Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Identification of Linear Control Systems via Gradient Descent
No full textIn this dissertation we use gradient descent and its variations, in the spirit of machine learning
to identify a linear control system. When the full state is observable, the most natural least
square cost function is convex. However, when the state is partially observable, this is no
longer the case. We propose two algorithms for later problem and show that the cost function
decreases as the iteration proceeds. The simulations are provided to support that theoretical
results. We also perform recursivity analysis when the amount of data increases. Finally we
provide an asymptotic analysis when a certain natural parameter goes to infinity
Image-guided Strategies to Improve Neuroblastoma Treatment
No full textNeuroblastoma (NB) is a pediatric malignancy that accounts for 15% of cancer-related childhood
mortality. High-risk NB requires an aggressive chemoradiotherapy regimen that causes significant
off-target toxicity. In spite of these invasive treatment measures, many patients do not respond
adequately or experience relapse after initial therapy. Boosting efficacy and reducing morbidity
are therefore key goals of treatment for afflicted children. We hypothesized that these could be
achieved by developing strategies to both focus and limit toxic therapies to the region of the tumor.
One such approach is the use of targeted image-guided drug delivery (IGDD), which is growing
in popularity in personalized therapy to simultaneously improve on-target drug deposition and
assess drug pharmacodynamics in individual patients. Over the last two decades, IGDD using
focused ultrasound with “microbubble” ultrasound contrast agents (UCAs) has been increasingly
explored as a method of targeting a wide variety of diseases, including cancer. This technique,
known as sonopermeation, mechanically permeabilizes the vascular endothelium, enabling
increased penetration of drugs into target tissue. However, to date, methods of monitoring the
vascular bioeffects of sonopermeation in vivo are lacking. UCAs are excellent vascular probes in
contrast-enhanced ultrasound (CEUS) imaging and are thus uniquely suited to track the effects of
sonopermeation in tumors. Recent studies further suggest that augmenting tumor vascularity
permits enhanced drug uptake and distribution within tumor tissue. Methods of transiently
increasing tumor perfusion prior to treatment could therefore be beneficial in the treatment of this
disease. Here we report the use of gene therapy to regulate nitric oxide synthase (NOS) expression
in the tumor vasculature. NOS catalyzes the chemical reaction that generates nitric oxide (NO), a
potent endogenous vasodilator. Our goal was to rationally design a state-of-the-art, non-viral
platform to efficiently deliver NOS-expressing plasmids to cells lining the tumor blood
vessels. To construct this gene delivery vehicle, we utilized cationic UCAs to carry plasmid DNA
(pDNA) in circulation and transfect tumor vascular endothelial cells in vivo using focused
ultrasound (FUS) energy. Our results suggest that significant drug uptake occurs by improving
tumor vascular permeability with microbubble sonopermeation without damaging the vasculature.
We have thus conceived a clinically viable methodology for improving neuroblastoma response
to treatment by successfully effecting increases in tumor perfusion volume and tumoral blood flow
rates
Customer-centric Pricing: Maximizing Revenue Through Understanding Customer Behavior
No full textAdvances in online retail infrastructure allows retailers and researchers to keep track of various attributes including customer demographics and online behaviors on the website. The
thesis studies the gaps between theoretical results and empirical data that can be attributed
to human behavior when faced with pricing and operational situations. In particular, we
study the effect of anticipated regret on purchase decisions of customers who face a markdown
pricing structure. In subsequent study, we assess the impact of a retailer’s frequent price adjustments on customers’ long term expenditure in the retailer. By this, we aim to fill the gap
in the revenue management literature that mainly focuses on retailer’s pricing optimization
problem and thus, leaves room for deeper understanding of customer’s perception towards
frequent price changes. I identify these behavioral motives that have previously received less
attention from conventional pricing and operations management literature, carefully model
them using probabilistic and/or statistical modeling methods, and estimate and quantify the
effects to operationalize these factors
CRISPR-Cas Systems and Bacteriophages: Alternative Therapies to Combat Antibiotic-resistant Enterococcus Faecalis
No full textThe increasing demand for antibiotics has created selective pressure for the emergence of
multidrug-resistant (MDR) pathogens. Enterococcus faecalis is normally found as a commensal
of the gastrointestinal (GI) tract of healthy humans. However, E. faecalis is also an opportunistic
