136879 research outputs found
Sort by
Phylogeography and Systematics of the Sand Shiner, Notropis stramineus (Cope, 1865)
No full textNotropis Rafinesque, 1818 is a species-rich genus of North American minnows that has been demonstrated to harbor ���cryptic��� diversity. The Sand Shiner, Notropis stramineus, is one of the most widespread North American minnows. Various subspecific classifications have been proposed for the Sand Shiner, the most widely accepted of these comprising two subspecies, N. s. stramineus and N. s. missuriensis, based largely on minor differences in scale-row counts between individuals inhabiting tributaries to the Great Lakes, upper Mississippi River and Texas Gulf Coast river systems (N. s. stramineus) and those inhabiting the Missouri and Arkansas River systems (N. s. missuriensis). Analysis of three loci revealed that the Sand Shiner, N. stramineus comprises five clades, and further, other members of Notropis were recovered between clades of N. stramineus, though gene tree discordance created difficulties for determining species boundaries. Ultra-conserved elements (UCEs), enriched from each clade of N. stramineus, related species, and several outgroup taxa, were then analyzed. All least-inclusive clades recovered were identical across analyses and corresponded to clades recovered in Chapter II, however, relationships between the clades differed. In contrast to Chapter II, most clades comprising N. stramineus were monophyletic, and are here termed the N. stramineus species complex. As in Chapter II, a separate clade of N. stramineus was recovered outside of the clade containing the N. stramineus species complex as the sister taxon to a species of Notropis not previously considered to be closely related (N. chihuahua). These results demonstrate that N. stramineus is likely comprised of five distinct evolutionary lineages, which are diagnosed and described (clades B1, B2 and D) or redescribed (clades A and C) in the final chapter, with species diagnosed using characters derived from aspects of pigmentation, body and head shape, tuberculation, and osteology
Essays on Social Preferences
No full textThis dissertation includes three essays in the field of behavioral economics, with a special focus on social preferences using laboratory experiments. The first essay investigates the influence of patient autonomy on doctors��� performance. Using a theory-driven laboratory experiment, I find that when the patient is not able to assess the doctor���s diagnostic precision, for those patients who are not fully compliant with doctors��� advice, doctors will reduce their investment of effort in the diagnosis. This reduction is the largest among those doctors who prioritize patients��� well-being. The experiment also investigates two institutional changes, communication and reputation, which both effectively improve patients��� well-being. This study contributes to experimental health economics by uncovering the potential detrimental effect of patient autonomy on a doctor���s performance and the patient���s health status.
The second essay explores the effectiveness of two punishment strategies for addressing the free-rider problem in public goods production. By varying the timing of punishment in a public good game, we differentiate between the Post-Punishment rule, expected to induce emotional arousal, and the Pre-Punishment rule, aimed at strategic considerations with minimal emotional impact. Pupil dilation data from eye trackers support our hypotheses, revealing that the Post-Punishment rule���s success relies on negative emotions, while the Pre-Punishment rule does not. This study sheds light on the role of negative emotions in punishment efficacy and introduces a novel punishment rule that operates independently of negative emotional responses within a group.
