Digital Commons Network^™

A Computational Approach Within Medical Research, Ryan Christopher Hogan

Theses and Dissertations

Within the context of medical image diagnosis, we explore novel computational models to facilitate the detection of two medical conditions that burden our society. In particular, this research focuses on the use of deep learning models for the detection of Alzheimer’s Disease in Magnetic Resonance images (MRI) scans, as well as the detection of heart arrhythmias from electrocardiogram (ECG) recordings. We propose a novel architecture that depends on the 3D-CNN model to classify between MRI scans of cognitively healthy individuals and AD patients. Moreover, we explore the use of LSTM deep learning models to detect abnormal heart arrhythmias that present …

A Multi-Resolution Graph Convolution Network For Contiguous Epitope Prediction, Lisa Oh

Dartmouth College Master’s Theses

Computational methods for predicting binding interfaces between antigens and antibodies (epitopes and paratopes) are faster and cheaper than traditional experimental structure determination methods. A sufficiently reliable computational predictor that could scale to large sets of available antibody sequence data could thus inform and expedite many biomedical pursuits, such as better understanding immune responses to vaccination and natural infection and developing better drugs and vaccines. However, current state-of-the-art predictors produce discontiguous predictions, e.g., predicting the epitope in many different spots on an antigen, even though in reality they typically comprise a single localized region. We seek to produce contiguous predicted epitopes, …

Advancement Auto-Assessment Of Students Knowledge States From Natural Language Input, Nisrine Ait Khayi

Electronic Theses and Dissertations

Knowledge Assessment is a key element in adaptive instructional systems and in particular in Intelligent Tutoring Systems because fully adaptive tutoring presupposes accurate assessment. However, this is a challenging research problem as numerous factors affect students’ knowledge state estimation such as the difficulty level of the problem, time spent in solving the problem, etc. In this research work, we tackle this research problem from three perspectives: assessing the prior knowledge of students, assessing the natural language short and long students’ responses, and knowledge tracing.Prior knowledge assessment is an important component of knowledge assessment as it facilitates the adaptation of the …

Enhancing Drug Overdose Mortality Surveillance Through Natural Language Processing And Machine Learning, Patrick J. Ward

Theses and Dissertations--Epidemiology and Biostatistics

Epidemiological surveillance is key to monitoring and assessing the health of populations. Drug overdose surveillance has become an increasingly important part of public health practice as overdose morbidity and mortality has increased due in large part to the opioid crisis. Monitoring drug overdose mortality relies on death certificate data, which has several limitations including timeliness and the coding structure used to identify specific substances that caused death. These limitations stem from the need to analyze the free-text cause-of-death sections of the death certificate that are completed by the medical certifier during death investigation. Other fields, including clinical sciences, have utilized …

Predicting Carcass Cut Yields In Cattle From Digitalimages Using Artificial Intelligence, Darragh Matthews

Theses

Beef carcass classification in Europe is predicated on the EUROP grid for both fatness and conformation. Although this system performs well for grouping visually similar carcasses, it cannot be used to accurately predict meat yields from these groups, especially when considered on an individual cut level. Deep Learning (DL) has proven to be a successful tool for many image classification problems but has yet to be fully proven in a regression scenario using carcass images. Here we have trained DL models to predict carcass cut yields and compared predictions to more standard machine learning (ML) methods. Three approaches were undertaken …

Fabrication, Characterization Of High-Entropy Alloys And Deep Learning-Based Inspection In Metal Additive Manufacturing, Wenyuan Cui

Doctoral Dissertations

"Alloying has been used to confer desirable properties to materials. It typically involves the addition of small amounts of secondary elements to a primary element. In the past decade, however, a new alloying strategy that involves the combination of multiple principal elements in high concentrations to create new materials called high- entropy alloys (HEAs) has been in vogue. In the first part, the investigation focused on the fabrication process and property assessment of the additive manufactured HEA to broaden its engineering applications. Additive manufacturing (AM) is based on manufacturing philosophy through the layer-by-layer method and accomplish the near net-shaped components …

Deep Learning For Hyperspectral Image Segmentation In Biosecurity Scanning, Ly Bui

University of Wollongong Thesis Collection 2017+

Biosecurity scanning plays a crucial role in preventing exotic pests, weeds and contaminants from entering a country through shipping containers. Exposure to biosecurity risks causes a substantial loss to the native environment, production value and public health. Currently, these threats are managed via manual inspection, detector dogs and x-ray scanners; however, these procedures are time-consuming, error-prone, or costly.

