based and data-driven methods have been investigated, and in recent years, deep learning has gained significant attention. When sufficiently large and diverse datasets are available, deep neural networks can achieve state-of-the-art performance in RUL prediction for a variety of systems. However, for end users to trust the results of these models, especially as they are integrated into safety-critical systems, RUL prediction uncertainty must be captured. This work explores an approach for estimating both epistemic and heteroscedastic …

Using A Systemic Skills Model To Build An Effective 21st Century Workforce: Factors That Impact The Ability To Navigate Complex Systems, Morteza Nagahi

Theses and Dissertations

The growth of technology and the proliferation of information made modern complex systems more fragile and vulnerable. As a result, competitive advantage is no longer achieved exclusively through strategic planning but by developing an influential cadre of technical people who can efficiently manage and navigate modern complex systems. The dissertation aims to provide educators, practitioners, and organizations with a model that helps to measure individuals’ systems thinking skills, complex problem solving, personality traits, and the impacting demographic factors such as managerial and work experience, current occupation type, organizational ownership structure, and education level. The intent is to study how these …

A Data-Driven Approach For The Investigation Of Microstructural Effects On The Effective Piezoelectric Responses Of Additively Manufactured Triply Periodic Bi-Continuous Piezocomposite, Wenhua Yang

Theses and Dissertations

A two-scale model consisting of ceramic grain scale and composite scale are developed to systematically evaluate the effects of microstructures (e.g., residual pores, grain size, texture) and geometry on the piezoelectric responses of the polarized triply periodic bi-continuous (TPC) piezocomposites. These TPC piezocomposites were fabricated by a recently developed additive manufacturing (AM) process named suspension-enclosing projection-stereolithography (SEPS) under different process conditions. In the model, the Fourier spectral iterative perturbation method (FSIPM) and the finite element method will be adopted for the calculation at the grain and composite scale, respectively. On the grain scale, a DL approach based on stacked generative …

Pranayama Breathing Detection With Deep Learning, Bikash Shrestha

Theses

Yoga, a complementary health approach, according to a 2017 National Health Interview Survey by the Center for Disease Control and Prevention (CDC), is a choice of around 14.3% adults in the US. Kapalbhati pranayama, a yoga practice of alternating fast exhales and longer passive inhales, is understood to improve our health. Incorrect and irregular practices, however, can cause injuries and adverse effects. To avoid these undesired effects, it is essential to maintain a pace fit for the practitioner. In the absence of any tools to observe a pace of practice, this work develops a deep learning method that listens to …

Developing Deep-Learning Methods For Diagnosis And Prognosis Of Pediatric Progressive Diseases Using Modern Imaging Techniques, Mahdieh Shabanian

Theses and Dissertations (ETD)

Purpose and Rationale. Central nervous system manifestations form a significant burden of disease in young children. There have been efforts to correlate the neurological disease state in tuberous sclerosis complex (TSC) neurological disease state with imaging findings is a standard part of patient care. However, such analysis of neuroimaging is time- and labor-intensive. Automated approaches to these tasks are needed to improve speed, accuracy, and availability. Automated medical image analysis tools based on 3D/2D deep learning algorithms can help improve the quality and consistency of image diagnosis and interpretation for cognitive disorders in infants. We propose to automate neuroimaging analysis …

Auto-Curation Of Large Evolving Image Datasets, Sara Mousavicheshmehkaboodi

Doctoral Dissertations

Large image collections are becoming common in many fields and offer tantalizing opportunities to transform how research, work, and education are conducted if the information and associated insights could be extracted from them. However, major obstacles to this vision exist. First, image datasets with associated metadata contain errors and need to be cleaned and organized to be easily explored and utilized. Second, such collections typically lack the necessary context or may have missing attributes that need to be recovered. Third, such datasets are domain-specific and require human expert involvement to make the right interpretation of the image content. Fourth, the …

Qualitative And Quantitative Improvements For Positron Emission Tomography Using Different Motion Correction Methodologies, Tasmia Rahman Tumpa

Doctoral Dissertations

Positron Emission Tomography (PET) data suffers from low image quality and quantitative accuracy due to different kinds of motion of patients during imaging. Hardware-based motion correction is currently the standard; however, is limited by several constraints, the most important of which is retroactive data correction. Data-driven techniques to perform motion correction in this regard are active areas of research. The motivation behind this work lies in developing a complete data-driven approach to address both motion detection and correction. The work first presents an algorithm based on the positron emission particle tracking (PEPT) technique and makes use of time-of-flight (TOF) information …

From Language Comprehension Towards General Ai, Binay Dahal

UNLV Theses, Dissertations, Professional Papers, and Capstones

Language comprehension or more formally, natural language understanding is one of the major undertakings in Artificial Intelligence. In this work, we explore a few of the problems in language understanding using fixed deep learning models. Specifically, first, we look into question generation. Asking questions relates to the cognitive ability of language comprehension and context understanding. For that reason, making progress in question generation is significant. We introduce a novel task called “question generation with masked target answer” and propose various models and present the baseline result for the task. Next, we extend on the question generation task and develop a …