The third essay examines the trade-offs individuals make between money, honesty, and altruism through a sender-receiver game that allows truth-telling, selfish lies, and altruistic lies. We propose a theoretical model identifying five unique types of senders. Experimental findings show prevalent yet diverse patterns of trade-offs among these three domains of concern. This research adds to the literature on lying behavior by revealing varied preferences across moral domains and a widespread
propensity for costly altruistic lies
San Giacomo di Galizia: The Digital Reconstruction of a Galleon of the Anglo-Spanish War of 1585 ��� 1604
No full textThe San Giacomo di Galizia was a late-16th century galleon employed by Spain during the Anglo-Spanish War that took place between 1585 and 1604. After a failed attempt to capture the English port of Falmouth, the ship returned to Ribadeo on the north coast of Spain, where it sunk due to the damage received in foul weather. The purpose of the thesis was to digitally reconstruct the vessel based on the archaeological evidence, primary historical sources, and contemporary naval treatises in order to analyze the hydrostatic features of the galleon to comprehend the seagoing performance of similar ships of this period. Likewise, the project aimed to determine the efficiency of this new methodology as a digital tool for nautical archaeology and historical research
Characterization of Ligand Binding Using Dissolution DNP Assisted NMR Spectroscopy
No full textBiomolecular interactions play essential roles in cellular processes including signaling, metabolism, and enzymatic synthesis of cellular components. Elucidating interactions between proteins and ligands using techniques such as nuclear magnetic resonance (NMR) spectroscopy provides fundamental insights into biological function, as well as guidance on the identification of new drug candidates. A significant NMR sensitivity improvement of several thousand-fold can be achieved by hyperpolarizing the ligand molecule using dissolution dynamic nuclear polarization (D-DNP). Spectra can be acquired in a reduced time, at or near physiological concentrations. Here, transverse (R2) NMR relaxometry is demonstrated to probe protein-ligand interactions. A 13C R2 relaxation dispersion measurement characterizes ligand binding epitopes through the observation of relaxation rates at different positions of the ligand, whereby the magnitude of the dispersion reflects the binding orientation of the ligand. The efficiency of the R2 measurement can be improved by an ultrafast approach to obtain the relaxation rates from all 13C spins in a single measurement. Numerous target proteins for pharmaceuticals are embedded in the cell membrane. Hyperpolarized 19F, due to its low NMR detection limit, is proposed for probing the interactions with membranes and the cell surface proteins. A model for the binding interaction combined with predictions of spin relaxation rates provides estimates of the binding affinity to membranes of different compositions in unilamellar vesicles. Applied to the measurement of ligand interactions with different cell types, the influence of the interactions between ligands and cell membrane proteins is identified
Advancing Stress Detection: Towards Real-Time, Naturalistic, and Personalized Artificial Intelligence Approaches for Stress and Mental Disorder Detection
No full textThis dissertation presents a comprehensive exploration of Machine Learning (ML) and Deep Learning (DL) methodologies for the detection, prediction, and analysis of stress and stress-related mental disorders (MDs). The research begins with a systematic review of existing ML algorithms, preprocessing techniques, and data types used in stress detection, highlighting the superior performance of Support Vector Machine (SVM), Neural Network (NN), and Random Forest (RF) models. The review also underscores the prevalent use of physiological parameters, such as heart rate measurements, as stress predictors, and identifies significant research gaps, including model interpretability, personalization, the incorporation of naturalistic settings, and real-time processing capabilities.
Addressing these gaps, the dissertation presents a series of studies. The first study explores a machine learning-based method for continuous monitoring and identification of stress in a naturalistic setting among college students, using a mobile health (mHealth) application and wearable wrist-worn sensors. The XGBoost model was found to be the most reliable in this context, with an accuracy of 84.5%.
The second study proposes a stress detection methodology that utilizes an array of deep learning models, including feedforward neural networks (FFDNN), Conv1D, long short-term memory (LSTM), hybrid Conv1D-LSTM, and Conv1D-LSTM with attention mechanism. The hybrid Conv1D-LSTM model augmented with an attention mechanism outperformed the other models, achieving a cross-validation accuracy of 0.89 and an area under the curve (AUC) score of 0.94. The final study explores the potential of individualized ML and DL algorithms in accurately identifying stress responses in individuals afflicted with Post-Traumatic Stress Disorder (PTSD). The findings demonstrate that individualized models, particularly advanced deep learning models such as LSTM and hybrid CNN-RNN models, consistently surpass standard models in accurately identifying stress in PTSD patients.