In this research, we propose a novel approach for biosecurity risk detection that utilizes hyperspectral imaging technology and semantic image segmentation. This approach segments the target objects in a hyperspectral image by analyzing their spatial and spectral signatures. The target objects in …

A Deep Learning Approach To Lncrna Subcellular Localization Using Inexact Q-Mer, Weijun Yi

Graduate Theses, Dissertations, and Problem Reports

Long non coding Ribonucleic Acids (lncRNAs) can be localized to different cellular components, such as the nucleus, exosome, cytoplasm, ribosome, etc. Their biological functions can be influenced by the region of the cell they are located. Many of these lncRNAs are associated with different challenging diseases. Thus, it is crucial to study their subcellular localization. However, compared to the vast number of lncRNAs, only relatively few have annotations in terms of their subcellular localization. Conventional computational methods use q-mer profiles from lncRNA sequences and then train machine learning models, such as support vector machines and logistic regression with the profiles. …

Texture-Driven Image Clustering In Laser Powder Bed Fusion, Alexander H. Groeger

Browse all Theses and Dissertations

The additive manufacturing (AM) field is striving to identify anomalies in laser powder bed fusion (LPBF) using multi-sensor in-process monitoring paired with machine learning (ML). In-process monitoring can reveal the presence of anomalies but creating a ML classifier requires labeled data. The present work approaches this problem by printing hundreds of Inconel-718 coupons with different processing parameters to capture a wide range of process monitoring imagery with multiple sensor types. Afterwards, the process monitoring images are encoded into feature vectors and clustered to isolate groups in each sensor modality. Four texture representations were learned by training two convolutional neural network …

Reliable And Interpretable Machine Learning For Modeling Physical And Cyber Systems, Daniel L. Marino Lizarazo

Theses and Dissertations

Over the past decade, Machine Learning (ML) research has predominantly focused on building extremely complex models in order to improve predictive performance. The idea was that performance can be improved by adding complexity to the models. This approach proved to be successful in creating models that can approximate highly complex relationships while taking advantage of large datasets. However, this approach led to extremely complex black-box models that lack reliability and are difficult to interpret. By lack of reliability, we specifically refer to the lack of consistent (unpredictable) behavior in situations outside the training data. Lack of interpretability refers to the …

Deep Learning Assisted Intelligent Visual And Vehicle Tracking Systems, Liang Xu

Theses and Dissertations

Sensor fusion and tracking is the ability to bring together measurements from multiple sensors of the current and past time to estimate the current state of a system. The resulting state estimate is more accurate compared with the direct sensor measurement because it balances between the state prediction based on the assumed motion model and the noisy sensor measurement. Systems can then use the information provided by the sensor fusion and tracking process to support more-intelligent actions and achieve autonomy in a system like an autonomous vehicle. In the past, widely used sensor data are structured, which can be directly …

Machine Learning Based Defect Detection In Robotic Wire Arc Additive Manufacturing, Yuxing Li

University of Wollongong Thesis Collection 2017+

In the last ten years, research interests in various aspects of the Wire Arc Additive Manufacturing (WAAM) processes have grown exponentially. More recently, efforts to integrate an automatic quality assurance system for the WAAM process are increasing. No reliable online monitoring system for the WAAM process is a key gap to be filled for the commercial application of the technology, as it will enable the components produced by the process to be qualified for the relevant standards and hence be fit for use in critical applications in the aerospace or naval sectors. However, most of the existing monitoring methods only …

Computational Intelligence For The Micro Learning, Jiayin Lin

University of Wollongong Thesis Collection 2017+

The developments of the Web technology and the mobile devices have blurred the time and space boundaries of people’s daily activities, which enable people to work, entertain, and learn through the mobile device at almost anytime and anywhere. Together with the life-long learning requirement, such technology developments give birth to a new learning style, micro learning. Micro learning aims to effectively utilise learners’ fragmented spare time and carry out personalised learning activities. However, the massive volume of users and the online learning resources force the micro learning system deployed in the context of enormous and ubiquitous data. Hence, manually managing …

Deep Learning Based Microatoll Detection From Drone Images, Zhexuan Zhou

University of Wollongong Thesis Collection 2017+

Visual object detection has made significant progress with the advent of deep neural networks and has been extensively applied. This thesis reports a novel application that aims to detect individual microatolls, which are circular coral colonies, from island images captured by drones. We first describe data collection and labelling to create a novel microatoll dataset for the microatoll detection task from drone images. Upon this dataset, the state-of-the-art object detectors are then evaluated for this task. To better integrate a detector with the characteristic of microatolls, we propose a modified detector called Microatoll-Net. It actively extracts features from the surrounding …

Adaptive Two-Stage Edge-Centric Architecture For Deeply-Learned Embedded Real-Time Target Classification In Aerospace Sense-And-Avoidance Applications, Nicholas A. Speranza

Browse all Theses and Dissertations

With the growing number of Unmanned Aircraft Systems, current network-centric architectures present limitations in meeting real-time and time-critical requirements. Current methods utilizing centralized off-platform processing have inherent energy inefficiencies, scalability challenges, performance concerns, and cyber vulnerabilities. In this dissertation, an adaptive, two-stage, energy-efficient, edge-centric architecture is proposed to address these limitations. A novel, edge-centric Sense-and-Avoidance architecture framework is presented, and a corresponding prototype is developed using commercial hardware to validate the proposed architecture. Instead of a network-centric approach, processing is distributed at the logical edge of the sensors, and organized as Detection and Classification Subsystems. Classical machine vision algorithms are …

