Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Artificial Software Diversification Through Program Synthesis
No full textA means of attaining richer, more comprehensive forms of software diversity on a mass scale
is proposed through leveraging and repurposing a closely related, yet heretofore untapped,
line of computer science research—automatic program synthesis. It is argued that the search-
based methodologies presently used for obtaining implementations from specifications can be
broadened relatively easily to a search for many candidate solutions, potentially diversifying
the software monoculture. Small-scale experiments using the Rosette synthesis tool offer
preliminary support for this proposed approach.
Dually, program synthesis-driven cyberattacks can potentially power a dangerous new level
of sophistication for malware mutation and reactively adaptive software threats. It is therefore important to develop synthesis-aware automated code analyses that can identify and
mitigate malware threats that leverage program synthesis as a strategy for code mutation
and obfuscation.
To address these needs, a purely relational definition of a central processing unit using
miniKanren is proposed as a new approach for realizing assembly code diversification. Software diversity has long been championed as a means of protecting digital ecosystems from
widespread failures due to cyberattacks and faults, but is often difficult to achieve in practice.
Using relational programming to simulate a processor allows large-scale automatic synthesis
of assembly-level code. Experiments with the technique indicate that such synthesis might
lead to better automation of code diversification by breaking the synthesis problem into
manageable chunks. An early prototype is presented, with some sample synthesis tasks and
discussion of possible future applications.
An introduction to miniKanren is also presented for users who have never seen the language,
along with an algorithm for introducing universal quantification to miniKanren, which is
central to solving the assembly code diversification problem.
Finally, to support formal reasoning about loops in functional code, the art of converting
Haskell’s lazy evaluation to the coinductive structural formalisms offered by the Coq automated theorem prover is introduced and explored. A Haskell-centric coinduction tutorial
first introduces newcomers to core concepts underlying formal analysis and verification of
lazily evaluated structures and algorithms. Simple code examples illustrate how to undertake conversions to Coq, and lessons learned from translating and verifying Haskell code are
related.
The analysis shows that Haskell data definitions actually correspond to greatest fixed-points
rather than the more typical inductive least fixed-points found in most algebraic data types.
Examples of coinductive definition, proof, and bisimulation illustrate and reinforce this correspondence
GMIP: A GWAS and Multi-Omics Integration Pipeline for Gene Prioritization and Interpretation
No full textGenome-wide association studies (GWAS) have become essential tools in identifying genetic variants associated with complex traits and diseases. GWAS plays a critical role in drug discovery by uncovering novel therapeutic targets. Despite these advancements, GWAS often falls short in directly identifying causal genes or pathways. Moreover, with the increasing availability of omics data, such as gene expression, protein-protein interactions, and pathway memberships, there is an urgent need to integrate GWAS with multi-omics datasets to enhance biological interpretation and facilitate drug development.
My Ph.D. work focuses on two main contributions to addressing these challenges. First, we developed GMIP: GWAS and Multi-omics Integration Pipeline, a unified and flexible framework for gene reprioritization tasks. GMIP is designed to incorporate widely used tools like PoPS, MAGMA, NAGA, and benchmarker, while remaining computationally robust, scalable, and easy to run on laptop, HPC, or cloud through its implementation in Nextflow.
Second, we applied Partial Least Squares Regression (PLSR) to address multicollinearity in PoPS-selected features, significantly enhancing gene reprioritization and result interpretability. We
demonstrated the versatility of GMIP-PLSR by applying it across a wide variety of GWAS, showing its effectiveness in multiple contexts.
To demonstrate the utility of GMIP, we applied it to the NAFLD GWAS. We reprioritized genes not only using the standard GMIP-PLSR pipeline and generalized PoPS features, but also by integrating scRNA-seq data from a NAFLD mouse model. The reprioritized genes showed a significant increase in heritability over baseline models and are highly enriched in known NAFLD pathway genes, validating the biological relevance of our approach.
