Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Unconventional Symmetry Breaking in Non-hermitian Floquet Systems
No full textTime crystals are systems that spontaneously break time-translation symmetry, exhibiting
repeating temporal patterns. Recent advancements have demonstrated that non-Hermitian
Floquet systems can host time crystalline phases with quasi-long-range order. This dis-
sertation introduces the background of non-equilibrium physics, non-Hermitian quantum
many-body systems, and quantum computing. It presents a comprehensive study of non-
Hermitian time crystals through two interconnected projects, bridging theoretical extensions
and experimental implementations. In the first project, we extend the non-Hermitian Floquet
transverse-field Ising model by introducing a non-integrable interaction term. Utilizing
numerical Time-Evolving Block Decimation (TEBD) simulations, mean-field analysis, and
perturbation theory, we discover that the interaction term induces a significant shift in the
phase diagram and leads to an unexpected symmetry-breaking transition not predicted by
mean-field approaches. Through average Hamiltonian theory, we trace this symmetry breaking
to a ferromagnetic transition in the anisotropic non-Hermitian XXZ model, highlighting the
intricate interplay between non-Hermitian effects and many-body interactions. The second
project focuses on implementing non-Hermitian quantum dynamics on noisy quantum com-
puters, specifically examining a non-Hermitian Ising Floquet model that exhibits persistent
temporal oscillations—a hallmark of time crystallinity. In a simplified two-qubit setup, we
identify conditions under which an infinitely long-lived periodic steady state emerges. How-
ever, our generalized Floquet analysis demonstrates that these oscillations are generically lost
in the presence of arbitrarily weak noise, and we compute the corresponding damping rates.
Simulations conducted using IBM’s Qiskit platform corroborate these theoretical predictions,
yet experimental executions on the ibmq-lima device fail to observe the anticipated oscillations,
underscoring the significant impact of quantum noise and hardware imperfections on the
realization of non-Hermitian dynamics. Collectively, these projects advance the understanding
of non-Hermitian time crystals by elucidating both their theoretical underpinnings and the
practical challenges of their experimental realization, thereby contributing to the broader
knowledge of non-equilibrium quantum systems and the potential of quantum computing
technologies in exploring complex quantum phenomena
Exploring the Impact of Chronic Health Conditions on US Maternal Mortality: a Comprehensive Analysis Beyond Racial Disparities
No full textMaternal mortality rates in many countries, including the United States, remain alarmingly high
despite advances in healthcare. While racial disparity has historically been a significant factor,
recent research suggests that pre-existing health conditions are a major factor in maternal
mortality. This thesis examines the shift from focusing solely on racial disparities to
acknowledging the influence of pre-existing health conditions on maternal mortality rates. By
analyzing various studies, healthcare policies, and interventions, this paper aims to highlight the
importance of addressing pre-existing health conditions to effectively reduce maternal mortality
rates and promote equitable maternal healthcare access
Advanced Approaches in NLP and Security: Addressing Catastrophic Forgetting Through Continual Learning and Resolving Data Imbalance in Semi-supervised Settings
No full textIn the rapidly evolving field of machine learning, particularly in applications demanding
continual or sequential learning, the phenomenon of catastrophic forgetting poses a significant challenge. This issue occurs when a model, trained on new tasks, inadvertently loses
information related to earlier learned tasks. Several innovative methodologies have been
developed to address this problem without relying on traditional methods that often require
additional memory or compromise privacy.
One such approach is the introduction of calibration techniques that adjust both parameters and output logits to balance the preservation of old knowledge with the acquisition of
new concepts, as exemplified in frameworks that incorporate Logits Calibration (LC) and
Parameter Calibration (PC). These techniques ensure the retention of previously learned
parameters while integrating new information, thereby maintaining performance across a
variety of tasks, such as those in the General Language Understanding Evaluation (GLUE)
benchmark.
Another promising method involves the use of Energy-Based Models (EBMs), which associate
an energy value with each input and allow the sampling of data points from previous tasks
during new task training. This method has been adapted in different solutions, with the
latter combining EBMs with Dynamic Prompt Tuning (DPT) to adaptively adjust prompt
parameters for each task, efficiently generating training samples from past tasks and thus
mitigating the effects of catastrophic forgetting.
In the realm of cybersecurity, particularly in analyzing imbalanced, tabular data sets such
as those encountered in industrial control systems and cybersecurity monitoring, semi-
supervised learning techniques have been employed. These methods leverage a mix of labeled
and unlabeled data and utilize novel data augmentation techniques triplet mixup to overcome the challenges posed by limited labeled data and the loss of contextual information.