Deep Learning-Guided Prediction Of Material’S Microstructures And Applications To Advanced Manufacturing, Jianan Tang

All Dissertations

Material microstructure prediction based on processing conditions is very useful in advanced manufacturing. Trial-and-error experiments are very time-consuming to exhaust numerous combinations of processing parameters and characterize the resulting microstructures. To accelerate process development and optimization, researchers have explored microstructure prediction methods, including physical-based modeling and feature-based machine learning. Nevertheless, they both have limitations. Physical-based modeling consumes too much computational power. And in feature-based machine learning, low-dimensional microstructural features are manually extracted to represent high-dimensional microstructures, which leads to information loss.

In this dissertation, a deep learning-guided microstructure prediction framework is established. It uses a conditional generative adversarial network (CGAN) …

Joint Linear And Nonlinear Computation With Data Encryption For Efficient Privacy-Preserving Deep Learning, Qiao Zhang

Electrical & Computer Engineering Theses & Dissertations

Deep Learning (DL) has shown unrivalled performance in many applications such as image classification, speech recognition, anomalous detection, and business analytics. While end users and enterprises own enormous data, DL talents and computing power are mostly gathered in technology giants having cloud servers. Thus, data owners, i.e., the clients, are motivated to outsource their data, along with computationally-intensive tasks, to the server in order to leverage the server’s abundant computation resources and DL talents for developing cost-effective DL solutions. However, trust is required between the server and the client to finish the computation tasks (e.g., conducting inference for the newly-input …

Survey And Analysis Of Deepfake Media As It Applies To The New Era Of Disinformation, Jade L. Wise

Theses

The world has become a place where technology and the internet is so integrated into every-day life that people rely on the media and internet access to stay informed and connected. But, that has exposed us to a whole new world of positives and negatives, and the negatives could be catastrophic for our society. It will very soon be very difficult for the average person to distinguish between figment and reality; primarily with video, which has always been seen as a credible form of evidence. Deepfake videos are a prime example of this growing phenomenon. They are videos in which, …

A Deep Learning Classifier For Detecting Atrial Fibrillation In Hospital Settings Applicable To Various Sensing Modalities, Kamil Bukum

Theses

Cardiac signals provide variety of information related to the patient's health. One of the most important is for medical experts to diagnose the functionality of a patient’s heart. This information helps the medical experts monitor heart disease such as atrial fibrillation and heart failure. Atrial fibrillation (AF) is one of the most major diseases that are threatening patients’ health. Medical experts measure cardiac signals usng the Electrocardiogram (ECG or EKG), the Photoplethysmogram (PPG), and more recently the Videoplethysmogram (VPG). Then they can use these measurements to analyze the heart functionality to detect heart diseases. In this study, these three major …

Heat Patrl: Network-Agnostic Cyber Attack Campaign Triage With Pseudo-Active Transfer Learning, Stephen Frank Moskal

Theses

SOC (Security Operation Center) analysts historically struggled to keep up with the growing sophistication and daily prevalence of cyber attackers. To aid in the detection of cyber threats, many tools like IDS’s (Intrusion Detection Systems) are utilized to monitor cyber threats on a network. However, a common problem with these tools is the volume of the logs generated is extreme and does not stop, further increasing the chance for an adversary to go unnoticed until it’s too late. Typically, the initial evidence of an attack is not an isolated event but a part of a larger attack campaign describing prior …

Contrastive Learning For Unsupervised Auditory Texture Models, Christina Trexler

Computer Science and Computer Engineering Undergraduate Honors Theses

Sounds with a high level of stationarity, also known as sound textures, have perceptually relevant features which can be captured by stimulus-computable models. This makes texture-like sounds, such as those made by rain, wind, and fire, an appealing test case for understanding the underlying mechanisms of auditory recognition. Previous auditory texture models typically measured statistics from auditory filter bank representations, and the statistics they used were somewhat ad-hoc, hand-engineered through a process of trial and error. Here, we investigate whether a better auditory texture representation can be obtained via contrastive learning, taking advantage of the stationarity of auditory textures to …

Deep Learning Based Speech Enhancement And Its Application To Speech Recognition, Ju Lin

All Dissertations

Speech enhancement is the task that aims to improve the quality and the intelligibility of a speech signal that is degraded by ambient noise and room reverberation. Speech enhancement algorithms are used extensively in many audio- and communication systems, including mobile handsets, speech recognition, speaker verification systems and hearing aids. Recently, deep learning has achieved great success in many applications, such as computer vision, nature language processing and speech recognition. Speech enhancement methods have been introduced that use deep-learning techniques, as these techniques are capable of learning complex hierarchical functions using large-scale training data. This dissertation investigates the deep learning …

