Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
A Big Data Framework for Unstructured Text Processing With Applications Towards Political Science and Healthcare
No full textMachine learning and deep neural networks have soared in popularity in recent years, allowing us to enhance many aspects of everyday life. While these methods are intuitive, they are
very reliant on the dataset being used to build the model. A high-quality dataset boosts the
model’s accuracy and validates the model’s output in the context of a real-world scenario.
Furthermore, continuous improvement on the dataset contributes in the tuning of the model
in a time-consistent way and the mitigation of temporal inconsistencies. However, preparing
datasets, particularly for text domains, is difficult due to the inherent unstructured nature
of the data and the use of multiple languages. Furthermore, the amount of text produced in
the form of news articles or social media posts is massive, necessitating large-scale processing. The velocity at which new texts are produced demands an elastic and scalable system
that can accommodate any surge of inputs while remaining resource efficient while not in
use. Texts are created in a variety of ways and must be preprocessed and analyzed in order
to provide well-structured, consistent data. This can be accomplished through the use of a
well-defined domain-specific ontology (rule-based approach) or machine learning approaches.
While rule-based systems can provide information that are more precise and are preferred
in a variety of circumstances, they lack flexibility as the ontologies are often fixed and does
not respond well with the continuous changes in respective domains. We propose associated solutions to the challenges described above in this dissertation. First, we go over a scalable
architecture for collecting news stories from around the world and utilizing a rule-based approach with the Conflict and Mediation Event Observation(CAMEO) ontology to generate
political events. We present a summary of the generated dataset, as well as some basic
analysis, to demonstrate how it relates to the real-world scenario. We present techniques
to dynamically adding information to the ontology using a mining approach for discovering
new political actors that works as a recommender system and retrieves more than 80% of
the missing information including political figures and their roles. We discuss an extended
data processing system for processing articles published in several languages, with a focus
on translation methodologies and tools developed. In comparison to the English language,
we demonstrate the efficacy of the coder in Spanish. When compared to equivalent events
in English articles, the revised event coder with translated knowledge-base was able to recognize 83% of information in Spanish.
For healthcare, we propose an alternative strategy in which we use several machine learning
algorithms and social media, such as tweets, to extract the location and severity of Road
Traffic Incidents (RTI). We highlight a pipeline that goes from collecting tweets to summarizing related tweets for an RTI. We also demonstrate how semi-automatic ontology learning
can be useful in determining severity and offer a simplified example in which 100% of the
target rules were identified using an iterative technique
Active Learning for Object Detection
No full textIn this thesis, we explore the interesting paradigm of Active Learning which is garnering
a lot of attention in recent years within the Machine Learning community, and introduce
approaches to how we can leverage Active Learning strategies to acquire significant performance gain on a niche task like Object Detection. From an application perspective, we
broadly focus on two aspects -
1. Apply Active Learning strategies on the standard Object Detection task
2. Introduce targeted Active Learning for Object detection that improves performance
on rare class/rare slices of data
We will begin with a brief introduction to standard object detection mechanisms and various Active Learning strategies available and then we will dive into the experimental setup
for implementing standard Active Learning and targeted Active Learning for Object Detection, discuss the empirical results obtained from different experimental settings and finally
conclude with key observations from the experiments
Laboratory Animation Productions: Strategies to Produce Customizable Animations to Be Used as Materials for Experimental Research on Media
No full textThe interventional experimental method is a standard method for investigating the effects of
media on children. Due to budget and time constraints, researchers use the lowest priced methods
to create visual content to be used as stimuli in the research. Surveys and correspondence with
researchers have been used to examine experimental research conditions and limitations. I
studied recent animation industry technologies and developments to create a cost-effective
animation production track compatible with experimental research conditions. This thesis shows
that animations can be produced in a way to meet experimental psychology research. Due to its
use of motion capture facilities, motion libraries, ease of use, and expertise in quickly creating
simple animations, I found Adobe Character Animator to be advantageous for creating
laboratory animations. Image generating capabilities of AI programs can also accelerate the preproduction phase. It takes 80–253 hours to make 12 minutes of 2D digital animation in an
academic animation lab. The animations could then be stored in an online library to be available
to other experimental researchers without extra cost or time
Microelectromechanical-system-based Scanning Tunneling Microscopy
No full textScanning Tunneling Microscope (STM) is one of the most versatile tools in nanotechnology.