Deep Learning For Compressive Sar Imaging With Train-Test Discrepancy, Morgan R. Mccamey

Browse all Theses and Dissertations

We consider the problem of compressive synthetic aperture radar (SAR) imaging with the goal of reconstructing SAR imagery in the presence of under sampled phase history. While this problem is typically considered in compressive sensing (CS) literature, we consider a variety of deep learning approaches where a deep neural network (DNN) is trained to form SAR imagery from limited data. At the cost of computationally intensive offline training, on-line test-time DNN-SAR has demonstrated orders of magnitude faster reconstruction than standard CS algorithms. A limitation of the DNN approach is that any change to the operating conditions necessitates a costly retraining …

Augmenting Structure/Function Relationship Analysis With Deep Learning For The Classification Of Psychoactive Drug Activity At Class A G Protein-Coupled Receptors, Hannah Willow Shows

Browse all Theses and Dissertations

G protein-coupled receptors (GPCRs) initiate intracellular signaling pathways via interaction with external stimuli. [1-5] Despite sharing similar structure and cellular mechanism, GPCRs participate in a uniquely broad range of physiological functions. [6] Due to the size and functional diversity of the GPCR family, these receptors are a major focus for pharmacological applications. [1,7] Current state-of-the-art pharmacology and toxicology research strategies rely on computational methods to efficiently design highly selective, low toxicity compounds. [9], [10] GPCR-targeting therapeutics are associated with low selectivity resulting in increased risk of adverse effects and toxicity. Psychoactive drugs that are active at Class A GPCRs used …

Texture-Driven Image Clustering In Laser Powder Bed Fusion, Alexander H. Groeger

Browse all Theses and Dissertations

The additive manufacturing (AM) field is striving to identify anomalies in laser powder bed fusion (LPBF) using multi-sensor in-process monitoring paired with machine learning (ML). In-process monitoring can reveal the presence of anomalies but creating a ML classifier requires labeled data. The present work approaches this problem by printing hundreds of Inconel-718 coupons with different processing parameters to capture a wide range of process monitoring imagery with multiple sensor types. Afterwards, the process monitoring images are encoded into feature vectors and clustered to isolate groups in each sensor modality. Four texture representations were learned by training two convolutional neural network …

Bickel-Rosenblatt Test Based On Tilted Estimation For Autoregressive Models & Deep Merged Survival Analysis On Cancer Study Using Multiple Types Of Bioinformatic Data, Yan Su

Browse all Theses and Dissertations

This dissertation includes two topics, Bickel-Rosenblatt test based on tilted density estimation for autoregressive models and deep merged survival analysis on cancer study using multiple types of bioinformatic data. In the first topic study, we consider the goodness of fit test the error density of linear and nonlinear autoregressive models using tilted kernel density estimation based on residuals. Bickel-Rosenblatt test statistic is based on the integrated square error of non-parametric error density estimation and a smoothed version of the parametric fit of the density. It is shown that the new type of Bickel-Rosenblatt test statistics behaves asymptotically the same as …

Implementing A Neural Network For Supervised Learning With A Random Configuration Of Layers And Nodes, Kane A. Phillips

Electronic Theses and Dissertations

Deep learning has a substantial amount of real-life applications, making it an increasingly popular subset of artificial intelligence over the last decade. These applications come to fruition due to the tireless research and implementation of neural networks. This paper goes into detail on the implementation of supervised learning neural networks utilizing MATLAB, with the purpose being to generate a neural network based on specifications given by a user. Such specifications involve how many layers are in the network, and how many nodes are in each layer. The neural network is then trained based on known sample values of a function …

Deep Fingerprint Matching From Contactless To Contact Fingerprints For Increased Interoperability, Alexander James Wilson

Graduate Theses, Dissertations, and Problem Reports

Contactless fingerprint matching is a common form of biometric security today. Most smartphones and associated apps now let users opt into using this form of biometric security. However, it’s difficult to match a finger-photo to a fingerprint because of perspective distortion occurring at the edges of the finger-photo, so direct matching using conventional methods will not be as accurate due to a lack of sufficient matching minutiae points. To address this issue, we propose a deep model, Perspective Distortion Rectification Model (PDRM), to estimate the fingerprint correspondence for finger-photo images in order to recover more minutiae points. Not only do …

Single And Differential Morph Attack Detection, Baaria Chaudhary

Graduate Theses, Dissertations, and Problem Reports

Face recognition systems operate on the assumption that a person's face serves as the unique link to their identity. In this thesis, we explore the problem of morph attacks, which have become a viable threat to face verification scenarios precisely because of their inherent ability to break this unique link. A morph attack occurs when two people who share similar facial features morph their faces together such that the resulting face image is recognized as either of two contributing individuals. Morphs inherit enough visual features from both individuals that both humans and automatic algorithms confuse them. The contributions of this …