Treasures @ UT Dallas
Not a member yet
7697 research outputs found
Sort by
Simple and Complex Manual Sequence Learning in School-Aged Children with Typical Development and with Developmental Language Disorder
No full textThe goal of this dissertation was to examine how children, both with typical development and
with developmental language disorder (DLD) learn two types of sequences, important for
language, in the manual domain. We sought to better understand the developmental trajectory of
statistical learning in the manual domain and investigate the extent to which cognitive
mechanisms underlie language learning in general and DLD in particular. Specifically, we
distinguished between two types of learning on a domain-general, modified Serial Reaction
Time (SRT) task: local transitional probabilities versus abstract exclusive disjunctive (XOR)
rules.
Typically developing (TD) infants can learn phonotactic XOR rules that adults cannot (e.g.,
Dell et al., 2021; Gerken et al., 2019), but the developmental trajectory of this ability
throughout childhood is not well understood. Research on rule-learning has predominantly
focused on phonotactic patterns; it remains unclear whether the learning process is specific to
language or applies more broadly across domains. Here, we assessed the extent to which TD
school-aged children learned both a simple pattern involving local transitional probabilities
(Word condition), and a complex pattern involving abstract XOR rules (Grammar condition), on
a domain-general modified SRT task.
This dissertation also served to inform theoretical accounts of DLD. Children with DLD are
classically identified by their grammatical deficits (e.g., Leonard, 2014), but often display co-
occurring weaknesses in other areas, including speech-motor organization (e.g., Benham et al.,
2018), and fine/gross motor skill (e.g., Hill, 2001). We anchored this dissertation in the
hypothesis that a domain-general sequential pattern learning deficit (of specific sequence types)
unifies language, speech, and motor difficulties attested in DLD. Critically, the sequences in the
patterned blocks of our SRT task are derived from components of language that are relative
linguistic strengths (i.e., word boundary parsing) or linguistic weaknesses (i.e., morphosyntactic
learning) among children with DLD. The rules governing the sequences are novel for SRT tasks
and are important for specifying the precise nature of a potential domain-general sequential
learning impairment in DLD. The second goal of this dissertation was to assess the extent to
which children with DLD learned local transitional probabilities (Word condition), and abstract
XOR rules (Grammar condition), on the domain-general modified SRT task.
Children aged 5-8 years with TD (n = 26) and DLD (n = 9) participated. TD participants
demonstrated evidence of learning in both the Word and Grammar conditions, though learning
appeared to be more protracted in the Grammar condition. There was not strong evidence that
participants generalized the XOR rule. Overall, these results suggest that TD children are
sensitive to local transitional probabilities and to abstract XOR rules in a domain-general task
into the early school years. Preliminary results revealed that school-aged children with DLD are
sensitive to local transitional probabilities, but not to a complex XOR rule, on a domain-general
SRT task. This supports an account of DLD in which specific sequence learning, conceptually
aligned with grammatical structure, is implicated across domains. Specifying a nonlinguistic
mechanism of DLD may lead to more targeted interventions and earlier identification across
dialects/languages using domain-general measures
Still and Seen
No full textThe MFA thesis exhibition, Still and Seen, is a site-specific and socially engaged art exhibition
that offers a safe space for connecting to vulnerable emotions. In this exhibition, both
installations Amidst Tears and Hurt to Heart encourage the audience to slow down and be in the
present moment, while fully immersed in a multi-sensory space filled with storytelling elements
in installation Boxes of Memories and performative art video Still Seen. I used clay, acrylic
boxes, paper, metal wires and shredders to create my installation and sound design, expressive
art therapy techniques and aromatic materials to enhance this aesthetic experience. Artifacts and
artworks in thesis exhibition are my experiments to tell my story while inviting meaningful
interaction through aesthetic experiences that also contribute to the interdisciplinary field of
neuroscience and aesthetics or neuroaesthetics
Diversity and Dynamics in Protest Movements: a Comprehensive Analysis of the Citizenship Amendment Act Protests in India
No full textThe dissertation analyzes how protester composition and diversity influence contentious political
movement dynamics through large-N comparative analysis and case studies. It focuses on protest
actions in India, particularly the Citizenship Amendment Act (2019) protest movement. It
examines the diversity of participants, motivations, and tactics that characterized the CAA
