80264 research outputs found
Sort by
How Food Insecurity Intersects With Health in Houston and Harris County
No full textIn Houston and Harris County, 2 in every 5 households are food insecure. As a foundational non-medical driver of health, food security plays a critical role in promoting an individual’s physical health and overall well-being. When households lack consistent access to enough nutritious food, known as food insecurity, residents are more likely to experience worse physical health. In spring 2024, over 5,200 members of the Greater Houston Community Panel (GHCP) shared their experiences with food insecurity, physical health, and health care access. This report compares findings between food-secure and food-insecure households to better understand how food insecurity status relates to physical health and health care differences in Houston and Harris County
WaLRUS: Wavelets for Long-range Representation using State Space Models
No full textState-Space Models (SSMs) have proven to be powerful tools for modeling long-range dependencies in sequential data. While the recent method known as HiPPO has demonstrated strong performance, and formed the basis for machine learning models S4 and Mamba, it remains limited by its reliance on closed-form solutions for a few specific, well-behaved bases. The SaFARi framework generalized this approach, enabling the construction of SSMs from arbitrary frames, including non-orthogonal and redundant ones, thus allowing an infinite diversity of possible "species" within the SSM family. In this paper, we introduce WaLRUS (Wavelets for Long-range Representation Using SSMs), a new implementation of SaFARi built from Daubechies wavelet
R3: Rice Research Review Winter 2025
No full textThis issue of R3: Rice Research Review focuses on research aligned with the university’s strategic plan, Momentous. The magazine covers four key areas: innovations in health, thriving urban communities, sustainable futures and responsible artificial intelligence (AI)
The Presence and Portrayal of Climate Change and Other Environmental Problems in Popular Films: A Quantitative Content Analysis
No full textHow have popular films responded to – or avoided – the climate and nature crises? To begin to answer this question systematically, quantitative content analysis was applied to 250 of the most-rated fictional films on IMDb that were released between 2013 and 2022. We found that climate change existed in 12.8% of these films, while a global environmental problem (climate change, freshwater pollution, marine pollution, air pollution, deforestation, species extinction and biodiversity decline, or toxic waste) existed in 26%; the presence of climate change, as well as common climate impacts, increased substantially over time; when climate change and other environmental problems were present, they were generally mentioned in just one or two scenes, and their gravity and/or urgency was not emphasized. As the first systematic, large-scale analysis of the presence and portrayal of climate change and other environmental problems in fictional narratives in the academic literature, this article illustrates a potentially productive area of future research, which is discussed
Amphiphilic Metal Complexes for Aqueous Photocatalysis and Photodynamic Therapy
No full textAmphiphilic metal complexes, also known as metallosurfactants, are metal-containing compounds that feature hydrophilic and hydrophobic components, enabling their interaction with aqueous and nonpolar environments. Integrating amphiphilic character into metal complexes has expanded their use in aqueous-phase reactions and biological systems. This thesis focuses on the design, synthesis, and application of ruthenium polypyridyl complex surfactants in two contexts: photoredox catalysis in water and photodynamic therapy.
Chapter 1 summarizes the fundamentals explored in this thesis: amphiphilic metal complexes, photoredox catalysis in aqueous solution, and photodynamic therapy. Chapter 2 presents an in-depth review of the amphiphilic ruthenium(II) complexes that have been developed and synthesized to date. Chapters 3 and 4 describe the experimental works related to the topics introduced in Chapter 1. In Chapter 3, we discuss the synthesis and application of a series of amphiphilic ruthenium(II) complexes with varying alkyl chain lengths attached to one of the pyridine ligands. The complex with the longest alkyl chain (16-carbon tail) has shown the most significant impact on the photophysical properties, e.g., photoluminescence quantum yield and excited-state lifetime. As a result, this surfactant has been utilized as a photocatalyst to drive photoredox transformations in water, with reasonable yields. Chapter 4 focuses on the design and synthesis of amphiphilic ruthenium(II) complexes tailored specifically for photodynamic therapy. For example, amphiphilic ruthenium(II) complexes with different hydrophobic tails were found to influence both their photophysical properties and their cellular uptake in cancer cells.