STM has made it possible to obtain three-dimensional topographic images from the surface of
conductive samples with sub-atomic resolution. With the help of STM, researchers are able
to study electronic properties of surfaces at atomic scale. Moreover, STM tip can be used
to manipulate atoms and molecules or even used as an electron-beam (e-beam) source for
patterning with atomic resolution and precision, putting atomically-precise fabrication into
practice. This superiority in precision and resolution presents STM-based lithography as a
viable alternative to conventional e-beam lithography for manufacturing the next generation
of nano-electronic devices with atomic precision.
In order for STM to have a widespread use as either an imaging or nanolithography tool,
however, its apparatus needs to be improved. The STM severely suffers from a limited
throughput due to its slow scan speed and single-tip structure. A number of STM components contribute to this shortcoming, with the nanopositioner being the main contributor.
The nanopostioners used in STMs, i.e. piezotubes, are bulky components with limited
bandwidth in all three axes. Their bandwidth is usually limited to about 1 kHz along Z axis
which is in direct interaction with sample surface and ultimately determines the scan speed.
Besides, the single-tip scheme in almost all STM systems severely inhibits the throughput.
These limitations imposed by the piezotube hinder the widespread use of STM despite all of
its unique atomic-scale applications.
To tackle this issue, we take advantage of micromachining to develop miniaturized STM
nanopositioners. In this approach, we propose one-Degree-of-Freedom (1-DOF) Microelectro-
mechanical-System (MEMS) based nanopositioners to replace the Z axis of STM piezotubes.
Thanks to micromachining, the proposed devices have less mass, and therefore, offer a higher
bandwidth compared to the piezotube, at least up to 10 kHz in this work, while keeping
the same range of motion along Z axis, i.e. 2 μm. Besides, due to a smaller footprint,
the miniaturized STMs open up the way for parallelism where an array of 1-DOF MEMS
STM nanopositioners can be placed closely to each other and engage with a sample surface
simultaneously, to increase the throughput multiple times.
To achieve these goals, we design MEMS devices to fit the tip holder of a commercial Ultra-
High-Vacuum (UHV) STM system. As a result, they can be used in place of the STM’s
regular tip, resulting a hybrid STM in which the motions in XY plane are carried out by
the piezotube while the Z-axis motion is delivered by the MEMS device, improving the Z-
axis positioning capabilities. Throughout the chapters in this dissertation, we describe the
background work, necessary design criteria, and steps for integrating the MEMS devices into
the UHV STM system. We fabricate the MEMS devices using silicon-on-insulator technologies and standard cleanroom processes, to make them suitable for batch fabrication. After
integrating the devices into the commercial UHV STM system, we then demonstrate their
capabilities for STM imaging and lithography on a hydrogen-passivated silicon sample, proving that MEMS-based STM systems are conducive to high-throughput scanning tunneling
microscopy
Practical Network Anomaly Detection: From Data Generation to Classification
No full textThroughout the Internet age, computer network-based threats have been commonplace, with
distributed denial-of-service (DDoS) attacks as a centerpiece. These attacks can knock network servers and even entire networks offline, potentially resulting in lost customers and
revenue. What once was a means-to-an-end, DDoS attacks more recently have been used in
coordination with other attacks. This threat ensemble, which we call DDoS-as-a-smokescreen
(DaaSS), leverages a DDoS attack to provide a distraction or smokescreen for another attack
or attacks. In addition to the damage a DDoS can inflict, entities targeted with a DaaSS
attack may suffer from theft of financial data or customer records. As DDoS attacks become
cheaper to launch, morphing into a service for hire, we expect DaaSS attacks to only increase
in prevalence.
