Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Blockchain Driven Technologies for Secure Data Sharing in Distributed Systems
No full textBlockchain technology is an emerging decentralized data collection, sharing, and storage technology
in peer-to-peer (P2P) networks, providing unique features such as data transparency, immutability,
security, and robustness for ledgering services in various real-world use cases. Recent years have
witnessed notable developments in blockchain technology and blockchain-enabled applications.
Blockchain has proven to be a remarkable success in cryptocurrency applications, such as Bitcoin,
Ethereum, and PeerCoin. The adoption of blockchain in many other fields continues to expand the
existing blockchain ecosystem.
The design of a blockchain system includes the specification of smart contracts, which contain
several policies, namely consensus mechanisms, incentive mechanisms, and additional functions for
each blockchain peer to follow, as well as the transaction and block architecture to specify what and
how data are stored in the blockchain.
In this dissertation, we focus on designing blockchain technology for secure data sharing in
distributed systems, particularly in two emerging and under-explored applications within the
ecosystem: healthcare and distributed machine learning in edge computing.
We first study a privacy-preserving contact tracing problem in pandemics, which has been demon-
strated as an effective way to control virus spreading. However, privacy concerns block the wide
implementation of contact tracing. Recognized for its decentralized and privacy-preserving capabili-
ties, blockchain technology holds potential for enhancing privacy in contact tracing applications. We
propose a comprehensive blockchain-driven framework, specifying the blockchain data structure, a
decentralized and lightweight consensus mechanism with integrated tracing data verification, and
an incentive mechanism to encourage voluntary participation and bear blockchain costs.
We then study how blockchain can enhance the robustness and security resilience for federated
learning (FL). FL is emerging as a sought-after distributed machine learning architecture, offering
the advantage of model training without the direct exposure of raw data. With advancements in
network infrastructure, FL has been seamlessly integrated into edge computing. However, the
limited resources on edge devices introduce security vulnerabilities to FL in this context. We
propose employing blockchain technology to bolster security and address the challenges brought
by constrained resources and heterogeneous data distribution on edge devices. We design the
Blockchain-empowered Heterogeneous Multi-Aggregator Federated Learning Architecture (BMA-
FL), including a novel lightweight Byzantine consensus mechanism, namely PBCM, to enable
secure and fast model aggregation and synchronization in BMA-FL, and a deep reinforcement
learning algorithm (MASB-DRL) for optimizing the FL training process
Testing the Standard Model of Cosmology Using Cosmological Data and Surveys
No full textThe last three decades of precision cosmology have been dedicated to understanding and
constraining the standard model of cosmology. Current tensions suggested by the most
up-to-date data motivate the study of extensions beyond the so-called standard ΛCDM
model. This dissertation focuses on testing the standard model of cosmology using existing
cosmological data while also presenting analyses to prepare the stage for forthcoming Stage-
IV surveys that will strengthen our understanding of cosmology. This dissertation is divided
into two main parts.
The first part of the dissertation focuses on modifications to the theory of general relativity
underlying the standard model. We introduce a formalism for conducting phenomenological
tests of modified gravity, accompanied by the presentation of the corresponding code named
ISiTGR. We applied this code to current cosmological data and performed various tests
of gravity. The first test involves singling out modified gravity parameters to improve the
constraints on such specific parameters and studying nested models for specific extensions
to the underlying theory of gravity. The second test emphasizes data-driven approaches,
employing binning methods to constrain modified gravity by considering bins in redshift and
scale. The first section concludes with a consistency analysis of current datasets as a further
test for the ΛCDM model.