protests, and explores the factors that contributed to this campaign. The second chapter provides
an overview of the theoretical and conceptual framework for understanding social movements,
with a focus on the CAA protests. It discusses theories related to collective identity, regionalism,
economic grievances, and protest strategies and offers a framework called regional identity-
based theory of revolution to qualitatively analyze the diversity and dynamics within the CAA
protests. The second chapter employs a comprehensive text analysis of over 3,000 English news
articles covering the Citizenship Amendment Act (CAA) protests and the Farmers' protests in
India. It uses topic modeling to map the underlying narratives, motivations, and fault lines within
and across these two protest campaigns. The analysis aims to provide insights into the factors
driving the unity and fragmentation of protesters, as well as the complex interplay of factors such
as shared identities, grievances, leadership, and framing processes. The fourth chapter introduces
a novel approach to measuring protest campaign fragmentation using Google Trends data. By
utilizing Google Knowledge Graph, it identifies relevant search queries and topics related to
specific protest campaigns. The study offers a behavior-based measure by providing a single
score for the campaigns based on their unification and fragmentation. This approach aims to
address the limitations of traditional methodologies and provide a scalable and comparable tool
for analyzing the internal dynamics of social movements. The dissertation contributes to the
literature on social movements and contentious politics by offering a contextualized
understanding of protest dynamics in diverse societies across the world
Millimeterwave Beamforming Antenna Arrays and Energy Harvesting Systems for the Next-generation of Sensing and Communication Applications
No full textThe rising demand and dependence on bandwidth-intensive wireless devices have led to a
global effort to create a mobile connectivity strategy that integrates satellites and high-altitude
drone-to-drone (D2D) platforms. Central to this initiative is the need for a highly efficient
antenna design that provides high gain, broadband operation, and high radiation efficiency
with a focused beam along the line of sight. This design also needs to be cost-effective for
mass production to serve a broad market. Despite recent advancements in antenna arrays
for high-frequency applications, few designs have optimized parameters for high gain, broad
bandwidth, and beam steerability. In addition, an antenna array for various vertically stacked
and linear configurations suitable for drone deployment is indeed required. There is a notable
research gap in designing millimeter-wave (mmWave) antenna arrays that are compact (less
than 3λ × 3λ) and broad bandwidth (greater than 50% fractional bandwidth).
Part 1 of this research focuses on designing optimal mmWave antenna arrays, specifically
Vivaldi antennas, to achieve high gain and broadband operation with at least 50% fractional
bandwidth. This helps mitigate the losses caused by atmospheric attenuations at mmWave
frequencies. Antenna array analysis for various vertically stacked and linear configurations
and implementation of active frontend based electronic beam steering is also included within
this portion.
Part 2 of the study explores bias-free energy harvesting techniques using Reverse Electrowetting on Dielectric (REWOD). This method involves using electrolyte impingement through
mechanical modulation for energy harvesting. Traditional REWOD research has used inflexible planar electrodes that require a high voltage bias for better power output. This study
introduces a novel approach using flexible electrodes made with sputtering-based physical
vapor deposition (PVD) on polyimide sheets. Flexible electrodes are essential to overcome
the limitations of traditional planar configurations. This flexible design for REWOD-based
energy harvesting opens up new possibilities for wearable, self-powered motion sensors by
effectively capturing energy from electrolyte impingement.
In summary, the research outcomes include novel mmWave Vivaldi antenna and corresponding
array design with a focus on active beamforming for D2D and military communication
applications, and advancements in bias-free energy harvesting using high-dielectric flexible
electrodes with the REWOD phenomenon. These two contributions of this dissertation
primarily pave a path towards advanced sensing and communication applications
Three Essays on Economic and Operational Challenges in Digital Advertising
No full textIT breakthroughs have been consistently reshaping the advertising industry landscape, creating new chances for advertising firms while presenting new challenges. A critical impact of
those IT breakthroughs is the shift toward digital advertising, which has distinctive features
compared to offline advertising, like targetability. My dissertation has three main chapters
and focuses on the questions faced by various firms with the rise of digital advertising.