Finally, Chapter 5 presents an overview of the key findings and conclusions based on the research described in this thesis
Control of Hybrid Neuromodulation- Exoskeletons for Upper Limb Rehabilitation and Assistance
No full textMotor deficits following spinal cord injury (SCI) can limit an individual’s independence; therefore, restoration of upper limb function is a top priority of individuals with cervical SCI. Hybrid neuromodulation-exoskeleton systems show promise as rehabilitative and assistive technologies for individuals with SCI because they can offer rehabilitative benefits, provide assessment of kinematic and torque production, and provide functional movement assistance. This thesis focuses on hybrid neuromodulation exoskeleton systems as both an assessment platform and as an assistive device for individuals after SCI. One form of neuromodulation, transcutaneous spinal stimulation (TSS), can excite spinal circuitries, leading to muscle activity in otherwise paralyzed muscles, offering long-term rehabilitation benefits that last after the intervention. I characterize the motor response following cervical TSS at the upper limb muscle and joints with electromyography and a robotic exoskeleton. Another form of neuromodulation, functional electrical stimulation (FES), can be used to drive users through trajectories that mimic functional movements with higher precision than neuromodulation-driven movement and reduced torque consumption compared to an exoskeleton alone. I developed and validated a model-based nonlinear trajectory optimization methodology to create personalized trajectories for a hybrid FES-exoskeleton system that ensure dynamic feasibility for individuals with limited musculature, often seen following SCI. The assessment and movement assistance techniques presented could be used to help translate hybrid neuromodulation exoskeletons to clinical or home settings for combinatorial rehabilitation interventions or functional movement assistance for individuals with SCI
Seeking Spirit in the Land of Atheism: Narrative Motifs and Social Functions in the Soviet Esoteric Underground
No full textSoviet official rhetoric rejected any notion of esotericism and the paranormal as irrational superstitions, vestiges of the past unfit for progressive Socialist societies. Nonetheless, esotericism existed in the Soviet Union and played a substantial role both as a part of underground counterculture and as an important component dissolved in Soviet culture more generally, including science fiction, cinema, poetry, and philosophy. Esoteric interests also played an important role in Soviet engineering culture characterized by its interest to fringe topics such as telepathy and UFOs.
The goal of this dissertation is to analyze some of the main building blocks of Soviet esotericism, recurring narrative motifs that resurfaced in conversations related to esotericism and served as the foundation for esoteric worldviews. In particular, the analysis focuses on five motifs – mythical ancient civilizations, global conspiracies, Gnostic dualism, telepathic connection, and contact with extraterrestrial civilizations.
While this list is by no means exhaustive, it includes some of the most common recurring motifs of Soviet esotericism that, together, were sufficient to answer five key questions: “Where did we come from?” “How did we end up here?” “What is our current condition?” “What is our future potential on earth?” and “How does humanity fit in larger, cosmic-scale processes?” The structure of the dissertation roughly follows these five questions, from hypothetical primordial origins of humanity in the lost continent of Atlantis to speculations about other galaxies and space-faring future of humanity.
In an attempt to understand the ubiquity of esoteric narratives in Soviet society, my analysis emphasizes their social functions. Focusing on social functions of esotericism helps to assess why, in a society that on the surface distanced itself from esoteric ideas and often punished individuals for engaging with them, esotericism persisted and remained an important element of culture. I argue that esotericism played an important role as subversive power within Soviet culture that challenged the established status quo of the official Soviet ideology. As such, it had an ambiguous position as a potential danger from the point of view of Soviet authorities and, at the same time, as necessary source of cultural and technological innovation
Development of Nanomechanical Testing Methods and Nanomechanical Behavior of Materials Under Extreme Conditions
No full textNano and micromechanical experiments allow for a fundamental
understanding of how small-scale materials behave and, thus, what applications
these materials can be practically applied to. For example, graphene and hexagonal
boron nitride (hBN) are structurally similar. However, graphene is stronger, but
hBN is tougher. This fundamental insight enables engineers to design composites
balancing strength versus toughness. The more nanomaterials studied and the
greater the number of methods used to study nanomechanics, the more
nanomaterials are applied, likely as composites. Due to the unique properties of
nanomaterials, they are strong candidates to reinforce composites in extreme
applications such as high radiation environments. However, testing methods are
lacking at the nano- and microscale, as manufacturing new setups is complicated
and niche, and some existing testing methods have yet to be applied to
nanocomposites.