GMIP is a comprehensive and scalable framework, freely and publicly available at https://github.com/mohammedmsk/GMIP, providing researchers with a powerful tool for integrating multi-omics data into GWAS summary statistics
Emotion Classification Using Biomechanical Analysis and Machine Learning
No full textGait contains many hints as to the health and mental state of the person performing it. Analysis
of how people walk can be used in clinical settings for both diagnosis and prognosis of different
pathologies such as knee osteoarthritis and Parkinson’s disease, in security settings to help
identify persons of interest, or in the improvement of human-machine interfaces. The goal of this
thesis was to apply machine learning techniques to develop an emotion classification model from
gait performance across five different emotions. The models can consistently identify the
specific emotional state during a given trial, the arousal or energy level of the emotional state,
and the valence or positivity level of the emotional state at higher rates than the a priori
probability (20% for specific emotion, 33% for arousal and valence). The models, however,
struggle with classifying emotional states that are underrepresented in the data set. Additionally,
the low number of participants (n = 12) leads to large variance across validation sets. While the
results of this initial study are promising, future work will employ methods to address class
imbalance and additional data collection to address its limitations
Essays on Options and Volatility
No full textThis dissertation consists of three essays on options and volatility.
First first essay, included in Chapter 1, is ”The Information Content in the Implied Volatility
Surface Around Earnings Announcements”. I show that information embedded in the implied
volatility surface (IV surface) has great explanatory power on stock price dynamics surrounding
earnings announcement dates (EADs). I show that a great component of return risk premium
can be extracted from the IV surface, denoted as RRPv. Prior to EADs, RRPv exhibits
remarkable explanatory capabilities, accounting for 73.21% of the cross-sectional return
variation. On EADs, the IV surface reveals a pronounced volatility feedback effect at the
aggregate level, offering compelling evidence that contributes to resolving the well-documented
earnings-return relationship puzzle. Post-EADs, stocks exhibit unexpected price behavior
that can be predicted by RRPv, underscoring the presence of a compelling idiosyncratic
component within the IV surface variation that is driven by firm-specific news. Implementing
a real-time RRPv sorted long-short portfolio after EADs generates an annualized abnormal
return of 13.68%.
The second essay, included in Chapter 2, is “Earnings Announcements: Ex-ante Risk Premia”,
co-authored with Hong Liu, Xiaoxiao Tang, and Guofu Zhou. We provide the first estimates
of the ex-ante risk premia on earnings announcements from the forward-looking information
of the options market. We find that the average earnings announcement risk premium is
highly significant at 15 basis points, with substantial variation across firm and across time.
Sorting by the estimated ex-ante risk premium generates a daily return spread of 32 bps
between high and low terciles. Moreover, the estimated ex-ante risk premia provide new
insights on what drives the well documented positive post-earnings-announcement drift and
offer profitable straddle strategies.
