Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Modeling Economic Mobility: a Machine Learning Approach
No full textThis thesis constructs and evaluates models of increasing complexity that aim to predict a
child’s income rank in adulthood based on the characteristics of the neighborhood in which
they grew up. Each model was trained on newly-released public data with estimates of outcomes for children based on their Census tract. A set of simple decision tree models provided
substantial performance in the prediction task, even with highly restricted depth. These decision trees, and a basic neural network model, were each restricted by an inability to handle
missing features in the input samples. To handle this restriction, a joint neural network
statistical model was created that utilizes learned distributions for each feature to provide
estimates of each missing value to the neural network, prior to prediction. The joint neural
network statistical model achieved a lower error rate than the more basic models. Assessing the features selected by the decision tree learning algorithm and comparing the relative
importance of features in the neural network through sensitivity analysis demonstrated that
each model relied on similar characteristics
Preparation of Functionalized Graphenes and Their Performance for Corrosion Resistance and Synthesis of Vanadium Nitridecarbon Nanofiber Mats and Their Application for Asymmetric Supercapacitor Electrodes
No full textCarbon materials have been researched in a wide range of applications, including energy storage,
corrosion protection, catalyst support, etc. Though many people have studied properties of
carbon with/without modification, research in the synthesis of functionalized carbon and carbon
composites therefrom will be important. This dissertation outlines the synthesis of functionalized
graphene, development of graphene composites, and metal nitride-carbon nanofiber composites
and their application for electrode for super-capacitor.
Graphene is single layer carbon-based material known for its multifunctional properties (e.g.
hydrophobicity, mechanical strength, etc.). This work studies application of graphene through
functionalization of mechanically exfoliated graphene and functionalized graphene use for
corrosion resistance. Functionalization of graphene is first discussed in the eco-friendly synthesis
of aminated mechanically exfoliated graphene (MEG). Here we are able to directly functionalize
MEG with amine groups using glycerol as the reaction medium and urea as the amine source.
The aminated MEG (AG) also showed enhanced dispersion stability in organic solvents and
aquous-solvent mixtures for >60 days. In the second area, we discussed how functionalized
graphenes affected surface properties and corrosion resistance when composited with a polymer
coating. Graphene, aminated graphene (AG), and fluorinated graphene (FG) were studied for
corrosion resistance. Both AG and FG exhibited enhanced corrosion resistance when composited
with a 2K urethane coating. Composite coating with FG showed a 94% increase in corrosion
resistance versus graphene at a concentration of 4% by weight of solids. These materials also
enhanced the surface properties of the coating. Electrochemical analysis of composite coatings
showed that through inclusion of functionalized graphenes (AG or FG), barrier and adhesion
properties were strengthened. Both FG and AG composite coatings showed an increase in
contact angle versus graphene, with FG resulting in a hydrophobic surface (>90o
) at 4wt%. This
project shows a step towards the potential removal of sacrificial zinc as a barrier for corrosion
resistance of steel substrate.
The increase in energy consumption over the last decade has led to research in the development
of alternative energy storage devices which can meet the demand. Supercapacitors have garnered
increased attention in this field due to their ability to provide high power and high energy.
Electrodes within these devices can consist of two different materials, carbon or
pseudocapacitive material. While carbon-based supercapacitors, or EDLCs, can provide high
power density, they suffer from low energy densities. This has led researchers to study
composite, or hybrid electrodes which combine the high power EDLC material with a high
energy pseudocapacitive material (e.g. metal oxide or metal nitride). In the third area, we
assembled and tested hybrid-asymmetric devices using activated vanadium nitride-carbon
nanofiber (VN-CNF) electrodes. The VN was made using vanadium oxide (V2O5) nanoflowers
by a new synthesis method. Composite electrodes were made by electrospinning of a
poly(acrylonitrile-co-itaconic acid) (PANIA) solution with the vanadium oxide (V2O5)
nanoflowers dispersed within it to produce freestanding mats. VN-CNF freestanding mats were
used as anode material and CNF as the cathode when assembling the device. Ionic liquid
electrolyte 1-butyl-1-methylpyrrolidinium bis(trifluoromethanesulfonyl)imide (Pyr14TFSI) with
0.5M lithium bis(trifluoromethanesulfonyl)imide (LiTFSI) was used, which widens the operating
voltage window (>3.5V) compared with aqueous (e.g. KOH or Na2SO4) or organic electrolytes
(e.g. TEA-BF4 in ACN).
Chapter 1 introduces the material graphene, its properties, and methods of synthesis. In this
section we also review methods of functionalization and application in the field of corrosion
protection.