pathogen and a leading cause of hospital-acquired infections (HAI). E. faecalis intrinsic and
acquired antibiotic resistance can make these infections very difficult to treat. Genomic analyses
have revealed that E. faecalis strains may become more virulent through horizontal gene transfer
(HGT) of mobile genetic elements (MGEs), such as the highly transmissible E. faecalis
pheromone-responsive plasmids (PRPs). MGEs can also decrease cellular fitness. Therefore,
bacteria encode many defense systems to limit their transfer. A well-studied defense mechanism
is the adaptive defense system CRISPR-Cas. In addition to the rise of MDR pathogens, the
development of new antimicrobial drugs has been limited, pushing modern medicine toward a
post-antibiotic era. The void created by the limited number of antibiotics that are still effective
against infections has led to exploring alternative therapies. Among them are CRISPR-Cas-based
antimicrobials. Multiple studies, including one in the Palmer lab, have established that CRISPR-
Cas can effectively remove antibiotic resistance from a bacterial population in a sequence-specific
manner. Most of these studies have used model strains; therefore, there is limited understanding
of how effective CRISPR-Cas antimicrobials are against non-model strains. Another proposed
alternative is bacteriophage therapy. Studies on the molecular mechanism involved in enterococcal
phage infection and phage resistance are limited. My research aimed to establish the efficacy of
CRISPR-Cas antimicrobials against non-model strains, which are the intended target for CRISPR-
based therapy, and to further elucidate the mechanisms involved in E. faecalis phage infection and
the host genomic alterations to become phage-resistant. Conjugation assays were used to compare
the efficacy of our previously engineered plasmid-based CRISPR-Cas antimicrobials against a
recent collection of E. faecalis fecal isolates, referred to here as “wild” isolates. It was discovered
that the wild isolates could i) antagonize the CRISPR-Cas antimicrobial donor strain via
competitive factors and ii) prevent CRISPR-Cas antimicrobial transfer, effectively avoiding
CRISPR-Cas targeting. Additionally, a 14-day coevolution study was performed using the E.
faecalis strain SF28073 and two genetically distinct E. faecalis phages, followed by whole-genome
sequencing to elucidate novel mutations resulting from the pressure imposed by phage infection.
The results revealed mutations in genes encoding macromolecules that may be associated with
phage infection, many of them previously unreported. Results from the first study emphasize the
need to continue studying CRISPR-Cas antimicrobials in the context of wild isolates to determine
potential limitations. The second study serves as a basis for the continued research of enterococci-
phage coevolution. Both studies are critical to developing viable CRISPR-Cas and phage therapies
Investigating Misconception Resolution and Learning From Scientific Texts Using a Cognitive Diagnostic Model
No full textResearch shows that refutation texts are more effective than expository texts at helping
students resolve science misconceptions. However, little work has been conducted on the
effects of related thought experiment discourse structures. To empirically investigate the
effects of thought experiments on student science learning, high and low prior knowledge
college students read refutation, expository, and thought experiment physics texts. Their
understanding was assessed using the Force Concept Inventory (FCI). Differences were evaluated using correct and misconception responses in ANOVA analyses. In addition, FCI
response data was analyzed using a novel cognitive diagnostic model of skills and misconceptions. Effects of prior knowledge on learning only were observed in the ANOVA analyses.
However, the item-specific probability parameters from the cognitive diagnostic model analysis indicated that both high and low prior knowledge students learned more from thought
experiment texts than expository texts for one of three topics that were presented. In addition, these parameters indicated that misconception possession was more prevalent among
low prior knowledge participants reading thought experiment texts for this topic as well.
Person-specific skill mastery probabilities indicated that high and low prior knowledge students learned the least from expository texts for a different topic but that there was little
difference in misconception resolution between discourse structures. The implications of
these results for improving student learning using various discourse structures are discussed
Polymeric Encapsulation of Vaccines for Enhanced Immunogenicity
No full textVaccines have been used for hundreds of years to provide protection against infectious diseases.
Early vaccine formulations provided strong protection as they were highly immunogenic but were
hindered by harsh associated side effects. In an attempt to improve the safety profiles of vaccines,
isolated components of pathogens were used instead of live-attenuated or whole-cell formulations.