The thrid essay, included in Chapter 3, is ”Improving the Performance of Volatility-Managed
Portfolios”, co-authored with Xiaoxiao Tang and Feng Zhao. Recent studies have criticized
volatility-managed portfolios for two reasons: poor out-of-sample performance and inaccessible
abnormal returns owing to transaction costs. We propose a simplified and more robust
method of volatility-timing by focusing on downside forecasting using past volatility. The
proposed trading strategies are free of look-ahead bias and have low trading costs. We show
that downside-managed portfolios outperform unmanaged portfolios and volatility-managed
portfolios. Lastly, we show that downside-managed portfolios embed a mechanism to ideally
balance between type I and type II errors in downside forecasting under return asymmetries
Digital Democracy: Navigating Free Speech Norms in the Cancel Culture Era
No full textIn the era of “cancel culture,” what role does freedom of speech play in online political
discourse? How does online cancel culture work, and what implications does it have for free
speech? Does cancel culture make users more considerate about the topics they comment about
online? This dissertation considers these questions by examining specific cases to provide
insights into the broader context of online interactions. Specifically, the cancel culture cases of
Gina Carano and James Gunn (Chapters 2 and 3), as well as discourse surrounding presidential
debates (Chapter 4), are analyzed to understand the dynamics, influence, and implications of
online political discourse and free speech. The study reveals that online platforms are used by
both elites and non-elites driven by ideological motivations. In both cases, a small but highly
vocal minority provoke backlash against individuals, while a larger majority push back against
“canceling.” Moreover, despite significant online engagement, these controversies are short-
lived, showcasing the need for companies to consider long-term impacts before taking drastic
actions. Analysis of YouTube comments of presidential debates indicate shared interests among
viewers across ideological lines, suggesting that viewers do not necessarily select content to
place comments based on ideological orientation. The dissertation concludes (Chapter 5) with a
discussion of the implications of these findings as well as potential avenues for further research
The Dimensional Stability of Deployable Composite Structures During Manufacturing and Stowage
No full textDeployable composite structures have been developed over the last 15-20 years. With spread-
tow carbon fiber enabling the ability to make much tinner textile plies and thus the ability to
make much thinner carbon fiber reinforced polymer composites also exist. These thinner
composites take advantage of the high strength and low weight capabilities that their thicker
counterparts have while also being more deformable and able withstand higher strains. This
combination of factors makes these composites ideal for certain deployable space structures that
need to be light weight and have high packaging efficiencies. Making deployable structures out
of these flexible composites can simplify many complex deployment systems currently used in
space. With the promising properties of these structures come unique challenges. These thin
composites are susceptible to manufacturing induced distortions as well as distortions caused by
packaging/stowage. These distortions affect the dimensional stability of the deployable structures
which in turn affects the structural performance. In this work the capabilities to understand and
predict these dimensional instabilities are improved. Material models are developed for the
aspects of theses thin composites that cause deformations during manufacturing and stowage.
New testing and experimental characterization methods are explored for testing these composites
under the high strains they are capable of sustaining during stowage. Finally testing methods are
developed and used to understand the deformations these structures will retain over the course of
their manufacturing, stowage, and deployment life cycles
BORNTOBEPERFECT
No full textMy adolescent preoccupations with choir, art, and English converged into a specific field of
study as an adult: sound art. Often, the work is the sound: especially as it relates to sound in a
physical space. Thus, one of my favorite methods through which to display work is installation.
Among other mediums, I also meld my background as a classically trained vocalist with the field
of live performance. The signature of a piece made by me is adaptive, live, or evolving
soundscapes, intentional feedback looping, and repetitive phrases. Often feeling bewildered by
the Herculean task of decoding myself, my pieces are a form of self-interrogation. This thesis
details the development of my artistic practice over the last three years as I explored the
mediums of sound, installation, and live performance
Reliability Challenges and Condition Monitoring of Power Devices
No full textThis dissertation presents a comprehensive analysis of reliability and performance issues
in silicon carbide MOSFETs (SiC MOSFETs) and isolated Laterally Diffused Metal Oxide
Semiconductor (LDMOS) devices, focusing on gate oxide degradation and package reliability.
The first part focuses on the third quadrant operation of SiC MOSFETs, examining the
factors influencing the behavior and reliability of the inherent body diode. The study shows
that current flows through both the channel and PN junction during third quadrant operation,
affecting reverse recovery current and power losses. By adjusting the gate voltage bias, the
distribution of current between the channel and PN junction can be controlled, thereby
regulating current flow during reverse conduction. The degradation of the gate oxide impacts
the third quadrant operation by reducing the reverse recovery current. A proposed circuit
monitors gate oxide degradation by observing the peak reverse recovery current (PRRC),
validated through experimental results on a double pulse tester (DPT) setup.