The second part of the dissertation is dedicated to laying the groundwork for upcoming data
analyses from the Dark Energy Spectroscopic Instrument (DESI) survey. We detail various
tests and analyses conducted to ensure high-quality outcomes from DESI. This includes
survey validation forecasts, systematic tests for baryon acoustic oscillations analysis, and
presenting the case for extending the DESI science by doing cross-correlations with DESI and
other photometric surveys. Finally, we present tests related to halo occupation distribution
systematics on the baryon acoustic oscillations analysis for the DESI 2024 results that underly
further cosmological analysis using such data
Generating Morals for Fables
No full textFables are short narratives that teach us valuable lessons in the form of morals; understanding
these morals is important because they are the central messages of their fables. This thesis
explores the challenging task of understanding the morals of fables using natural language
processing. We propose features, focused on the characters and events of a fable, that include
both semi-structured subject-predicate-object triplets, as well as with the Story Intention
Graph (SIG; Elson, 2012b); we develop the first fully automated pipeline for building SIGs
from narratives. Using the triplet features, we train a neural classification model to label a
fable with one of twelve abstract categories of morals, demonstrating that neural networks
are capable of extracting moral information. Using SIGs, we tackle the more challenging,
open-ended task of generating a fable’s moral from scratch. We develop graph-to-sequence
models for this task, proposing a novel global meta-node for initializing a sequence decoder
from a graph encoder. Our best-performing model achieves state-of-the-art results for the
task of moral generation
An Exploration of Cell Type-Specific Enhancer RNAs
No full textMost of modern medicine is based on cell agnostic treatments which ignore how off-target cell
types are affected, resulting in a wide range of pleiotropic effects. Enhancer-derived RNAs
(eRNAs) have been shown to affect target gene expression and computational studies indicate
that eRNAs are expressed in a highly cell type-specific manner, providing a potential target for
cell type-specific gene modulation, which could lead to the development of more targeted
therapeutics.
While many computational studies examine eRNA cell type specificity, there are very few
experimental studies that explore cell type-specific eRNAs and the effects of eRNA inhibition
across different cell types or in different contexts. In this work, we explore the effects of eRNA
inhibition in multiple cell types in both a virus-infection and cancer context. We developed a
computational workflow to identify virus-induced transcripts and provide downstream
comparative analysis. RNAi was used to inhibit an array of eRNAs across cell types to
understand cell type-specific effects. While computational results indicate cell type-specific
transcription, experimental results demonstrate that most eRNAs were actually transcribed in
multiple cell types. While there were differences in eRNA abundances between cell types, the
effects on target transcript expression were mostly consistent across cell types, with difference in
the magnitude of knockdown. A comparison of experimental eRNA copy number with
computational expression revealed that eRNAs clustered into two groups, suggesting distinct
mechanisms. Analysis of the combined results revealed that eRNAs with low expression tended
to yield a more robust knockdown effect.
Exploration in the cancer context revealed that, similarly, effects on target transcript expression
were consistent across cell types, but inhibition produced dramatic differences in downstream
protein abundance and proliferation in the different cell types.
Enhancer RNAs remain a compelling target for cell-type specific gene modulation, but the
effects of stochastic expression and the differential downstream effects of inhibition need to be
thoroughly understood before they can be utilized for precision therapeutics
Topological Machine Learning in Medical Image Analysis
No full textMedical image analysis is a critical component of modern healthcare, aiding in the detection, diagnosis,
and treatment of various diseases. In recent years, the fusion of topological data analysis
(TDA) and machine learning has brought about a transformative approach to understanding and
leveraging the complex patterns inherent in medical images. This thesis explores the application of
TDA, with a focus on persistent and cubical homology, as a powerful tool for extracting topological
features from 2D and 3D medical images. By capturing the birth and death of these features
through a filtration process and representing them as feature vectors, we enhance the interpretability
and performance of machine learning models.
Our journey begins with the transformation of color images into grayscale, RGB, and HSV channels,
laying the foundation for versatility in our approach. The three-step process of persistent
homology unfolds through filtration, persistence diagram generation, and vectorization, leading to
feature vectors that outperform traditional image inputs in machine learning models.
In our investigations, we utilized this methodology across various medical imaging contexts such
as cancer histopathology, chest X-rays, retinal images, and MEDMNIST a collection of various
medical image datasets. Our findings highlight the effectiveness of topological features in analyzing
images, particularly in classifying and identifying abnormalities in these settings. Additionally,
when we merged these topological aspects with deep learning characteristics, we observed
an improved model efficiency. This presents a novel, comprehensive approach to medical image
analysis.
This research further demonstrates the potential of TDA in image analysis and classification, providing
valuable insights into complex structural information not readily apparent through conventional
techniques. In the realm of histopathological cancer detection, we introduce a groundbreaking
Topo-ML model that adapts to multiple cancer types, offering remarkable computational efficiency.
In the evaluation of chest X-ray images, our Topo-CXR model excels in terms of reliability
and interpretability, outperforming deep learning models on benchmark datasets to detect Pneumonia
and Tuberculosis. Additionally, our analysis of fundus images for retinal diseases presents
a dual contribution, introducing effective feature extraction methods and topological deep learning
models that surpass existing models.