Deepfakes Generated By Generative Adversarial Networks, Olympia A. Paul

Honors College Theses

Deep learning is a type of Artificial Intelligence (AI) that mimics the workings of the human brain in processing data such as speech recognition, visual object recognition, object detection, language translation, and making decisions. A Generative adversarial network (GAN) is a special type of deep learning, designed by Goodfellow et al. (2014), which is what we call convolution neural networks (CNN). How a GAN works is that when given a training set, they can generate new data with the same information as the training set, and this is often what we refer to as deep fakes. CNN takes an input …

Handypose And Vehipose: Pose Estimation Of Flexible And Rigid Objects, Divyansh Gupta

Theses

Pose estimation is an important and challenging task in computer vision. Hand pose estimation has drawn increasing attention during the past decade and has been utilized in a wide range of applications including augmented reality, virtual reality, human-computer interaction, and action recognition. Hand pose is more challenging than general human body pose estimation due to the large number of degrees of freedom and the frequent occlusions of joints. To address these challenges, we propose HandyPose, a single-pass, end-to-end trainable architecture for hand pose estimation. Adopting an encoder-decoder framework with multi-level features, our method achieves high accuracy in hand pose while …

Gourmetnet: Food Segmentation Using Multi-Scale Waterfall Features With Spatial And Channel Attention, Udit Sharma

Theses

Deep learning and Computer vision are extensively used to solve problems in wide range of domains from automotive and manufacturing to healthcare and surveillance. Research in deep learning for food images is mainly limited to food identification and detection. Food segmentation is an important problem as the first step for nutrition monitoring, food volume and calorie estimation. This research is intended to expand the horizons of deep learning and semantic segmentation by proposing a novel single-pass, end-to-end trainable network for food segmentation. Our novel architecture incorporates both channel attention and spatial attention information in an expanded multi-scale feature representation using …

Impact Of Image Complexity On Early Exit Neural Networks For Edge Applications, Sharan Vidash Vidya Shanmugham

Theses

The advancement of deep learning methods has ushered in novel research in the field of computer vision as the success of deep learning methods are irrefutable when it comes to images and video data. However deep learning methods such as convolutional neural networks are computationally heavy and need specialized hardware to give results within a reasonable time. Early-exit neural networks offer a solution to reducing computational complexity by placing exits in traditional networks bypassing the need to compute the output of all convolutional layers. In this thesis, a reinforcement learning-based exit selection algorithm for early-exit neural networks is analyzed. The …

Multi-Object Localization In Robotic Hand, Tsing Tsow

USF Tampa Graduate Theses and Dissertations

We have developed a machine learning approach to localized objects inside a robotic hand using only images from 2D cameras. Specifically, we used deep learning method (You Only Look Once, YOLO) and Iterative closest Point (ICP) to estimate the 3D coordinates of the objects in a robotic hand. This method will also output the number of objects inside the robotic hand in addition to the coordinates of the objects. We have demonstrated the performance with simulation and obtained typical accuracy within a few pixels (couple mm) and counting accuracy of about 76%. We have also applied it to real images, …

Deep Learning Models For Irregularly Sampled And Incomplete Time Series, Satya Narayan Shukla

Doctoral Dissertations

Irregularly sampled time series data arise naturally in many application domains including biology, ecology, climate science, astronomy, geology, finance, and health. Such data present fundamental challenges to many classical models from machine learning and statistics. The first challenge with modeling such data is the presence of variable time gaps between the observation time points. The second challenge is that the dimensionality of the inputs can be different for different data cases. This occurs naturally due to the fact that different data cases are likely to include different numbers of observations. The third challenge is that different irregularly sampled instances have …

Towards More Trustworthy Deep Learning: Accurate, Resilient, And Explainable Countermeasures Against Adversarial Examples, Fei Zuo

Theses and Dissertations

Despite the great achievements made by neural networks on tasks such as image classification, they are brittle and vulnerable to adversarial example (AE) attacks, which are crafted by adding human-imperceptible perturbations to inputs in order that a neural-network-based classifier incorrectly labels them. Along with the prevalence of deep learning techniques, the threat of AEs attracts increasingly attentions since it may lead to serious consequences in some vital applications such as disease diagnosis.

To defeat attacks based on AEs, both detection and defensive techniques attract the research community’s attention. Given an input image, the detection system outputs whether it is an …

Traffic Sign And Light Detection Using Deep Learning For Automotive Applications, Humaira Naimi

Electronic Theses and Dissertations

Traffic sign and light detection are core components of Advanced Driver Assistance Systems (ADAS) and self-driving vehicles. The automotive industry is widely employing numerous approaches for automation through computer vision techniques. Object detection algorithms based on deep learning can be divided into two main categories, two stage and single stage detection algorithms. Two stage algorithms are designed to improve detection accuracy. While single stage algorithms are constructed to be faster, this increases their suitability for real time applications. This thesis presents a lightweight traffic sign and light detector by adapting a single stage, Single Shot Multibox Detection (SSD) algorithm by …