The second part of the study investigates the reliability of TO-263 SiC MOSFETs, emphasizing
gate oxide and package degradation. Accelerated aging tests (AATs) such as high temperature
gate bias (HTGB), high temperature reverse bias (HTRB), and DC power cycling (DCPC)
reveal that a 10nm reduction in gate oxide thickness significantly impacts its reliability.
Detailed analyses using focused ion beam (FIB) and transmission electron microscopy (TEM)
highlight the critical role of gate oxide thickness. Additionally, the reliability of the TO-263
package is assessed, identifying solder fatigue failure between the device and the printed
circuit board (PCB) as a major issue, leading to potential device detachment and loss of
connection. This underscores the necessity of optimizing gate oxide thickness and addressing
thermal-mechanical stresses to enhance the reliability of SiC MOSFETs in high-power
applications.
The final part of the dissertation investigates the robustness and performance degradation of
isolated LDMOS devices under repetitive avalanche breakdown and dynamic reliability tests.
Three custom LDMOS devices with different isolation configurations are designed to study
the impact of isolation well connections on aging. The results from a large-scale accelerated
aging setup indicate a gradual increase in on-state resistance and drain leakage current during
repetitive avalanche tests, with only the drain leakage current increasing during dynamic
tests. This identifies the drain leakage current as a reliable precursor for condition monitoring.
Among the three configurations, the most reliable isolation configuration is determined, with
failure locations and device lifetime varying based on isolation well connections. Furthermore,
the forward characteristics of the body diode are analyzed, showing that the distance between
n-well and substrate terminals significantly enhances diode robustness.
Through these investigations, the dissertation provides insights into optimizing device perfor-
mance and reliability for high-power and high-reliability applications
Search Costs, Working Capital Management and Asset Pricing
No full textHow might the cost of searching for information influence asset prices? This paper develops
a production-based asset pricing model with imperfect information, in which firms can wait
to buy inputs and sell inventories while searching for low cost suppliers and high value
customers. Investors, meanwhile, can wait to buy assets while searching for cheap investment
opportunities. The model predicts that when search costs rise, profits fall and the marginal
utility of consumption spikes, leading investors to discount firms that are especially exposed
to search cost risk ex ante. I identify observable firm characteristics related to search costs,
and find that some of these variables do predict stock returns. Results suggest that excess
inventory can be a hedge against the risk of higher search costs accompanying supply chain
fragility or information suppression
Knowledge-intensive Natural Language Generation
No full textIn an era of information overload, the ability to generate knowledge-intensive natural language (KI-
NLG) has become critical for efficiently processing and transforming domain-specific data. This
dissertation presents two interrelated lines of research in KI-NLG: related work generation (RWG)
and knowledge-grounded dialogue response generation. These studies explore the challenges of
generating accurate and informative text based on diverse and evolving knowledge sources.
Our first research line focuses on the under-explored task of RWG, which involves automatically
composing short literature reviews. We propose novel approaches for citation text generation,
shifting the focus from citation sentences to spans for more precise generation. To facilitate re-
search in this area, we introduce the CORWA dataset, which annotates citation spans and their
discourse roles, and demonstrate the utility of retrieving cited text spans for citation generation,
improving both accuracy and faithfulness. Additionally, we propose a planning-based approach
to RWG by extracting relationships between research papers through features derived from large
language models.
The second line of research addresses established KI-NLG tasks, mainly focusing on dialogue gen-
eration, but also including fact verification and question-answering. We present the multi-source
Wizard of Wikipedia (Ms.WoW) benchmark, designed to evaluate a model’s ability to adapt to new
knowledge sources in a zero-shot setting. We further analyze the interplay between knowledge
retrieval, selection, and retrieval-augmented generation performance. In addition, we introduce
the minimal evidence group (MEG) framework to reduce redundancy among knowledge pieces.
Finally, we develop a planning-based approach for dialogue generation in e-commerce settings,
leveraging decision trees for product-focused conversation generation, and present the Wizard-of-
Shopping (WoS) dataset to enhance conversational product rankers