These approaches have demonstrated effectiveness in detecting vulnerabilities and attacks
within cyber-physical systems, highlighting their potential in sectors where high stakes and
high data imbalance are common.
Across these diverse applications, the overarching goal remains consistent: to develop machine learning models capable of continual learning without sacrificing previously acquired
knowledge. By harnessing innovative strategies such as parameter calibration, energy-based
sampling, and semi-supervised learning with data augmentation, we are setting new benchmarks in the field, ensuring that models not only retain old knowledge but also seamlessly
integrate new information, thereby paving the way for more robust, adaptive machine learning applications
Machine Learning Solutions for Wind Turbine and Wind Farm Applications
No full textThis research is mainly focused on machine learning applications in wind turbines and wind
farms. Two applications are explored: the prediction of icing on wind turbines and wind farms
using available Supervisory Control and Data Acquisition (SCADA) data, and the estimation
of incoming wind direction travelling across the farm using Large Eddy Simulation (LES)
data and SCADA data. SCADA data contains ten-minute averages of all the performance
and key signals from the wind turbines in a wind farm. LES data contains any signal, variable
or parameter from turbines in a wind farm; currently LES provides the highest level of fidelity
with reasonable computational cost.
Ice detection and ice prediction in early ice formation stages are important and beneficial.
Ice detection and prediction offer the potential for mitigating power losses due to icing events.
In this dissertation, the prediction of icing on wind turbines is achieved using available
features from the SCADA data such as generated power, gear bearing temperature, and
other time dependent variables. A Temporal Convolutional Network (TCN) is used as the
prediction model due to its recent popularity in various prediction tasks when long-range
dependencies in data sequences are of significance. The prediction of icing at the wind farm
level is accomplished by using fusion methods among all the wind turbines in the farm. The
farm level prediction of icing has better accuracy than the single turbine level prediction. The
variance across different prediction horizons is also reduced at the wind farm level prediction,
which indicates that uncertainty in predictions is decreased when the TCN models use data
from all wind turbines.
Knowledge of wind direction within wind plants is critical for their operation. Real-time
estimates of wind direction are required to properly orient a single wind turbine for power
maximization (greedy control). Wind direction estimation is performed using the rotor
angular velocity of the turbines in a wind farm. Long short-term memory (LSTM) is used as
the estimation model since it is able to capture complex patterns from input data. LSTM is
also widely used in capturing long-range dependencies. While the TCN architecture could
deliver better accuracy, the work on wind direction estimation was completed before TCN
was discovered. The estimation of wind direction is posed as a classification problem, i.e.,
the model estimates the wind direction among a discrete and finite set of predefined wind
directions. Large eddy simulations (LES) data and SCADA data are used to validate our
methods. LES data are generated from high-fidelity simulations that are used for data analysis
and control algorithms validation without carrying out expensive field experiments. Testing
results show that our LSTM model can estimate incoming wind direction with satisfactory
accuracy.
This dissertation concludes with experimental demonstration of a specific flow control tech-
nique for increasing power production in wind farms - wake steering via yaw angle control.
In this method, the yaw angles of the upstream turbines are intentionally misaligned relative
to the incoming wind direction to steer the wakes away from the downstream turbines. This
dissertation describes the application of an existing wake steering algorithm on experimental
scaled wind turbine models. The wake steering algorithm tested is the so-called Log-of-Power
Proportional Integral Extremum Seeking Control (LP-PIESC), which controls the yaw angles
of the turbines to maximize the wind farm total power output. Experimental tests were
conducted on scaled wind turbines that were designed and fabricated at the Technical Uni-
versity of Munich. These scaled turbines have a 0.6 m rotor diameter and represent a scaled
down version of the DTU 10 MW wind turbine. Experimental results show that the total
power of a cluster with two scaled experimental turbines can be increased using wake steering
LP-PIESC yaw control.
Lessons learned in this dissertation are summarized in the following: icing prediction can be
achieved using our data-driven framework based on SCADA data. Long-term predictions
(e.g., a few days) would allow operators to prepare for shifting supply to generators that are
less impacted by icing event in order to reduce energy loss from wind turbines. While short
predictions (e.g., minutes to hours) can be used to make operational changes on the turbines
to mitigate the impact of icing. Our icing prediction framework can be extended to an entire
farm by using fusion approaches, including decision fusion and feature fusion. Both decision
fusion and feature fusion approaches can enhance the prediction accuracy at the farm level.