Chapter 2 describes the eco-friendly method of functionalization of mechanically exfoliated
graphene (MEG) with amine groups and the effect of these groups on dispersion stability in
organic solvents and aqueous-solvent mixtures are studied.
Chapter 3 studies the effect of graphene, aminated graphene (AG), and fluorinated graphene
(FG) on surface properties and corrosion resistance when composited with a urethane coating.
Contact angle measurements and electrochemical analysis were performed to determine which
graphene and concentration provides best corrosion resistance.
Chapter 4 introduces supercapacitors and the concepts behind the different types. This section
describes charge storage mechanisms of the pseudocapacitors and discusses different aspects of
metal oxides and nitrides.
Chapter 5 describes the preparation of vanadium nitride-carbon nanofiber (VN-CNF) composite
electrodes and analysis of their electrochemical properties. Charge contributions and storage
mechanisms for each sample was studied to understand the mechanism in which their charge is
stored (e.g. intercalative or pseudocapacitive/capacitive)
3D Printing Tough, Isotropic, and Sustainable Polythiourethanes for Fused Filament Fabrication
No full text3D printing, especially fused filament fabrication (FFF), has captivated the imagination of
thinkers, hackers, doers, makers and manufacturers alike for its ability to rapidly convert digital
designs into tangible objects.. Many industries such as biomedical, aerospace, and automotive
have been early adopters of the technology because of FFF's ability to print complex geometries,
its cost-saving ability for custom parts, prototypes and low lot manufacturing runs, and its
potential to be a more sustainable manufacturing practice. However, despite these benefits, FFF
is still limited to prototyping and non-functional parts in these industries because of its limited
selection of engineering-grade filaments and its typically weak mechanical properties. It also
shows limited ability to be recycled and reused multiple times, similar to other plastics. Past
research has explored various ways to widen the material selection, improve the mechanical
properties of FFF filament, and make it more recyclable, but research has yet to show a material
with these properties in an easy-to-use filament. This work leverages a thiol-isocyanate click
chemistry that produces tough, isotropic, and sustainable FFF filaments. These filaments are
comparable to many polymers used in traditional manufacturing, and widely outperform
common 3D printed materials in all print directions. This filament also retains its mechanical
properties when recycled multiple times without significant changes to the polymer's mechanical
or materials properties. While further development is needed for FFF printing to flourish across
all industries, this work shows the potential of polythiourethane filaments to offer solutions to
some of the most significant drawbacks of FFF technology
Do Women Directors Improve Firm Performance? Evidence From India
No full textPrior studies have documented, in various countries, negative effects of gender quota laws on firm
value, suggesting that such quotas are detrimental for firm outcomes. In this paper, I revisit this
question using gender quota laws in India. The Companies Act, 2013 required all listed firms and
public firms that exceed a size threshold to mandatorily appoint at least one woman director on the
board. The rule was first applicable for fiscal year 2015. In contrast to prior studies, I document a
positive effect of quota legislation in India. I find that the cross-section of the firms that appointed
non-promoter women directors or did not belong to a business group, have a significant
improvement in operating performance, relative to the control group. Additional cross-sectional
analyses show that firms that have relatively lower bias against women or face lower supply
constraints, experience a significant positive effect from the quota legislation. Overall, my results
suggest that the quota legislation in India had positive effects on firm performance
Multiscale Modeling of Dislocation and Grain Boundary Mechanics in Small Scale Metals
No full textMetals are of great importance for structural applications due to their high yield strength and fracture toughness. In recent years, efforts have been undertaken to further improve these properties,
accelerated by advances in materials research and manufacturing processes. The conventional
strategy to achieve high strength is to reduce the average grain size, but this is inevitably followed
by the loss of ductility. Deformation mechanisms for plastic flow and ductility are largely dependent on microscopic defects such as dislocations, grain boundaries (GBs), and triple junctions
(TJs). It is necessary to obtain a fundamental understanding of the correlation between defect mechanics and macroscopic properties across a variety of time and length scales so as to overcome
the strength-ductility trade-off. With this motivation, a computational and theoretical approach
has been taken to investigate the complex interplay between defects and macroscopic material
response.
In the first part of this dissertation (Chapters 2-3), dislocation mechanics within single crystals are
examined to understand the role of sample size, crystallographic orientation, and loading conditions on the mechanism response. The focus is drawn to the plastic deformation which occurs at
the mesoscale, wherefrom material properties are determined.