Chapter 2 analyzes the problem of advertising coordination across different brands and
advertising media. The growing online retail market has led to the prevalence of multichannel retailing. Meanwhile, retailers are increasingly combining multichannel retailing with a
multibranding strategy. While this can further increase the retailer’s sales, it brings new
advertising challenges. Multibrand, multichannel retailers usually launch advertising campaigns for different brands on multiple media. Thus, the retailer’s advertising efforts fall
into a set of brand-media units. Each unit’s advertising efforts can affect the sales of all
brands on all channels. Therefore, retailers need to coordinate the advertising efforts of different units to maximize advertising efficiency in propelling sales. So far, the optimization
problem of multibrand, multimedia advertising has not been analyzed in the literature, and
our study aims to bridge this gap. We develop a stochastic differential equation model to
estimate the impact of multimedia advertising on sales in a multibrand, multichannel context. Using the data from a jewelry retailer in the U.S., we show that our model is effective
in predicting future sales driven by advertising. Afterward, we formulate the advertising
optimization problems under four coordination strategies: (i) non-coordination, (ii) brand
coordination, (iii) media coordination, and (iv) global coordination. By solving the problem
for each strategy, the retailers can obtain the optimal expenditure for each unit under that
strategy. Finally, we compare the retailer’s profits under four strategies.
Chapter 3 analyzes the potential of ad exchanges to increase revenues by subsidizing advertisers to acquire data. Large volumes of online impressions are sold daily via real-time
auctions to deliver targeted advertisements (ads) to consumers. Advertisers use data to learn
about user preferences and select the most appropriate ad for each user, which also helps
them optimize their bids in an ad auction. While ad exchanges may provide some user data
to advertisers, it is usually limited, and advertisers often acquire data from various sources
to improve targeting performance. The acquisition of such data can significantly influence
the revenue of the ad exchange, which has mainly been passive about advertisers’ data acquisition process. Previous studies have examined the impact of ad exchanges revealing their
data to advertisers, but little attention has been paid to the active role that ad exchanges
can play when advertisers acquire data themselves. To address this gap, we propose three
subsidy frameworks to increase ad exchange revenue by inducing more advertisers to acquire data: All-subsidized (AS), Winner-subsidized (WS), and Loser-subsidized (LS). Using
a stylized model, we analyze the impact of subsidy provisions on the platform’s net revenue.
Chapter 4 analyzes how many advertisers a bidding agent should work with to maximize
its profit while ensuring a high likelihood of acquiring the number of impressions requested
by advertisers. The digitization of billboards has facilitated the sale of advertising slots
through real-time auctions. Consequently, there has been a rise in the number of agents
who assist advertisers in bidding and acquiring these slots. These agents enter into contracts with advertisers to secure a specific number of slots within designated time periods.
To fulfill these contracts, agents participate in auctions to win the desired slots. However,
accepting numerous contracts may require the agent to place higher bids in auctions, potentially impacting their profitability. Therefore, it is crucial to strike a balance between
contract acceptance and the bidding strategy employed. Motivated by these observations,
we address two key aspects: (1) identifying the optimal set of advertisers for the agent to
contract with, and (2) determining the appropriate bidding strategy based on the contracted
advertisers’ demands. We formulate a two-stage optimization problem for agents, followed
by the proposal of a near-optimal solution to the bidding optimization problem. Through
this near-optimal solution, we demonstrate that the objective function in the resulting advertiser selection problem exhibits the properties of a submodular set function. To solve the
advertiser selection problem, we introduce a greedy algorithm
A Peer-to-peer Access Control Infrastructure for IoT Systems With Efficient Blockchain Solutions
No full textThe Internet of Things (IoT) integrates a vast array of sensor-equipped devices across various
networks, boosting daily operation efficiencies and bringing great social benefits. However,
together with the advantages, IoT systems also present stringent security requirements. A
fundamental security requirement for IoT systems is the management of the accesses to the
shared IoT devices and their collected data.
Many existing access control models, such as Discretionary Access Control (DAC), Role-Based
Access Control (RBAC), Attribute-Based Access Control (ABAC), etc. have been designed
and used in practice. These models are applicable to IoT systems, however, most of the
IoT systems in practice use centralized approach to manage accesses based on these models.