Incoming wind direction can be estimated using time samples of rotor speed of the turbines.
A classification neural network model (LSTM model) can be used to estimate the incoming
wind direction into a wind farm. Both LES data and SCADA data have been used to validate
our model. Testing results demonstrate that incoming wind direction can be estimated using
our model with satisfactory accuracy. Finally, a wake steering algorithm (LP-PIESC) is
applied on a cluster of scaled wind turbine models (G06 model). Wind tunnel experimental
results indicate that the total power of the cluster can be increased using LP-PIESC algorithm
for experimental turbines that match the aeroelastic properties of utility-scale machines
Examining Confidence for Same-race and Other-race Face-matching Decisions
No full textIn applied settings (e.g., forensic face examination, passport security), face-matching is car-
ried out by determining whether face images portray the same or different identities. When
expressed with a high level of confidence, face-identification decisions can determine the
outcome of legal proceedings. Despite the importance of confidence judgments in applied
settings, very little research has considered the relationship between confidence and face-
matching accuracy (Phillips et al., 2018; Hahn et al., 2021). In addition, a large body of
research has shown that people are less accurate at recognizing faces of a different race than
faces of their own race (the other-race effect [ORE] Malpass and Kravitz, 1969) (Meissner
and Brigham, 2001). However, no prior research has examined the effect of race on the
relationship between confidence and face-matching accuracy. The goal of my dissertation is
to examine 1) whether people can evaluate the correctness of their face-matching decisions
using comparative-confidence judgments and 2) whether participant- and stimulus-race af-
fect this ability. In addition, I examined the extent to which item-difficulty level informs
comparative-confidence decisions. The study is organized in two parts, as follows. The first
part reports completed work (Jeckeln et al., 2022) (Experiment 1) establishing a baseline un-
derstanding of the relationship between confidence and face-matching accuracy. In this work,
confidence was measured via comparative judgements: Upon completing two face-matching
trials, participants were asked to compare the two trials and select the decision on which
they felt more confident to be correct. Note that stimulus race and participant race were
not controlled. Additionally, I examined whether the difference in difficulty (measured using
Item Response Theory [IRT]) between the paired trials predicts the comparative-confidence
judgments. The results revealed that: 1) comparative-confidence judgments are a good in-
dicator of accuracy and 2) item difficulty predicts confidence decisions. In the second part
(Experiment 2 and 3), my goal was to extend Experiment 1 to same-race and other-race
faces. To achieve this, I constructed a new Cross-race Triad Identity Matching test com-
posed of a balanced set of African American (n = 25) and Caucasian (n = 25) face-image
triads (Experiment 2) with comparable difficulty. Additionally, I modelled (using IRT) the
responses of African American and Caucasian participants on the new test to measure item
difficulty (Experiment 2). In Experiment 3, I implemented the new test and comparative-
confidence judgments (akin to Experiment 1) to examine the relationship between confidence
and accuracy for same-race and other-race faces. In addition, I tested whether item difficulty
predicts confidence decisions for same-race and other-race faces. Results from Experiment
3 showed that the relationship between confidence and accuracy observed in Experiment 1
extends to cross-race identification. Consistent with Experiment 1, item difficulty predicted
confidence decisions for same-race faces. Item difficulty also predicted confidence decisions
when African-American participants identified Caucasian faces, but not when Caucasian par-
ticipants identified African-American faces. This work provides insight into people’s ability
to use comparative-confidence judgments to evaluate the correctness of their face-matching
decisions for same-race and other-race faces
Spinning Stories to Gold: the Grimm Brothers and Their Female Support System – Creating a Visual Narrative
No full textThis creative dissertation consists of a photographic exhibition titled “Spinning Stories to Gold”
and a scholarly dissertation, which highlights the contributions of the women who supplied the
Grimm brothers with over two-thirds of the fairy tales that they published in their famous fairy
tale collection. The photographic exhibition focuses on four fairy tales, “Snow White,” “Little
Red Cap,” “The Frog King,” and “Rumpelstiltskin,” and features emotional characteristics of
Expressionism such as anxiety, horror, and isolation, as well as visual aspects of Expressionism
such as exaggeration, abstraction, and forestry. The research examines Jacob and Wilhelm’s
collection process, insights into their female supporters, characteristics of fairy tales, narrative
photography, German Expressionist films and their connection to the Romantic period, artists
such as Cindy Sherman, Miwa Yanagi, Gregory Crewdson, and Gerhard Richter, and my creative
process. The primary focus is to shed light on some of the female storytellers and to present their
contributions to a wider audience in a contemporary visual format
Enhancing the Power of Quantum Electronic Design Automation
No full textQuantum Computing furnishes a potential exponential acceleration compared with classical
counterparts. Many potential applications that require heavy computing resources, including
machine learning, molecule simulation, encryption algorithms, and fluid dynamic simulation,
are under exploration. However, the fundamental structure of quantum computation is still
under exploration. Currently, quantum computers have over 100 qubits available publically
with an error rate of around 1% on each quantum gate execution. Quantum algorithms, on
the other hand, are limited to many aspects and are difficult to address real-world complex
tasks. To this end, quantum Electronic Design Automation (QEDA) is introduced to assist
with the design and automation of the quantum algorithms. This approach allows users to
focus less on the intricacies of quantum algorithms and hardware and more on leveraging
quantum computing’s potential to address complex challenges effectively.