Despite their prevalence, DaaSS attacks have been largely ignored by industry. One potential
reason is the nature of DaaSS attacks themselves. The attack which the DDoS is providing a
smokescreen for is often a zero-day, or something previously unknown. Detection of zero-day
threats require anomaly-based intrusion detection systems (IDS), introducing a potentially
unacceptable false positive rate. This issue is further complicated with DaaSS attacks, as
a false positive could divert IT support staff mitigating the DDoS to a non-existent threat.
Instead, signature-based IDS that are able to detect threats with no false positives are
typically used in practice, but with a trade-off of being blind to zero-day attacks.
Anomaly-based network intrusion detection systems that can detect DaaSS attacks need to
be capable of detecting DDoS and the other attack or attacks which the DDoS is providing
a smokescreen for. Our goal is to devise methodologies and techniques for anomaly-based
DaaSS detection which minimizes, preferably to zero, the false positive rate, and maintaining
this performance over time as network usage patterns shift. The detection systems should
be straightforward to train without labeled training data or hyper-parameter guesswork. We
refer to this entire process as practical network anomaly detection, to emphasize that our
methodologies and techniques can produce an IDS that is capable of practical deployment
and performance.
This process incorporates network data generation techniques, data structuring, and design
of detection systems. Network intrusion detection is inherently data-driven, but there do not
exist network datasets that capture DaaSS attacks which we are aware of. This requires us
to generate our own, and within the budget, time, and space constraints of an academic lab.
To this end, we develop a computer network traffic generation and monitoring framework
which is able to generate realistic benign network traffic along with DDoS and other attacks
as packet captures. These packet captures are then compiled into propositional network
flow datasets, using attributes and augmentation methods suitable to train and test DaaSS
detection models.
For DaaSS detection, we generalize cluster-based models to represent multiple known classes
plus an unknown anomalous class. The baseline approach is an extension to k-Prototypes
hard clustering which utilizes bounded densities to support classification of anomalies. Using
network flow augmentation to structure the data, we demonstrate that this detection model
is capable of detecting DaaSS attacks often with zero false positives, and can keep consistent
performance in the face of shifting network traffic patterns. Utilizing flow segmentation
and concurrency to structure the data, we are able to significantly decrease DDoS detection
times, potentially leading to faster detection of the other threat or threats which comprise a
DaaSS attack.
We then develop a detection model utilizing soft clustering which encodes structure of the
DaaSS threat ensemble itself, enabling model training from a small subset of benign traffic,
simplifying the data generation and collection process significantly. This detection model
outperforms the baseline, and furthermore does not incorporate any hyper-parameters, eliminating any trial-and-error involved with hyper-parameter tuning. Overall, this approach
demonstrates a means by which anomaly-based network intrusion detection may be practical for real-world use, providing practicality from data generation through model training
and performance once deployed
Stability of Softening Neural Interfaces With a-SIC Thin Film Interlayer
No full textNeural interfaces are implantable devices that enable communication between a computer and
nervous tissue to read, write and block neural activity within targeted nerves. To improve the
chronic use of neural interfaces, the materials used to develop them have been evolving with time,
leading to softer and thinner layers of the involved materials to minimize the foreign body response
from the body caused by the implanted device. Recently, researchers have studied many
biocompatible polymers that promise to extend the lifetime of neural interfaces. An emerging
materials class of interest, softening polymers (SPs), has performance advantages (while stiff and
rigid) similar to Parylene-C and Polyimide during fabrication, handling, and insertion, but after
softening (e.g. once implanted into the body), this class of polymers demonstrates enhanced
conformability. This dissertation work (1) describes the flexibility and performance as an insulator
of thiol-ene based softening polymers, (2) details a fabrication process of SP-based devices
integrating amorphous silicon carbide (a-SiC) as an encapsulation layer and (3) elucidates
structure-property-processing relationships of a-SiC SP neural interfaces via long-term electrical
stability after accelerated aging and cyclic bending for future use in chronic animal studies
Post-transcriptional Controls in Nociceptive Signaling
No full textChronic pain is a condition wherein pain continues beyond the completion of the healing
process. It is a debilitating condition that diminishes quality of life and is highly
prevalent. Persistent pain is characterized by nociceptor plasticity. Dorsal root ganglion
(DRG) neurons are responsible for generating nociceptive signals and undergo
plasticity changes following injury. These changes are intimately linked to persistent
pain. Translational regulation of mRNA permeates pain plasticity. Yet, the identity of
translationally regulated mRNA that mediates plasticity is unknown.