Many mechanical properties have yet to be directly measured at the
nanoscale (e.g. high strain rate tension and out-of-plane shear of a 2D material).
Nanoscale high-strain rate methods have been devised for 1D materials, but not all
methods are applicable to 2D materials. In this thesis, a high strain rate tensile
method is developed and validated on 1D PMMA. With this proof of concept,
multilayer hBN was subjected to high strain rate tension. This validated the
applicability of using a push-to-pull microdevice with a 2D material at high strain
rates. This method found a strain rate sensitivity in PMMA but not in hBN. Designs
for in- and out-of-plane shear devices are explored as well. Two separate material
platforms were investigated to understand how 1D and 2D reinforcing
nanomaterials affect the mechanical properties of composites with radiation
exposure at the microscale. Those composites are carbon nanotube-reinforced
silicon carbide (SiC) exposed to radiation and an hBN-reinforced covalent organic
framework (COF) exposed to radiation. Pillar splitting, a simple microscale fracture
toughness test, has yet to be performed on a nanomaterial-reinforced composite.
Pillar splitting SiC showed the method's limitations when samples are highly
defective. Nanotube reinforcement led to a weaker material likely correlated to
processing more than reinforcement, while radiation increased the toughness. The
hBN/COF composite showed that neutron radiation can alter the detectable bonds
in the composite and strengthen, harden, and toughen the composite
An SVD Guided Active Learning Approach for Medical Image Segmentation
Full text linkThe motivation behind this work is to two pronged: first, we seek to reduce the human effort required to create voxel level segmentation masks for medical images; second, we present a novel approach to improve the interpretability of convolutional neural networks used in medical image segmentation. This approach involves computing the truncated singular value decomposition of the entire model. To do this, we begin by providing the relevant background to convolutional neural networks and prove how they can be interpreted as input-dependent matrices for later discussion.
Following this, we provide enough relevant background to interactive image segmentation and active learning in order to present our `human-in-the-loop' approach to creating segmentation masks for medical images starting with few to no labeled images. We present results from three of our experiments.
The first experiment uses a `click-based' interactive convolutional neural network. We simulate a human user who follows one of three clicking strategies - placing centered clicks, random clicks, and boundary clicks. We determine that the clicking strategy does play an important role in model performance with centered clicks performing best. We additionally report that model performance can be significantly improved if the user is allowed to provide subsequent clicks to refine and edit predictions.
The second and third experiment evaluate our novel active learning sampling strategy against entropy and random sampling for binary liver segmentation and binary tumor segmentation. We investigate if it is worthwhile to have a user supply positive and negative patches for the unlabeled pool. To accomplish that, we define a metric to directly evaluate the model's performance in user-supplied clicked regions. This is in contrast to many common active learning strategies that attempt to estimate the model's \textit{confidence}. We use the dice loss function restricted to clicked regions to estimate the model's performance for the whole image. Our sampling strategy then selects the images with the largest dice loss in its clicked patches. We find that for liver segmentation, one of our click-based strategies offered the largest initial improvement in model performance and that all strategies outperform random sampling. For tumor segmentation, which is a much more challenging task, the results are noisier.
Next, we present our application of the singular value decomposition to convolutional neural networks. We first describe how we use the matrix formulation of a convolutional neural network to solve for a low rank approximation. Since the matrix is never explicitly formed, these solvers rely on implicit Krylov subspace solvers.