Chapter 2 reports on DD simulations conducted to examine plastic deformation in single crystalline
Cu micropillars subjected to two types of combined loading conditions: tension after torsion and
torsion after tension. These combined loadings are then compared with simple tension and pure
torsion, respectively. In metallic materials, the activation of one slip system increases the flow
strength of other slip systems, which is a phenomenon known as latent hardening. This latent
hardening behavior has been understood by the “forest hardening” mechanism arising from mutual dislocation interactions at the continuum length scale. As the size of a sample decreases to the
submicron scale, the interactions between dislocations become increasingly sparse, so plastic deformation is instead governed mainly by dislocation sources. We find that there exists a transition
from latent hardening to latent softening in intermediately-sized 600 nm samples undergoing the
combined tension after torsion loading. The systematic computational and theoretical model described here suggests explosive multiplication causes dislocation density to greatly increase, giving
rise to latent softening in those micropillars under tension after torsion.
At the continuum length scale, mechanical properties of metals show relatively weak orientation
dependence; however, Chapter 3 shows how strong anisotropic behaviors are exhibited as the size
of sample decreases to micron and nanometer length scales. DD simulations are performed to investigate the orientation-dependent plasticity in submicron face-centered cubic (FCC) micropillars
subjected to torsion. Accommodating results from atomistic modeling, updated surface nucleation
schemes in DD models have been developed for three orientations ([001], [101], and [111]), allowing investigation of the dislocation microstructure evolution and the corresponding anisotropic
mechanical response upon torsional loading and unloading. The DD simulation results show that
the coaxial and hexagonal dislocation networks formed in [101]- and [111]-oriented nanopillars, respectively, exhibited excellent plastic recovery, while the rectangular dislocation network formed
in the [001] crystal orientation was more stable and did not experience as much plastic recovery.
Following work on isolated dislocation mechanics within a single crystal, the second part of this
dissertation, Chapter 4, transitions into the exploration of defect mechanisms within bicrystals.
Mechanical properties of metals such as strength and toughness are strongly correlated to complex
interactions between various defects in the crystalline structure. While elementary interactions
between these defects have been investigated using recent micro- and nano-characterization techniques, understanding of the detailed interaction mechanisms has hardly been obtained. To model
plasticity in polycrystals at larger time and length scales, it is necessary to formulate a general
guideline to predict both the interaction type (transmission or reflection) and the dislocation’s subsequent slip system after the interaction. Many criteria based on the geometric alignment of the
defects have been developed to predict this phenomenon, but these have not been found to be
accurate when applied to general data sets of grain boundaries (GBs). With this motivation, we
conduct a systematic study using molecular dynamics (MD) models of bicrystals to analyze defect
interaction process between a prismatic dislocation loop and eleven different grain boundaries of
the following character: three tilt, three twist, and five mixed. Based on the MD observations,
two new prediction methods are developed: the first is a new data-driven parametric score function
based on the classical geometric criteria, and the second is by applying Gaussian process machine
learning methods to find the probability distribution of a hidden function. The proposed methods
could pave a new way to predict the unit interaction of dislocation with various GBs, which could
show much higher accuracy compared to pre-existing geometric criteria.
Finally, additional work on paving the way to polycrystalline modeling at the mesoscale is detailed,
followed by an overall summary in Chapter 5
Multistream Mining and Its Applications
No full textDevelopment of Internet technology has created a lot of stream data in our daily life. Streams
of data may come from a variety of sources, such as online shopping, social media, smart sensors, transportation, and etc. Data streams has some unique properties, such as fast changing,
time stamped, numerous, and potentially infinite. Hence, traditional machine learning approaches
have limited ability to handle the whole data stream. Since data stream usually has tremendous
volume. When we consider a supervised learning scenario, the availability of sufficient labeled
data is the most important thing. However, collecting ground truth data is time consuming and
very expensive in many real world applications. Particularly when performing prediction over a
non-stationary data stream, limited labeled data affects the classifier’s long-term performance by
limiting its adaptability to changes in the data distribution along with time. In this dissertation,
the approaches we propose to solve this problem can be divided into two directions: Multistream
Classification and Multistream Regression. For the first direction, we propose a transfer learning framework which address the covariate shift and concept drift challenges over a data stream
setting. We consider two independent non-stationary streams. One stream contains labeled data,
called source stream. The other stream contains unlabeled data, called target stream. Data instances in source stream have a biased distribution compared to data instances in target stream.
Label prediction task under above scenario is called Multistream Classification. In this task, data
instances in source stream and target stream occur independently. While previous studies have
addressed various challenges in this multistream setting, it still suffers from large computational
overhead mainly due to frequently employed bias correction and drift adaptation methods. In this
dissertation, we propose a multistream framework called MSCRDR. In MSCRDR, we focus on
utilizing an alternative bias correction technique, called relative density-ratio estimation, which is
known to be computationally faster. Importantly, we propose a novel mechanism to automatically
learn an appropriate mixture of relative density that adapts to changes in the multistream setting
along with time. In addition, we theoretically study its properties and empirically illustrate its
superior performance.