In this study, we used ribosome profiling to determine the protein landscape of
sensory neurons after a brief exposure to the inflammatory mediators, NGF and IL-6.
We observed preferential translation of a variety of transcripts. We focused on two
immediate early genes, Arc and c-Fos that play a role in neuronal plasticity. These
proteins have two very distinct functions: Arc regulates neuroinflammation while c-Fos
regulates neuronal excitability. We also observed ribosomal occupancy on long non-
coding RNAs as well as uORF utilization in certain mRNA transcripts. Among the
various uORF containing transcripts, we identified a novel peptide generated from a
uORF present in the 5ˊUTR of Calca mRNA. This short peptide generated from Calca is
responsible for nociceptor sensitization via Gq signaling. Finally, we also identified that
the 3’UTRs of preferentially translated mRNA contained a U-rich element for a RNA-
binding protein called HuR. This protein contributes to nociceptor firing and mechanical
hypersensitivity in mice. Our work provides insights into new key players that govern
neuronal functions
Laser Processing and Additive Manufacturing of Metallic Alloys: Laser Impact Welding, Laser Shock Peening, and Directed Energy Deposition
No full textThe objectives of this work are to develop a more accurate numerical simulation of the laser
impact welding (LIW) process, to investigate the effects of the laser shock peening (LSP)
process as a post-treatment to LIW, and to examine the effects of interlayer machining (IM)
on the microstructure and residual stress (RS) in builds manufactured via the directed energy
deposition (DED) additive manufacturing (AM) process.
First, a method for capturing the laser-generated plasma pressure load is proposed to increase
the accuracy in numerical LIW simulations. LIW is a recently developed technology for the
fully-mechanical joining of thin metallic foils. LIW is of great interest as it could be used
to join any pair of similar or dissimilar metals without any melting required, thus avoiding
the formation of brittle intermetallic compounds. However, the LIW of thin metallic foils
is a complex process, and the underlying physical phenomena involved in the mechanical
interlocking of the foils are not yet fully understood. Therefore, to benefit from its full potential in the near future, extensive research on experimental implementation and numerical
simulation of the LIW process is of paramount importance. Thus far, only a few articles
on numerical simulations of the LIW process have been published in the literature. All of
these works have made over-simplifying assumptions such as using a pre-defined deformed
flyer foil shape with a uniform initial velocity, that diminish the accuracy of the simulation.
In contrast, the research in this work proposes the idea that the incorporation of the actual
spatial and temporal profiles of the laser beam and modeling of the corresponding pressure
pulse based on an LSP approach could provide a more realistic prediction of the LIW process
mechanism. In this study, spatial and temporal profiles of an Nd:YAG laser beam pressure
pulse are experimentally characterized and fully captured for use in numerical simulations of
LIW. Both axisymmetric, arbitrary Lagrangian-Eulerian, and Eulerian dynamic explicit numerical simulations of the collision and deformation of the flyer and target foils are created.
The effect of the standoff distance between the foils on impact angle, velocity distribution,
springback, the overall shape of the deformed foils, and the weld strength in lap shear tests
are investigated. In addition, the jetting phenomenon (separation and ejection of particles at
very high velocities due to high-impact collision) and interlocking of the foils along the weld
interface are simulated. Preliminary work indicates very similar deformation and impact
behaviors in simulation results compared to experiments performed for validation.
Next, using the same laser system configuration, an experimental methodology is proposed
to investigate the effects of the extremely high strain rates present in the LSP process on
the strength and interface geometry of welds obtained from LIW of dissimilar metallic foils.