Then we detail how to model a `low rank convolutional neural network' using the low rank matrix approximation. We argue the singular value decomposition can be used to improve the interpretability of a convolutional neural network since it provides us with a means to analyze its range and nullspace.
Specifically, we extract three metrics from the singular value decomposition: the Left Projection Ratio, the Right Projection Ratio, and the Low Rank Fidelity Score. The Right Projection Ratio evaluates how well the input image is represented by its projection onto the computed right singular vectors. We show how the Right Projection Ratio provides a relative measure of how `close' an image is to the nullspace of the convolutional neural network. The Left Projection Ratio is constructed the same way but compares the image's label to the left singular vectors.
The Low Rank Fidelity Score directly compares the prediction from the convolutional neural network and its low rank approximation formulation.
We then present two experiments to validate our implementation.
In the first, we apply our approach to 2D image classification using the MNIST handwritten digits dataset since the rank of the associated matrix is at most the number of classes, 10. We show that the predictions from our low rank convolutional neural network steadily improve as the rank increases. Furthermore, we find that the rank 10 convolutional neural network obtained by our approach agrees perfectly with the original convolutional neural network. This validates our implementation and the theory that convolutional neural networks can be cast as matrices.
In the second experiment, we demonstrate how the Right Projection Ratio can be used as an a priori estimate for the dice score.
Finally, we revisit active learning with the Right Projection Ratio and Low Rank Fidelity Score in three experiments.
We first attempt to find a coreset using the Right Projection Ratio. While we do not observe a meaningful difference compared to random sampling, we report that the models trained on images with relatively larger Right Projection Ratios appeared to overfit, unlike the models trained on images with relatively smaller Right Projection Ratios.
Second, we create two new sampling strategies using both the Right Projection Ratio and the Low Rank Fidelity Score. These sampling strategies may have slightly outperformed random sampling, however the difference was clinically irrelevant.
Third, we compute larger rank approximations in an attempt to reach convergence between the low rank convolutional neural network and the original convolutional neural network
Navigating the End: Environmental Apocalypse and Responses of Muslims in Malaysia and Indonesia
No full textThis dissertation examines Islamic apocalyptic narratives in Indonesia and Malaysia, with a particular focus on how environmental crises are interpreted as divine warnings within these eschatological frameworks. Rather than viewing natural disasters, climate change, and ecological degradation as purely material phenomena, many Muslims in Indonesia and Malaysia understand these events through narratives of fear, anxiety, hope, and salvation—signs of the impending end times that demand moral reckoning and spiritual renewal. This study investigates how these apocalyptic interpretations shape religious consciousness, identity formation, and sociopolitical engagement, using a multidisciplinary approach that integrates anthropology, ethnography, sociology, and philology.
A central argument of this study is that Islamic apocalypticism functions as a dual adaptive mechanism: it can serve both as a stabilizing force and a source of social tension. On one hand, these narratives inspire movements centered on justice, ethical reform, and environmental responsibility, compelling individuals and communities to align their actions with divine mandates of righteousness. On the other hand, extremist groups and radicalized individuals have weaponized apocalyptic rhetoric to justify militant actions, framing violence as part of a sacred struggle. This inversion—where the same eschatological framework mobilizes both constructive and destructive responses—underscores the complex and contested nature of Islamic apocalyptic thought in Southeast Asia.
By examining case studies such as Darul Arqam, claimants of the al-Mahdī, and eschatological representations in children’s literature and comics, this dissertation traces how apocalyptic themes permeate both textual traditions and lived religious experiences. The study also engages with classical Malay and Indonesian manuscripts to contextualize contemporary narratives within a longer intellectual history. Beyond its scholarly contributions, this research offers critical insights into how religious responses to environmental crises influence social cohesion, radicalization, and resilience. These findings have broader implications for policy and practice, particularly in shaping community engagement strategies, counter-radicalization initiatives, and environmental advocacy efforts within faith-based contexts