Beyond these specific applications, we conducted a comprehensive evaluation of TDA methods
across various medical image datasets, demonstrating competitive performance with deep learning
models in certain cases. These findings underscore the potential of topological features in
boosting the capabilities of future machine learning models in diverse medical domains, ultimately
advancing our understanding and utilization of digital images in healthcare
An Asian American Cookbook: the 汤圆(Tang Yuan) Chocolate Chip Cookie
No full textThe 汤圆(Tang Yuan) Chocolate Chip Cookie is a three-dimensional diorama that captures and
guides viewers through the process of combining two drastically different ethnic desserts into
one final product. Through a video presentation of the diorama, this project is designed so that
each part of the 3D environment contains hints and clues about the baker's identity. It aims to
introduce readers to embark on a treasure hunt to piece together the story of this new cookie and
how it came into existence.
By employing dynamic composition with baking ingredients, the implied motion directs the
reader's gaze towards key elements of the baking environment. The scene's composition and prop
placement begin with the camera opening to the center of the room and progressing to display
the cooking process from left to right. A slow pan captures the viewer's attention, allowing them
to explore the baker's story elements scattered throughout the kitchen
Relationships Between Age, Subjective and Objective Measures of Recollection, and Pre-stimulus Neural Correlates of Memory Encoding and Retrieval
No full textIt is well established that episodic memory (i.e., memory for personal unique events) declines
with age. There is some evidence, however, that the decline might be moderated by the
estimation methods. That is, while the impact of age on recollection is consistently evident when
using ‘objective’ memory procedures (e.g., a source memory task), recollection estimated from
‘subjective’ procedures such as the Remember/Know procedure gives rise to inconsistent impact
of age on recollection. Therefore, it was suggested that the null impact of age on recollection
reported in some prior studies might be due to a differential impact of age on subjective versus
objective aspects of recollection. In the first study covered in Chapter 2, we examined this
hypothesis by directly contrasting recollection estimated from the source memory procedure with
recollection estimated from the R/K procedure in the same sample of young and older adults.
Our results revealed an equivalent impact of age on recollection estimated from both procedures.
Hence, we found no evidence that the impact of age was moderated by the subjective and
objective aspects of recollection. In Chapter 3, we conducted an ERP study to examine the
relationship between retrieval mode (i.e., pre-retrieval processing suggested to facilitate episodic
retrieval) and recollection on a trial-by-trial basis and whether this relationship is moderated by
the impact of age. Our results revealed that the pre-retrieval ERP effects associated with
recollection was distinct from those associated with non-semantic judgements. The pre-retrieval
ERP effects took the shape of more negative going ERPs preceding recollected items compared
to that preceding non-episodic semantic items. The results are in line with the suggestion that
retrieval mode facilitates episodic retrieval. Of importance, we found no evidence that this
pattern of effects was moderated by age. Chapter 4 covers the impact of age on a different aspect
of episodic memory. Here, we examined the impact of age on pre-stimulus subsequent memory
effects (pre-stimulus SMEs). Only a few studies examined age-related differences in prestimulus
SMEs and reported mixed findings. Our results revealed an impact of age on prestimulus
SMEs. That is, the pattern of results demonstrated more negative-going ERPs for
correct than incorrect source judgements in the young age group only. An important question is
whether the pre-stimulus SMEs predict behavioral age-related differences in recollection. Indeed,
the impact of age on the pre-stimulus SMEs corresponded to the impact of age on recollection
and recognition memory in our behavioral findings. Taken as a whole, the present dissertation
provides the field of cognitive aging and memory with valuable insights in understanding the
impact of age on recollection from different methodological and memory stages’ perspectives
Convex Hull Simplification and Geometric Hardness
No full textAs data can often be viewed as geometric points, using geometric algorithms for extracting
properties of the data is natural. Relevant examples include simplifying the structure of the
data or generating a clustering of the data.
This dissertation studies two problems concerning geometric data. The first is the problem
of simplifying convex hulls by selecting a subset of the original point set such that its convex
hull minimizes certain error functions. The error measures are defined in terms of distances
between the selected convex hull and the full set of points. In particular, the focus in this dissertation is on the maximum distance, and the sum of distances. Results for these measures
include exact algorithms in two dimensions and hardness results in higher dimensions.