LSP is a processing technology capable of improving fatigue life and performance by inducing plastic deformations and, thus, compressive RS into the near-surface depth of metallic
components. Due to this unique ability, in recent years, LSP has been explored as a post-
treatment to improve performance in metallic welds fabricated via conventional fusion-based
techniques. However, in high-velocity impact welding (HVIW) methods, specifically LIW,
LSP has never been explored as a post-welding treatment. Therefore, in this work, LSP’s
potential to improve the weld strength and integrity in dissimilar metallic joints fabricated
via the LIW technique is investigated for the first time. Single and double LSP shots are
applied to LIW foils using three different metallic material combinations. Subsequent lap
shear testing show that single-shot LSP increases the average weld strength by 12% to 25%,
depending on the flyer and target material combination. In contrast, with double-shot LSP,
the average weld strength decreases regardless of the flyer and target materials involved.
Scanning electron microscope images reveal wavy weld interfaces and increased interlocking
between the foils for the single-shot LSP treatments as compared to the initial “flat” weld
interface geometry, thereby leading to greater flyer/target weld strength. In the double-shot
LSP treatments, however, separations and melting are observed along the weld interface
due to rebounding and excessive plastic heat dissipation of the foils. The findings of this
study reveal the first insights and effects regarding the application of LSP as a post-welding
treatment beyond conventional fusion-based welding to HVIW methods.
Last, an experimental procedure is presented to examine IM’s impact on the microstructure
and RS in metallic components manufactured via the powder-based DED AM process. DED
is one of the major additive manufacturing processes for producing and repairing large-size
and high-value metallic components. At smaller size scales (micro to millimeters), DED
could be potentially used in conjunction with LIW to manufacture small devices such as
micro/mini robots. However, IM may be necessary to provide a flat surface on the DED
build for its successful joining to metal sheets/wires via LIW. Recently, it has been shown
that the grain structure of the materials influences the in-situ LIW behavior and likely
performance. Therefore, to predict the performance of assemblies manufactured via hybrid
processes that combine DED, IM, and LIW, it is important to understand the IM effects
on the DED build’s microstructure. Therefore, in this work, for the first time, the influence
of IM on the processing-structure-properties relationships in powder-based DED of stainless
steel 316L is investigated. Four types of single-track builds are manufactured on stainless
steel 316L substrates: single-layer, double-layer, machined single-layer, and double-layer
with IM. The effects of IM on the microstructure and residual stress before and after the
second layer’s deposition are studied via metallographic imaging and neutron diffraction. In
single-layer samples, due to induced plastic strains and heat generated during the machining
process, the microstructure undergoes dynamic recrystallization, which results in smaller,
more equiaxed grains. In double-layer samples, IM results in greater tensile stresses at the
interface of the two deposited layers, where a considerable variation in the microstructure is
also observed. This is attributed to the delay caused by IM resulting in the second layer’s
deposition onto a cooler first layer and thus a higher temperature gradient. However, the
overall build height remains almost unchanged, with a slight reduction in build width. This
study’s results show that IM has important and influential effects that should be considered
in the design and control of the processing-structure-properties-performance relationships in
the DED AM
Tracking the Biochemical Activities in Cultured Cancer Cells Using Nuclear Magnetic Resonance
No full textThe human body, on average, is made up of approximately 30 to 40 trillion cells which are
divided into subgroups as organs and tissues with specific roles and functions, along with a
regular and controlled growth mechanism. Cancer forms when some of these cells mutate and
become rebellious, manifesting in uncontrollable growth and abnormally rapid proliferation. As
cancer cells multiply rapidly, there is an immediate need for new raw materials and nutrients to
sustain its hyperactive metabolic machinery. Thus, most of the biochemical pathways in cancer
are abnormally hyperactive to satisfy its voracious appetite to multiply into new cells. This Ph.D.
dissertation entails a discussion of using nuclear magnetic resonance (NMR) spectroscopy to
track the aberrant biochemical activities of cancer cells at the molecular level.