In conclusion, this dissertation underscores the importance of employing advanced ML and DL techniques, such as hybrid models and attention mechanisms, and individualized approaches to improve stress detection performance. The findings contribute significantly to the field of digital health, suggesting innovative strategies for monitoring and managing stress in various populations
Planning Under Uncertainty with Unreliable Robotic Actuators
No full textWe focus on a critical aspect of autonomous robotics: the challenge of decision-making under the uncertainty of inevitable actuator degradation. Through the lens of physically embodied decision-makers, this research explores the complexity in modeling and planning for robotic actuator deterioration and failure. By an analogy to biological aging, we explore the necessity for agents to anticipate and plan for their own senescence, thus embracing their finite lifespan to maximize their utility. This shift towards acknowledging and planning for actuator frailty is particularly crucial for robotic explorers on interplanetary and interstellar missions, where autonomous, resilient decision-making is paramount.
Central to our approach is the introduction of Fallible Actuator Markov Decision Processes (FA-MDPs), an extension of the traditional MDP framework that incorporates actuator reliability into planning. This allows for the anticipation of failures, enabling strategic actuator usage and rapid adaptation post-failure. Our methodology leverages the inherent structure of FA-MDPs to decompose the problem into manageable sub-problems which increases solver efficiency. Furthermore, we explore the concept of actuator dominance and introduce virtual actuators to model k-shot and degrading actuators, thereby extending our failure model and improving planning performance.
The contributions of this thesis include: (1) A novel framework for incorporating actuator reliability into planning, enabling proactive planning for actuator failures. (2) An improved solution methodology for FA-MDPs through problem decomposition and a value function lattice, demonstrating superior performance over naive solvers. (3) An analysis of actuator relationships to further enhance planning performance and address actuator degradation. This work represents a step towards the development of long-lived autonomous robots capable of navigating the uncertainties of dynamic environments and their own inevitable deterioration
AI and Machine Learning Using Wearables for Diabetes Care
No full textThe emergence of Internet of things (IoT) devices and technological advances have revolutionized healthcare, particularly the advent of AI-based wearables have enabled frequent monitoring of glucose levels in patients. This is critical for an incurable disease like diabetes which can only be managed through frequent monitoring of glucose values. Despite the progress made, making AI-based solutions more patient-centric remains a challenge. To address this, we present methods for efficient application of AI/ML solutions for improving diabetes care with an emphasis on patient needs.
This dissertation has two primary objectives: (1) To build robust machine learning models to improve diabetes care, and (2) develop alternatives to current glucose monitoring technologies to enhance the accessibility and reduce the intrusiveness. To achieve the Objective 1, we focus on developing machine learning algorithms for prediction of impending hypoglycemia risk in patients. A feature-based machine-learning model is built based on previous CGM values that gives real-time predictions for hypoglycemia risk, enabling patients to take intervening actions. We subsequently work on improving the quality of hypoglycemia predictive alerts in a real-world setting by focusing on predictive alerts that are based on sustained hypoglycemia. This drastically reduces instances of false alerts, a major deterrent for technology adoption among patients.
Machine learning models with robust performance rely on large corpus of data for training. However, healthcare data is sensitive, and its accessibility is restricted with many regulations in place. To this, we develop FedGlu, a machine learning model trained in a federated learning framework that simultaneously addresses the dual challenge of data availability and model performance. FedGlu also incorporates a customized loss function that improves the model���s predictive capabilities in the glycemic excursion regions.