Centralized approach may work for data centers and other similar systems, but will incur
problems in IoT systems. This is due to the pervasive nature of the IoT systems, i.e., IoT
devices are dispersed all over the edge of the Internet. Decentralized solutions have been
considered in academia, mainly using blockchain technology.
Though there are existing works applying blockchain for access control in IoT systems, there
are limitations in these works, including insufficient peer-to-peer networking support and
inefficient access control protocol. We develop a novel Blockchain Embedded Access Control
(BEAC) framework to tackle these.
First, we use libp2p library with additional protocols to achieve a virtually universal domain.
It supports decentralized identity authentication and facilitates peer discovery, addressing the
common networking problems such as dynamic IP address resolution and firewall navigation.
Importantly, our peer-to-peer network solution ensures continuous service for roaming users,
maintaining functionality even during temporary disconnections from the blockchain.
Second, we design our blockchain based access control (BEAC) mechanism with several
desirable features. It supports multiple domains, each governed by its specific access control
model. To demonstrate the flexibility of our BEAC protocol, we apply it to Discretionary
Access Control (DAC), Attribute-Based Access Control (ABAC) and Role-Based Access
Control (RBAC) models. Also, with the support from our peer-to-peer overlay, our BEAC
facilitates user mapping across domains with varying access control models. Moreover, our
BEAC design decision is to embed access control policies in the blockchain, thus, it achieves
resilience, can fully recover from crashes by reconstructing their state from the blockchain.
Third, the blockchain based BEAC protocols in our BEAC framework is designed with
efficiency as a critical goal. Unlike some other works that utilize platforms such as Ethereum or
Hyperledger Fabric, our BEAC framework employs a custom consortium blockchain optimized
for access control, leveraging a Byzantine Fault Tolerant (BFT) consensus protocol and a
Jellyfish Merkle Tree. These design choices bring the basic performance advantages over other
platforms. More importantly, to ultimately enhance our BEAC protocol performance and avoid
the high messaging overhead in general blockchain based access control protocols, we introduce
a shortcut protocol and design several device hierarchy based protocols that effectively cutdown
the number of message rounds and achieve two to three folds of performance gain.
In our shortcut protocol, trusted users are allowed to directly access IoT devices in parallel
with registering the acidity in the blockchain. This is achieved by the design of secure
authorization tokens and corresponding token issuing and validation protocols. For access
request handling, we require that both the device domain and the blockchain service to
validate the access rights of the request against the policies. With a little additional overhead
due to duplicated processing, we enable a significant reduction in access latency (specifically,
full access validation is done in parallel with local access authorization). This can benefit
accesses by permanent users, such as the owner, and by recurrent accessors, which occur
frequently in real-world access patterns. This shortcut protocol not only greatly enhances
protocol performance, but also address internet connectivity issues, allowing users to access
local IoT devices even when they lose Internet connectivity.
In enterprise settings, where managing a large number of IoT devices is challenging and time
consuming, we design a Resource and Role Hierarchy-Based Access Control (RRBAC) model
which not only considers role hierarchy as in the RBAC model but also organizes IoT devices
into a resource hierarchy. The access rights assigned to a parent resource group is propagated
to the entire subtree. This design enables permission assignment and validation being done
in a group base. Accordingly, we design highly efficient data structures and algorithms
for the RRBAC model to realize the potential performance gain for RRBAC performance
assignments and validations. Our RRBAC design is shown to reduce computation time by
63% compared to traditional flat RBAC models in a resource hierarchy of size 10,000,000.
We extend the concept of RRBAC and design the R&D-BAC access control algorithms
for performance enhancements in our BEAC framework. Similar to the shortcut protocol,
our R&D-BAC algorithms allow accesses to the IoT devices in a designated subtree to be
authorized together, instead of individually as in traditional protocols. A user can cache
one authorization token and use it to access the group of IoT devices. The IoT devices, in
turn, are able to validate the token and allow the accesses. This protocol greatly reduces
the network message rounds and cutdowns the communication costs. Performance studies
show that the R&D-BAC access control algorithms can reduce the access times by 43% for
internet-based shortcut requests and 56.7% for local accesses.