We extend our research in Multistream Regression task. In this dissertation, we propose a multistream regression framework which unify concept drift detection and covariate shift adaptation. We use multiple real world datasets and synthetic datasets to evaluate the performence of our proposed
framework. We use multiple state-of-the-art approached in our experiments. Empirical evaluation
results indicates the effectiveness of our proposed framework. We demonstrate our approaches by
several applications: game cheating application, flight delay project and Covid-19 spread analysis
The Averaging Problem in Cosmology and Macroscopic Gravity
No full textThe dynamics of the universe are traditionally modelled by employing cosmological so-
lutions to the Einstein field equations. In these solutions, the matter distribution is taken
to be averaged over cosmological scales, and hence, the Einstein tensor needs to be av-
eraged as well. To construct such an averaged theory of gravity, one needs a covariant
averaging procedure for tensor fields. Macroscopic gravity (MG) is one such theory. It
gives the macroscopic Einstein field equations (mEFEs) where the effects due to averag-
ing are encapsulated in a correction term, the so-called back-reaction. This additional term
accounts for the non-commutativity between the averaging operation and the calculation
of Einstein tensor.
In this dissertation, we analyse how to deal with inhomogeneities within macroscopic
gravity. First, we model the inhomogeneities as linear perturbations around the spa-
tially homogeneous Friedmann-Robertson-Lemaître-Walker (FLRW) geometry. Then, we
analyse exact inhomogeneous models with plane and spherically symmetric geometries.
We calculate the back-reaction in these models and analyse how it modifies observations
done within them.
First, we explore the application of the MG formalism to an almost-FLRW model. Namely,
we find solutions to the field equations of MG taking the averaged universe to be almost-
FLRW modelled using a linearly perturbed FLRW metric. We study several solutions with
different functional forms of the metric perturbations including plane waves ansatzes.
We find that back-reaction terms are present not only at the background level but also at
perturbed level, reflecting the non-linear nature of the averaging process.
To analyse how observations get modified by the back-reaction, we derive the expres-
sions for distance measures in MG. We analyse two cases. In the first one, the back-
reaction modifies distances only through the expansion history. In the second one, the
back-reaction density parameter enters the distance formulae in such a way that, phe-
nomenologically, it is degenerate with a spatial curvature. Turning to the perturbations,
we derive an equation for growth of structure and analyse how back-reaction modifies
the linear growth rate. Thus, the averaging effect can extend to both the expansion and
the growth of structure in the universe.
Then, we turn our attention to inhomogeneous models with plane and spherically sym-
metric geometries. We calculate the MG correction term for such models and find that
it takes the form of an anisotropic fluid with a qualitative behaviour of an effective cur-
vature in the field equations. We categorise the solutions according to the source for the
space-time – vacuum, dust and perfect fluid. Within these three categories, we treat, in
detail, the cases of the static spherically symmetric vacuum solution (Schwarzschild exte-
rior), the static spherically symmetric perfect fluid solutions (Schwarzschild interior and
Tolman VII) and the non-static spherically symmetric dust solution (Lemaître-Tolman-
Bondi (LTB)). This is a first step towards analysing back-reaction in inhomogeneous cos-
mology with MG
Multidimensional Uncertainty Quantification for Deep Neural Networks
No full textDeep neural networks (DNNs) have received tremendous attention and achieved great success in various applications, such as image and video analysis, natural language processing,
recommendation systems, and drug discovery. However, inherent uncertainties derived from
different root causes have been realized as serious hurdles for DNNs to find robust and trustworthy solutions for real-world problems. A lack of consideration of such uncertainties may
lead to unnecessary risk. For example, a self-driving autonomous car can misdetect a human
on the road. A deep learning-based medical assistant may misdiagnose cancer as a benign
tumor.
In this work, we study how to measure different uncertainty causes for DNNs and use them
to solve diverse decision-making problems more effectively. In the first part of this thesis,
we develop a general learning framework to quantify multiple types of uncertainties caused
by different root causes, such as vacuity (i.e., uncertainty due to a lack of evidence) and
dissonance (i.e., uncertainty due to conflicting evidence), for graph neural networks. We
provide a theoretical analysis of the relationships between different uncertainty types. We
further demonstrate that dissonance is most effective for misclassification detection and
vacuity is most effective for Out-of-Distribution (OOD) detection. In the second part of the
thesis, we study the significant impact of OOD objects on semi-supervised learning (SSL) for
DNNs and develop a novel framework to improve the robustness of existing SSL algorithms
against OODs. In the last part of the thesis, we create a general learning framework to
quantity multiple uncertainty types for multi-label temporal neural networks. We further
develop novel uncertainty fusion operators to quantify the fused uncertainty of a subsequence
for early event detection
Hardware-based Malware Detection in Modern Microprocessors: Formal and Statistical Methods in System-level Security Assurance
No full textPrevious studies in workload forensics have relied on retrospective assessments using comprehensive process execution profiles, hindering the ability to take prompt action against
ongoing cyber threats. In this dissertation, we put forth a hardware-centric approach for
real-time workload forensics, enabling the identification of processes during their execution.