CGM devices are valuable, but accessibility is limited as they are expensive and also available based only on prescriptions. In addition, they are invasive which can be painful for patients. In Objective 2, as an alternative to CGM devices, we extend the use of IoT based noninvasive wearables for glucose monitoring. For this, we evaluate hyperglycemia detection along with hypoglycemia detecting using ECG and accelerometer signals that are collected noninvasively. We comprehensively evaluate the proposed algorithms on people with and without diabetes and demonstrate the efficacy of the proposed approach. In closing, a summary of the contributions and directions of future work are presented
Investigating the Impact of Timing of Basal Leaf Removal and Fruit Thinning on Potassium Accumulation in Red Wine Grapes
No full textIn hot climates such as Texas, high juice/wine pH represents a serious challenge for wineries. The objective of this study was to evaluate the impact of timing of basal leaf removal and cluster thinning on potassium (K+) concentration in ���Tempranillo��� and ���Camminare Noir��� grapes as a possible vineyard management practice to mitigate high pH. This research took place during the 2021 and 2022 growing seasons in two vineyards, one in the Texas Gulf Coast and the other in the North Texas region. Treatments consisted of basal leaf removal (removal of lowest three basal leaves), cluster thinning (thinning to one cluster per shoot), and leaf removal plus cluster thinning conducted at either berry set or veraison. Differences in fruiting zone canopy density were consistently observed between treatments, across years, cultivars, and sites. Leaf removal treatments averaged a 35.83% reduction in canopy density in the fruiting zone, determined by occlusion layer, leading to an increase of 124.17% in cluster exposure flux availability. However, differences in yield (yield per vine, clusters per vine), berry chemical composition (alpha amino nitrogen, ammonium, fructose, malic acid, soluble solids, tartaric acid, titratable acidity, and pH) and tissue nutrient content were only observed in singular instances, with no consistent differences between years, cultivar, or site. Berry K + concentrations and juice pH varied by rootstock, year, and cultivar, but there was no clear impact of leaf removal or cluster thinning on berry K+ or pH. In both years of the study, very low yields were observed as a result of a severe winter event and possibly negated any effects from leaf removal or cluster thinning
Machine Learning-Based Automated Fault Detection and Diagnostics in Building Systems
No full textAutomated fault detection and diagnostics (AFDD) analysis in commercial building systems using machine learning (ML) can improve the building���s efficiency and conserve energy costs from inefficient equipment operation. Boolean rules-based analysis is standard in current AFDD solutions but limits analysis to the rules defined and calibrated by energy engineers. As part of this dissertation, an automated process was developed to provide ML-based building analytics to building engineers and operators with minimal training in ML. The process can be applied to buildings with a variety of configurations, which reduces time and manual effort required for fault analysis when compared to Boolean rule-based systems. The developed procedure introduces advanced diagnostics with automatically generated metrics to validate the ML model���s predictions and rank detected faults in order of fault severity. Explanations of the methodology used for the ML analysis include a description of the algorithms used.
The analysis was applied to a building on the Texas A&M University campus where the results are shown to illustrate the performance of the process using measured data from a commercial building. Three case studies which analyze the building���s equipment are presented to show ML���s advantages over rule-based analysis. ML can detect faults in the system caused by degrading components. ML can also detect faults in system components with missing sensors by modeling expected system operation and making comparisons to actual system operation. An example of ML detecting a failure in a building is shown along with a demonstration of the decision boundaries of ML-based FDD in comparison with Boolean rule-based analysis. The results from these examples are used to show the strengths and weaknesses of using ML for AFDD analysis
Effect of Sample Size on Linear Elastic Fracture Toughness of FFF-Processed PLA Compact Tension Specimens
No full textThe "Effect of Sample Size on Linear Elastic Fracture Toughness of FFF-Processed PLA Compact Tension Specimens" investigates the impact of sample size on the fracture toughness of 3D printed Polylactic Acid (PLA) using Fused Filament Fabrication (FFF) technology. ASTM D5045 standard is widely recognized and employed as a benchmark for evaluating the fracture toughness of materials within the linear-elastic fracture mechanics (LEFM) domain. This research aims to assess the applicability of this standard for evaluating fracture toughness in additive manufacturing thermoplastic materials focusing on the influence of specimen size, among other parameters, on the material's mechanical properties. Through a comprehensive experimental setup involving various sample sizes, infill sizes, and layout patterns, the study provides a detailed analysis of fracture toughness behavior through numerous tests. The findings reveal significant dependencies of fracture toughness on the sample size, indicating a deviation from true plane strain conditions and highlighting the necessity of reporting conditional stress intensity factors (����_����) for this particular case. The research contributes valuable insights into the mechanical characterization of 3D-printed thermoplastic materials and suggests practical implications for improving additive manufacturing testing and evaluation practices