Furthermore, we explore future research directions in BEAC. This includes examining the
potential of role-based tokens to enhance batch operation performance by assigning access
tokens to user roles instead of individual users. Additionally, we investigate the use of
smart contracts within BEAC, emphasizing the importance of deterministic execution and
programming simplicity. The structure of smart contracts in the Ethereum Virtual Machine
(EVM) is reviewed, and we propose eBPF (extended Berkeley Packet Filter) as an alternative.
eBPF offers a secure, resource-constrained environment that integrates XACML policies more
effectively than current smart contract languages such as Solidity and Vyper
Determining How ECM Mechanics Regulate Corneal Keratocyte Behavior in an in-vitro Model of Corneal Wound Healing
No full textDuring corneal wound healing, keratocytes located within the stroma are activated into a repair
phenotype by the release of soluble growth factors, such as transforming growth factor-beta 1
(TGF-β1) and platelet-derived growth factor-BB (PDGF-BB). The fibrotic response is often
accompanied by an increase in the corneal tissue stiffness. Previous studies have shown that TGF-
β1-mediated myofibroblast differentiation of corneal keratocytes is regulated by changes in
stiffness and prolonged myofibroblast presence can lead to corneal fibrosis and scarring, which
are leading causes of blindness worldwide. In vivo, corneal keratocytes encounter multiple growth
factors concomitantly, and PDGF signaling is known to be involved in regulating TGF-β1-induced
myofibroblast differentiation. Although previous work showed striking stiffness-dependent
phenotypic changes in TGF-β1-treated corneal keratocytes, it is unclear how changes in ECM
mechanics influence the keratocyte response in the presence of either PDGF-BB alone or when
treated with both TGF-β1 and PDGF-BB, simultaneously. The comprehensive transcriptional
profile underlying these phenotypic changes, their temporal dynamics, and their dependence on
stiffness are also not fully understood. Here, we used a polyacrylamide (PA) gel system to fabricate
susbtrata of tunable stiffness to determine how changes in substratum stiffness modulate the
corneal keratocyte behavior in response to either PDGF-BB alone or in the presence of both TGF-
β1 and PDGF-BB. To investigate the time-dependent transcriptional response during TGF-β1
treatment, we conducted bulk RNA sequencing on keratocytes cultured on collagen-coated glass
coverslips in the presence of TGF-β1 and quantified gene expression on days 1, 2, and 5.
Additionally, to study the effect of varying substratum stiffness on gene expression, we cultured
cells in the presence and absence of TGF-β1 on PA gels of 1 kPa and 10 kPa stiffnesses, as well
as collagen-coated glass coverslips, for 2 days, and in other experiments, for 5 days. Taken
together, our findings suggest that various biochemical and biophysical cues synergistically
regulate the behavior of corneal keratocytes during wound healing. Treatment with PDGF-BB
along with TGF-β1 appears to decouple molecular markers of myofibroblast differentiation from
the elevated mechanical phenotype typically associated with these cells. This implies potential
crosstalk in mechanotransductive signaling pathways downstream of TGF-β1 and PDGF-BB.
Additionally, although TGF-β1-treated keratocytes exhibit striking phenotypic differences in
response to changes in substratum stiffness, we do not observe corresponding transcriptional
changes associated with these stiffness-dependent responses. This suggests that the nuanced
relationship between ECM stiffness and cellular response extends beyond regulation solely
through mRNA transcript production
AI Driven Wireless Networks: Advances in Learning, Localization and Sensing
No full textMobile data has experienced phenomenal growth in recent years, with most of this data
being generated in real-time and distributed to edge nodes such as smartphones and vehicles.
The central processing unit (CPU), memory, and battery level in the client equipment (CE)
enable these devices to run complex artificial intelligence (AI) algorithms. Each client collects
large amounts of private data that may have sensitive information. In addition, current AI
algorithms are predominantly centralized requiring a server to collect data from the clients
to train a powerful learning model. However, this centralized approach is often impractical
due to limited communication bandwidth together with latency and privacy constraints.