Chapter 1 of this dissertation includes an introductory discussion on the 3 cancer cell types that I
have investigated: pancreatic ductal adenocarcinoma (PDAC), colorectal cancer (CRC), and
glioblastoma multiforme (GBM) cell lines, and the experimental techniques that I used to study
these cells: NMR spectroscopy, electron spin resonance (ESR), Western blot, and the NMR
enhancing technique Overhauser effect dynamic nuclear polarization (DNP). In this chapter, I
also discussed the fundamental principles of NMR and the basic details of the other experimental
techniques.
Chapter 2 of this thesis is about the effect of the chemotherapeutic drug beta-lapachone on the
metabolism of the biochemical tracer [1,3-13C2] ethyl acetoacetate (EAA) in PDAC and CRC
cells. The main finding of this study was that the metabolism of EAA is reasonably rapid in these
cells, with acetate and beta-hydroxybutyrate as some of the metabolic byproducts. The enzyme
NQO1 converts β-lapachone into a cancer-killing reactive oxygen species; details of its effects
on EAA metabolism in PDAC and CRC cells will be discussed.
Chapter 3 discusses the use of the biochemical tracer [1-13C1]α-ketoisocaproate (KIC) to study
the hyperactivity of the branched-chain amino acid transferase (BCAT) enzyme in glioblastoma
cells. 13C NMR results revealed that 13C-labeled KIC was converted abundantly into the amino
acid leucine due to overexpressed and hyperactive BCAT enzymatic activity in SFxL
glioblastoma cells. Further study was done using a BCATc inhibitor in which 13C NMR unveiled
specific details as to the disruption of this metabolic pathway by this inhibitor ting details from
microscopy and Western blot results.
Chapter 4 details the fundamentals, instrumental setup, and preliminary results of the Overhauser
effect DNP. Herein, the mechanism of electron spin polarization transfer to the nuclear spins is
provided as well as the detailed instrumental assembly and setup of the homebuilt Overhauser
DNP machine. This DNP setup is a combination of NMR and microwave technologies in pursuit
of enhancing the NMR signals at room temperature. This project is still ongoing, but preliminary
proton NMR results are presented.
The rest of the dissertation entails technical details on the operation and data acquisition of
various instruments and techniques used in this thesis. Overall, this PhD thesis provides a
compilation of research works on tracking the abnormal biochemical activities in cancer cells
using 13C NMR spectroscopy to turn these metabolic aberrations into diagnostic advantages for
early detection and metabolic assessment of this disease
Essays on Mutual Funds
No full textThis dissertation consists of two essays on mutual funds.
The first essay, included in Chapter 1, is “Trump Tweeting Sentiment, Mutual Fund Portfolio,
and Investor Preference”. I show that online social media affects how mutual fund managers
construct their portfolios and how investors make decisions across funds. By analyzing the
tweets of Donald J. Trump, the 45th president of the United States, I find that funds following
Trump’s recommendations at industry level attract extra flows and generate superior
performance. Significant effects are observed only during Trump’s presidency and only for
positive sentiment tweets. I present evidence that investors’ political preference, their
sophistication level, the fund manager’s age and the size of the fund management, all affect fund
level response to Trump’s tweets. I also discover that Trump’s tweets have the most impact on
politically-sensitive industries including guns, energies, and utilities. This study sheds light on
how active fund managers utilize information from online social media for their portfolio
decisions, and suggests that investment decisions are highly affected by investor’s characteristics.
The second essay, included in Chapter 2, is “Rise from the Dead: An Empirical Analysis of
Mutual Fund Manager Turnover, Demotion, and Rehiring”. Using comprehensive fund manager
turnover data, I classify turnovers by the long-term past performance of their funds. Management
structure and manager experience have significant impact on fund manager replacement, as
single managers and rookie managers have less incentive to leave voluntarily, but they also face
higher risk of dismissal when funds are underperforming. I also document a spillover effect
across employments triggered by one employment dismissal. Finally, I find that performance and
experience affects short-term rehiring, while manager age affects long-term rehiring. A rehired
manager typically works for a smaller, less competitive fund, and for less pay