These subunit vaccines were much more tolerable but were unfortunately less effective at
providing protection. To overcome this, several methods to adjuvant vaccines have been
investigated. One method is to improve cellular uptake of subunit antigens by the components of
the innate immune system, specifically antigen presenting cells (APCs) whereas another is to
prolong exposure of the antigen to the immune system via an antigen depot. Both of these
mechanisms have been effective in inducing long-lasting immunity via activation of T-cells and
B-cells. Herein a class of polymeric scaffolds, metal-organic frameworks (MOFs), have been
employed to improve both APC uptake of model vaccines and provide an antigen depot. This has
been shown to improve the immunogenicity of immunoadjuvants, subunit antigens, and wholecell vaccines
Towards Trustworthy Machine Learning
No full textIn an era marked by ubiquitous machine learning (ML) applications, the question of trust has
risen to the forefront of concern. From financial institutions using ML for credit risk modeling
to the rapid adoption of Large Language Models (LLMs) such as ChatGPT, the reliance on
ML systems has become a defining characteristic of our lives. However, as ML’s influence
grows, so do the implications of trust; affecting individuals, organizations, and society at
large. Trustworthiness in ML transcends mere performance; it involves the intricate balance
of privacy, fairness, robustness, and more. This dissertation addresses these core issues,
aiming to enhance the trustworthiness and reliability of ML applications.
Privacy is a paramount concern in the context of LLMs. Concretely, LLMs possess the ability
to memorize segments of their training data, and can reproduce memorized content when
given appropriate prompts. This becomes particularly significant when models are trained
on data containing sensitive and private information. In our work, we show how we can
discover prompts capable of eliciting memorized content from LLMs which corresponds to a
data extraction attack. Additionally, by deriving valuable insights from our attack, we create
a defense mechanism that reduces the chances of an LLM generating memorized content.
Our defense is efficient as it does not need re-training of models, and offers adjustable
privacy-utility trade-offs.
Fairness in ML is another critical area as models increasingly inform decision-making processes. Current research highlights that, models may amplify and propagate societal biases
encoded in data. Most fairness-focused techniques rely on extensive demographic data, which
is not always accessible due to privacy concerns. We introduce fair training methods that
works effective with limited demographic information. Our findings suggest that, with demographic attributes for just 0.1% of the training data, we can still achieve competitive
fairness-utility trade-offs.
Further, we investigate robustness of Federated Learning (FL) against backdoor attacks.
FL allows a set of agents (e.g., hospitals) to collaboratively train a model without sharing
their potentially sensitive data. However, the decentralized and unvetted data makes FL
particularly susceptible to attacks. In this dissertation, we propose a lightweight defense
against backdoor attacks in FL. Empirical evidence supports the effectiveness of our defense
against backdoor attacks under various settings, outperforming existing defenses.
Finally, we explore how variations in local data distributions affect the fairness and robustness properties of models trained with FL. The existing literature shows that, as local data
distributions differ, accuracy of the trained models drop. Our findings reveal that, robustness and fairness might degrade much faster than accuracy. All in all, we reveal that, small
variations that have little impact on the accuracy could still be important if the trained
model is to be deployed in a fairness and security critical context
ANN Crowds: Harnessing Collective Wisdom in Design Prediction
No full textThis thesis presents and evaluates an approach to early-stage product performance prediction by harnessing the "wisdom of the crowd" embodied in Artificial Neural Network (ANN) Crowds. Unlike traditional crowd wisdom, which leverages responses from large groups of people, this research explores the concept of using 189 distinct ANN architectures, each replicated 100 times, as a collective decision-making entity, thus forming an ANN Crowd.
The central inquiry in this study revolves around the notion of whether every agent within the ANN Crowd should possess an equal influence. To address this question, the research conducts a comprehensive exploration of the sensitivity of key influencing factors, including training set selection, the configuration of nodes and layers, and architectural attributes, on the performance of the ANN Crowd. This investigation aims to refine the ANN Crowd, making it more universally applicable.
The first aspect explores the impact of training set selection in predictive accuracy. The results clearly demonstrate that training set selection has a statistical and practical influence prediction accuracy, especially when it includes edge cases. This finding provides a crucial guideline for
decision-makers, advocating for the strategic inclusion of challenging examples in training sets to improve predictive accuracy.
The second facet investigates the complex interplay of architectural attributes within the ANN Crowd. By categorizing architectural attributes based on Normality, Centrality, and Width, the analysis shows that the number of nodes within architectural configurations does not have a statistically significant impact on prediction accuracy. This challenges the conventional belief that complexity leads to improved performance, providing practical insights for architectural design.