In order to address the challenges mentioned above, QEDA is leveraged as a subdomain of
quantum computing as a low-level software stack to assist with the commercialization of
quantum computers. It includes the study of quantum circuits’ simulation, transpilation,
high-level synthesis, equivalence checking, automated quantum error correction embedding,
and several security issues that may lead to privacy information leakage to the attackers. This
dissertation provides many efficient frameworks to enhance the capability of QEDA tools,
which includes 1) design and optimization of complex arithmetic quantum circuits, where it
facilitates the execution of various algorithms that require the mentioned algorithm to reduce
the communication between classical and quantum end; 2) high-level synthesis of quantum
circuits to automatically convert C code to quantum circuits, which reduce the complexity
to design quantum circuits; 3) efficient quantum circuits equivalence checking algorithms,
including formal verification and simulation-based verification, where it can verify whether
two quantum circuits are functionally equivalent; 4) timing-based side-channel attack of
quantum cloud services, where it enhances the security property of quantum computation
on cloud services.
To conclude, this dissertation builds a bridge between the application layer of different
quantum applications and the low-level quantum circuits design automation, as well as the
verification of the designs, where it facilitates the broadness of application of quantum com-
puters, and ensures the compilation process of quantum circuits design. Our contributions
push the development of QEDA to the next level, where the usage of quantum computers
worries less about the circuits and hardware level to finish the desired task and potentially
inspire many future works for more efficient QDEA toolsets
Optical Blood-spinal Cord Barrier Modulation to Enhance Intravenous Delivery to the Spinal Cord
No full textDiseases affecting the central nervous system (CNS) pose substantial costs to society due to
their impact on individual autonomy. The total lifetime cost of care for a patient with dementia
was $321,780 in 2015, with families shouldering 70% of that cost in America. Unfortunately,
delivery to the brain and spinal cord is complicated by the blood-brain and blood-spinal cord
barriers (BBB and BSCB), respectively. These structures exist within CNS vasculature to protect
against toxins within the bloodstream. However, the barriers also prevent the passage of about
98% of small molecule drugs and almost all large molecules, stymieing potential treatments for
CNS diseases.
One such disease, amyotrophic lateral sclerosis (ALS), is a disorder that induces motor neuron
degeneration and causes early death most often by respiratory failure. Mutations in the SOD1
(superoxide dismutase 1) gene can cause ALS by producing mutant SOD1 proteins, which
cannot properly reduce reactive oxygen species or protect motor neurons from oxidative stress.
Current treatments for ALS include Rilutek (riluzole), a pharmaceutical that prolongs survival by
interfering with excess glutamate causing excitotoxicity but cannot reverse motor neuron
degeneration; Radicava (edaravone), an antioxidant that counteracts oxidative stress in ALS to
slow disease progression; and Qalsody (tofersen), an intrathecally injected antisense
oligonucleotide that targets SOD1 mRNA to reduce SOD1 protein translation.