Additionally, the collected dataset may contain private information, making it a valuable
target for malicious attacks. Federated learning (FL) has recently emerged as a powerful
alternative approach to solve these problems by enabling edge devices to collaboratively train
the learning model using real-time data. In FL, the server orchestrates the participation of
clients in the training process while keeping their data locally. Specifically, a server updates
a global model by averaging local models computed using local data and transmitted by
participating clients. The updating of the global model using local models and the reverse are
iterated until convergence. Transmitting high-dimensional models over wireless links is very
challenging due to the scarcity of radio resources and the uncertainty of wireless channels.
Moreover, FL performance depends on the reliability of the wireless links, which we quantify in
this dissertation by the desired outage probability level, and on the communication resources
(transmit power and bandwidth) of each FL client. These considerations motivate us to
design reliable, low-latency, multi-access, privacy-preserving, and energy-efficient resource
allocation schemes by jointly considering edge learning and wireless communication design
aspects.
Furthermore, radio frequency (RF) based indoor localization becomes critical for many
applications including asset tracking and indoor navigation, when the Global Positioning
System (GPS) is unreliable indoors. WiFi technology is an attractive solution for indoor
localization due to its ubiquity and low-cost in addition to enjoying wider coverage and
bandwidth compared to Bluetooth. We propose an AI based approach to utilize the channel
state information (CSI), which contains more information about the environment, to accurately
estimate the location. We first extract localization-related features such as time of flight
(ToF) and angle of arrival (AoA) and utilize them to determine the target device’s location.
Finally, to maintain privacy and security compared to camera-based systems, wireless sensing
is another promising technology. Integrating frequency modulated continues wave (FMCW)
radar with AI can effectively extract information about the surrounding objects and humans,
which can be used to count the number of people in a room or building, infer daily activities,
and track fall detection for elderly people. To demonstrate the potential of applying AI
to FMCW Radar data, we designed a framework to monitor the number of passengers in
a vehicle and to detect if a baby is left behind in the backseat. This framework provides
more advantages over the camera-based approach, such as privacy and robustness to different
illuminations condition
American Sign Language in Modern 2D Animated Media: Studying the Fundamentals of Deaf Representation in Cartoons
No full textThe intersection of animation and American Sign Language (ASL) presents a unique opportunity
to enhance communication accessibility for d/Deaf individuals and increase the visibility of Deaf
culture among hearing audiences. This research thesis explores the concept of animating ASL in
cartoon form as a way to bridge communication gaps and promote inclusivity. By leveraging the
expressive potential of 2D animation, ASL can be visually represented in dynamic and engaging
ways, capturing the nuances of sign language that static illustrations or written descriptions often
fail to convey. This thesis delves into the technical and artistic considerations involved in
animating ASL, including the use of character design, movement, and facial expressions to
convey meaning effectively. Additionally, it examines the intersection of the parameters of ASL
with the fundamentals of 2D animation—as established by the golden age of American
cartoons—to emphasize the beauty of the language in this medium. Furthermore, this thesis
discusses the cultural significance of properly animated ASL representations in mainstream
media, advocating for greater visibility and proper representation of d/Deaf and signing
characters and communities. Overall, animating ASL in cartoon form holds promise as a
powerful tool for fostering communication accessibility and promoting cultural understanding in
an increasingly diverse world
Social Connection Interspace: an XR Social Engagement Builder
No full textIn this project, I built a platform designed to help people interact with others worldwide, but with
the look and feel of being in the same room without language barriers in the Metaverse. With
Social Connection Interspace (SCI), my groundbreaking platform powered by Virtual Reality
(VR) and Large Language Models (LLMs), communication transcends borders and cultures,
fostering real-time connection and understanding. The SCI virtual environment redefines
collaboration and engagement. By harnessing the power of Mixed Reality, Artificial Intelligence,
Blockchain, and Big Data, it creates a seamless experience where people can connect, share
ideas, and reach consensus. SCI is more than just a Metaverse platform, it is also a doorway to
evolve the engagement ecosystem. This paper will share what I discovered and what is needed
when building a Metaverse platform, including AI, hand tracking, and consideration of
verisimilitude, how I prototyped all these elements to develop into one project, and how SCI can
benefit users and future researchers moving forward