In conclusion, this research offers valuable insights into the predictive capabilities of ANN Crowds. It extends practical implications to engineering design and decision-making processes, positioning ANN Crowds as a vital tool across diverse industries. The findings and guidelines provide a foundation for data-driven practices, enhancing efficiency, and adding value to businesses. Acknowledging its limitations, this research paves the way for future work, encompassing a broader range of datasets, architectural factors, and validation studies
Phase Transition of Community Detection Under Efficient Algorithms, Expressive Generative Models, and Confidentiality Constraints
No full textWe formulate a semi-definite relaxation for the maximum likelihood estimation of node
labels, subject to observing both graph and non-graph data. This formulation is distinct
from the semidefinite programming solution of standard community detection, but maintains
its desirable properties. We calculate the exact recovery threshold for three types of non-
graph information, which are called side information: partially revealed labels, noisy labels,
as well as multiple observations (features) per node with arbitrary but finite cardinality. We
find that semidefinite programming has the same exact recovery threshold in the presence
of side information as maximum likelihood with side information.
Empirical observations suggest that in practice, community membership does not completely
explain the dependency between the edges of an observation graph. The residual dependence
of the graph edges are modeled in this dissertation, to first order, by auxiliary node latent
variables that affect the statistics of the graph edges but carry no information about the
communities of interest. We then study community detection in graphs obeying the stochastic block model and censored block model with auxiliary latent variables. We analyze the
conditions for exact recovery when these auxiliary latent variables are unknown, representing unknown nuisance parameters or model mismatch. We also analyze exact recovery when
these secondary latent variables have been either fully or partially revealed. Finally, we
propose a semidefinite programming algorithm for recovering the desired labels when the
secondary labels are either known or unknown. We show that exact recovery is possible
by semidefinite programming down to the respective maximum likelihood exact recovery
threshold.
Releasing graph structures containing nodes with multiple latent variables might cause privacy issues and confidential information leakage of the users. This dissertation investigates
the confidentiality in community detection in networks with multiple latent variables. Focusing on stochastic block model and censored block model with multiple latent variables,
we address the leakage of confidential information by changing the connectivity of nodes. To
this end, we first propose a new metric for evaluation of confidentiality based on Chernoff-
Hellinger divergence. An optimization is introduced to minimize the required changes on
the edges of the graph realization
Between God and Homeland: the Diaries of Victor Klemperer and Willy Cohn
No full textThis dissertation examines the self-identities of two German-Jewish academics through the lenses of
faith and homeland. Victor Klemperer and Willy Cohn transmitted their thoughts and feelings
through their extensive diaries kept during the period of the Third Reich. As Nazi Germany
progressed from virulent antisemitism, to eliminationist antisemitism, to mass murder, Klemperer,
and Cohn, recorded their experiences under radical assault by a regime determined to make
Germany Judenrein, cleansed of Jews. However, the Nazi project did not seek to only make Germany
free of Jews, it sought to erase Jews and Judaism, past, present, and future, far beyond the borders of
the Fatherland. The assault on Jews encompassed the religious, social, cultural, economic, and
political spheres. No measure was too draconian, nor too petty, if it added to the marginalization
and demonization of Jews. Each diarist had his own view of religion, while both men were fully
committed to their German heritage, a heritage encompassed within the German word heimat
(homeland).
In eleven chapters, I examine Klemperer’s and Cohn’s words for the fracturing of each man’s self-
identity as they attempted to navigate a world in which the ground under their feet was constantly
shifting: attempting to live one more day, one more week, one more month, under a regime that
intentionally pushed Jews to the limits of despair. As Nazi Germany progressed from virulent
antisemitism to eliminationist antisemitism and ultimately mass murder, Klemperer and Cohn
recorded their experiences under assault by a regime determined to make Germany Judenrein—
cleansed of Jews.
Both diarists struggled with their German-Jewish self-identity, but in different ways. Both
Klemperer and Cohn were also deeply committed to German nationalism, but each man had his
own perspective on his “Germanness” as well. Klemperer’s was more culturally and socially
oriented, while Cohn’s conservative nationalism turned on his love of Germany as both political
entity and heimat. Klemperer and Cohn most contrast one another in their religious beliefs. While
Cohn was a deeply committed and observant Orthodox Jew, Klemperer, who converted to
Protestantism, rarely mentions God, or his religious beliefs in his diary.
The parallels and contrasts found in the diaries of these two learned academics underscore the
complex nature of the German-Jewish population in the Third Reich, revealing the German Jews
that confronted the Nazi threat were no monolithic block of people acting in concert against the
existential assault on Judaism, but instead were internally torn by their love for their country and
their hatred of the regime and its antisemitic actions. As with the broader population of German
Jews, Victor Klemperer and Willy Cohn confronted Nazism without foreknowledge of the tragic
events that unfolded before them. As part of their response to a heretofore unthinkably murderous
assault, they wrote