We aim to investigate needs in the field of gene therapy delivery to the CNS and ultimately
improve therapeutic delivery to the spinal cord to delay or halt ALS disease progression. We
have found that systemic administration is noninvasive, but the highly regulated blood-spinal
cord barrier (BSCB) limits current therapeutic efficacies. A popular method for barrier
modulation, focused ultrasound, has been used for transient blood-brain barrier (BBB)
modulation for systemic drug delivery, though this method may also cause tissue damage and
inflammation. It is not suited for BSCB modulation due to the irregular geometry of the spinal
bone creating standing waves and causing thermal deposition. We have developed a novel
high-resolution method by stimulating endothelial cell-targeting plasmonic nanoparticles with
ultrashort laser pulses, allowing for effective, transient, and safe modulation of blood-brain
barrier (BBB) and BSCB permeability for delivering AAVs to CNS parenchyma.
In this project, we conducted an extensive literature review to understand the current
landscape of CNS gene therapy. Using this background, we identified weaknesses in current
CNS delivery using AAVs, the most promising carriers in CNS delivery due to their unique
advantages over alternative carriers. This review informed the next steps to our experimental
investigation: applying our optoBBB modulation techniques to the BSCB. This involved
determining the degree of enhanced delivery to spinal cord tissue after pulsed picosecond
laser-mediated BSCB modulation in combination with endothelial-targeting plasmonic
nanoparticles. Due to current limitations in CNS treatments, particularly for the spinal cord, we
researched the therapeutic applicability of optoBSCB in enhancing ALS treatment, as our
method offers high spatiotemporal control, minimal invasiveness, and quick barrier recovery.
This project’s successful conclusion may then contribute to enhancing the therapeutic delivery
of intravenous agents to the spinal cord capable of affecting the disease outcome
Investigating the Impact of Organizational Resilience and Turnover in U.S. Local Governments: the Mediating Effect of Strategic Human Resources Management Practices
No full textResearch on resilience has mainly focused on crisis, or disaster management. However, more
recently the concept of resilience has emerged into management research and organization
studies, with scholars examining how organizations can prepare for the daily challenges that
come with an ever changing and complex environment, thrive and capitalize on change, and
uncertainty. Against the challenges that come with change and a complex environment,
organizations are looking for means to maintain staff performance, ensure high performance and
employee wellbeing. Resilient organizations manage the everyday stressors, and help employees
learn, adapt and bounce back from setbacks. Organizations should therefore proactively prepare
for future challenges and change, and continuously review policies and practices to create
positive and efficient work environments. What is largely missing from studies is the
examination of organizational level resilience, and particularly the role that human resource
management can play in attaining organizational outcomes. This research investigates
organizational resilience and its impact on turnover in U.S. local governments, and the mediating
effect of Strategic Human Resources Management (SHRM) practices. A survey questionnaire
was distributed to HR Directors in U.S. cities’ to determine organizational resilience, application
of strategic human resources management practices, and employee turnover rate in 2018 in
selected cities. The research adopted the Resilience Benchmark Survey developed by Resilient
Organizations (2012) which is a tool that measures organizational resilience by the leadership
and culture, networks and relationships, and change readiness. From the structural equation
(SEM) results, leadership and change readiness were negatively associated with turnover in U.S.
cities, and the mediating effect of the three SHRM practices showed both positive and negative
relationship with turnover. The findings from this research will help advance the existing
literature and improve our understanding of organizational resilience and the role of strategic
human resources practices on employee turnover. The research findings have significant
implications for HRM practitioners and researchers, for example, HR practitioners that attend to
factors that contribute to organizational resilience create an organization’s adaptive capacity for a
range of day-to-day challenges in an ever- changing environment
Silence, Society and Self-hatred: Anti-semitic Female Characters in American Literature: 1863-1947
No full textThis dissertation examines anti-semitic female characters in American literature, and how that
anti-semitism relates to each of the female characters in the works studied, differs from the male
characters, and relates to the period in which it was written. The objective of this study is to
investigate the intersectionality of anti-semitism and gender, to show that while female
characters are anti-semitic they can be so in their own way, and are not only different from the
male characters, but other female characters as well. By focusing on a few specific works I have
shown that these characters are given compelling reasons why they are anti-semitic, regardless of
wealth, income, occupation or education. This dissertation begins with a discussion of the
limited scholarship already done on the subject of anti-semitic female characters in American
literature and how existing scholarship focuses mainly on anti-semitic male characters, or anti-
semitism in general. This includes the study of authorial intent in writing a novel whose theme is
or includes anti-semitism, and why the subject is associated with the literary period in which it
was written. The resulting analysis shows that anti-semitic female characters in American
literature are just as compelling as male characters, if not more so, and allowing for the
expansion of current scholarship and the study of anti-semitism in general