The second problem considered in this dissertation is the problem of k-center clustering of
disjoint convex objects. The results include approximation algorithm for disks, and hardness
results for line segments, unit disks, and axis aligned unit squares
Toward Applying Variability-oblivious Static Analyses to Software Product Lines
No full textSoftware variability has been shown to introduce bugs which only exist in certain variations
of a software product line (SPL). A SPL is formed with configuration options which change
select parts of the code. These configuration options lead to exponential growth in the
number of variations, causing exhaustive analysis of an SPL to be infeasible. Existing techniques to analyze SPLs have used sampling techniques, created variability-aware analyses,
and systematically lifted analysis techniques. Sampling SPLs allows the use of state-of-the-
art analyzers, but only a small subset of program variations can be tested, leaving blind-spots
for bugs to remain in. Variability-aware analyses solve this problem by analyzing an entire
SPL at once. However, since these analyses must be adapted or made from scratch, their
capabilities lag behind state-of-the-art tools. Systematic lifting allows for existing techniques
to operate on a SPL, but no implementation has been shown to practically lift existing tools
due to limitations of analysis type or input format. This dissertation explores approaches to
systematically lift existing state-of-the-art tools to work on the entirety of a SPL.
Our first approach is the desugaring approach, which aims to rewrite an entire SPL as one
individual piece of code, and then apply existing analyzers the result to the desugared code.
This approach is split into two works. The first work discusses the process and implementation of rewriting C and C Preprocessor code into a pure C representation. Rewriting has been
attempted before, but prior attempts fail to handle the complexities of real-world preprocessor usage, and how they effect constructs such as user defined objects. We improve upon
these with our tool SugarC, which implements a novel syntax-directed translation approach
to rewriting, and discuss new optimizations to user defined objects which allow for real-world
use. We evaluate SugarC on axTLS, Toybox, and Busybox, where it was shown to scale to
real-world preprocessor usage for highly configurable programs. The second work presents
a framework that applies existing variability-oblivious tools to the desugared representation
and extracts implicit details. The desugaring process introduces infeasible paths, renames
key methods like malloc and free, duplicates warnings that appear in multiple configurations,
and obscures the configuration options requires for a given path. Our framework, Sugarlyzer,
enables the analysis of SugarC output by addressing these issues and using alarm reports
generated by existing analyzers to intuitively deduce the presence condition and feasibility
of alarms. Through the generic design of Sugarlyzer, we implement three popular bug finders, Clang static analyzer, Infer, and Phasar, and use them to analyze axTLS, Toybox, and
Busybox. We explore the findings from this approach and compare them to state-of-the-art
sampling and variability-aware analyses.
The desugaring approach reveals inherit unsoundness in static analyzers due to the multitude
of paths that are added on top of their inability to know what configuration options are
mutually exclusive. The third work of this dissertation addresses this as we explore the idea
of generically lifting a language through variability input interpretation. Past lifting through
variability-aware interpretation work has been stifled by reliance on variability-aware data
structures as input, which prevents them from truly lifting many existing tools. We devise
and implement a new approach to read any variable input and combine it with simultaneous-
interpretation to lift any program at the language level. To highlight the modularity we
present a case study where we lift a parser to being variability-aware, thus showing the
feasibility of applying our approach to replicate some of the most powerful and widely use
variability-aware tools
Home-based Lower Limb Immersive Serious Game for Phantom Pain Management
No full textWith recent advances in mixed reality technology, Head Mounted Devices (HMDs) have
become a preferred option for immersive gaming at home. Immersive games using HMD have
been applied for serious purposes, such as improving motivation for performing rehabilitation
activities at home. Mr.MAPP is a virtual mirror therapy system designed for amputees with
phantom limb pain. In this work, we first investigate the Mr.MAPP system through a clinical
patient study and further improve the system design based on the uncovered insight. In
addition, to allow users to experience mixed reality in the first-person perspective using their
realistic body reconstruction, we prototype a system to aggregate the body’s texture in realtime.
Through simulation using the reconstruction system, we created an optimal camera
setup guideline for amputees’ in-home setup. Lastly, a Home-based Lower Limb Immersive
Exergame System (HILLES) is developed to address challenges such as system setup, safety,
and game engagement for in-home usage. A stomping game using the system is developed
for an exploratory user study to evaluate the system and the proposed enhancements. The
discovered insight could be helpful for future lower limb exergame design