We present a universal framework that formalizes real-time forensic analysis and malware
detection procedures, incorporating hardware-level feature extraction and machine learning-
based data analysis. To showcase the effectiveness of our proposed framework, we explore
hardware-based workload forensics and hardware-based Spectre attack detection. Our experimental findings indicate that our system can successfully identify Spectre attacks across
thirteen intentionally vulnerable victim code patterns. Beyond machine learning techniques,
we also utilize formal analysis to ensure the secure execution of machine-level binaries on
specific hardware configurations. This method offers a more extensive coverage of the state
space compared to verification techniques dependent on testbenches, potentially encompassing the entire state space
Improve Planning Efficiency by Problem Reformulation to Facilitate Automated Service Composition
No full textThe Internet of Things (IoT) is an emerging paradigm where practically everything, including both physical and cyber things, is interconnected via the Internet to offer a wide
spectrum of physical and cyber services. Traditionally, atomic services are composed to
create more complex and value-added business processes. However, since many tasks in the
IoT world arise dynamically, IoT services also need to be composed dynamically. To deal
with this dynamic requirement, service composition in IoT should be automated to reduce
human effort, especially for complicated real-world problems. In the literature, AI planning
techniques have been widely applied to automate service composition. However, some major challenges still exist in automated IoT service composition research. First, IoT systems
involve physical services which are quite different from software services, and a modified service model is required to deal with the mix of physical and software services. Also, to apply
AI planning to automate service composition, an automated mapping framework is essential for converting service composition problems into planning problems. Existing mapping
techniques neither consider physical services nor QoS-related properties and constraints on
physical objects, while these missing elements add significant complexity to the framework
and demand advanced research. The second challenge is the scalability of the planning algorithms. The vast collections of physical things and their services result in huge action and
state sets in the planning problems, making it infeasible to derive solutions in a reasonable
time limit. The third challenge is related to the traditional two-stage handling of QoS service composition (QSC) problems. Existing techniques assume that a predefined workflow
is available. Though the workflow can be derived based on functional requirements, the
independent processing potentially impacts the problem solvability and solution optimality.
To overcome these challenges, this dissertation aims at developing solutions to facilitate
feasible and efficient automating service composition using AI planning, especially for IoT
applications. Correspondingly, we have developed three solutions in this dissertation. First,
to facilitate applying existing planners in IoT service composition, we have developed a
service model to properly define IoT services and physical things and the techniques that
automatically map service composition problems to planning problems.
Second, for the scalability challenge, we propose a Two-Level Planning (TLP) framework,
which partitions a planning problem with a huge action set into two stages, each involving
a significantly smaller action set. The first stage applies a planner to identify the actions
relevant to the problem, and the second stage performs the planning process on the relevant
action set to derive the final solution. With TLP, since the problems in both levels are
much simpler than the original one, the overall planning efficiency is significantly improved.
Experimental results show that TLP can effectively reduce the action sets by 39% and reduce
the problem-solving time by up to more than two times.
For the third challenge, we adopt the existing numeric AI planning techniques. To cope with
the scalability issue, we introduce Provider Cardinality Reduction, a novel problem reformulation technique to improve planners’ efficiency in handling QSC problems and achieve
efficient integrated QoS-based service composition. Our approach focuses on reducing the
object set in the planning problems corresponding to QSC problems. Particularly, we estimate the number of providers (objects) required for the problems by analyzing the service
influence relation. Also, we develop techniques to force the planners to select up to the
estimated number of objects when searching for plans. As a result, the ground action set
and the state space are greatly reduced, and the planners achieve significant speed-up. Experimental results show that our approach can reduce the QSC problem-solving time by up
to 25 folds while retaining a competitive plan quality.
Our work significantly enhances the state-of-the-art technologies in automated service composition, especially in improving planning efficiency by planning problem reformulation.
Also, our problem reformulation solutions can be applied not only to automated service
composition but also to a wide variety of application domains, allowing much more efficient
problem solving for both classical and numeric planning problems