19304 research outputs found
Sort by
A Multimodal Approach to Detecting and Assessing Objectionable Content in Online Media
No full textThis dissertation presents an innovative approach to the automatic evaluation of online video content, encompassing various modalities such as video, audio, and text. Online videos appear everywhere in people's daily lives, providing entertainment and knowledge; however, not all content is appropriate for every age group. For instance, videos with explicit or violent content may be inappropriate for younger audiences. To address content suitability, systems like the Motion Picture Association (MPA) film rating system and the Parental Advisory Label (PAL) for music have been developed. These systems aim to guide the age-appropriateness of media products, and streaming services have also undertaken efforts to offer guidance to their users. Traditional rating systems depend on expert analysis, which can be costly and time-consuming. Moreover, such systems typically yield a broad age-based suitability rating, lacking in detailed content insights that users may desire. In this dissertation, we address the challenge of detecting questionable content in online media, specifically the subcategory of comic mischief. This type of content combines elements such as violence, adult content, or sarcasm with humor, making it difficult to detect. Employing a multimodal approach is vital to capture the subtle details inherent in comic mischief content. To tackle this problem, we propose a novel end-to-end multimodal system for the task of comic mischief detection. As part of this contribution, we release a novel dataset for the targeted task consisting of three modalities: video, text (video captions and subtitles), and audio. We also design a HIerarchical Cross-attention model with CAPtions (HICCAP) to capture the intricate relationships among these modalities. We employ a hybrid pretraining approach that merges contrastive learning with multimodal matching tasks allowing for the joint learning of language, audio, and video representations. Our system also innovatively employs automatic video captioning to compensate for missing subtitles, thereby enriching the dataset and improving detection efficacy. The results show that the proposed approach makes a significant improvement over robust baselines and state-of-the-art models for comic mischief detection and its type classification. This emphasizes the potential of our system to empower users, to make informed decisions about the online content they choose to see
Multi-Dimensional Orientable Framelets with Compact Support
No full textSignal processing, evolving from its classical roots, integrates contributions from applied mathematicians to form a robust theoretical framework applicable across diverse signal types. Harmonic analysis, pivotal in signal denoising and compression, bridges disciplines like electrical engineering, applied mathematics, and computer science. Emerging in the late 1980s, wavelets provided sophisticated tools for analyzing discontinuous signals and multidimensional data. Shortcomings in wavelet designs spurred innovations like shearlets and curvelets, which excel in approximating multivariate functions with edge-type discontinuities by incorporating multiple orientations. However, conventional shearlets and curvelets lack compact support in space, limiting their utility. Addressing this, recent research introduced discrete compactly supported shearlet-like functions forming frames or Parseval frames. This approach combines advantages of compactly supported wavelets—small support and vanishing moments—with curvelets’ directionality and anisotropy, using Singular Value Decomposition for computation efficiency. This dissertation extends previous work by Karantzas et al. and addresses critiques by Thacker regarding compactly supported framelets, verifying constructions in Sobolev spaces. Contributions include novel proofs for space identification techniques and a main theorem on small reconstruction errors in Sobolev spaces. The theorem shows that we can uniformly control the reconstruction error E(f) on the Sobolev space Hs, when s > 32. It is supplemented by a corollary and simulations with real Sobolev functions highlighting practical implications
Foundational Reading Skills: Implementation in Secondary Classrooms
No full textBackground Across the nation, many states require educators to take professional development in the science of reading, a body of scientific research explaining how students learn to read. However, this requirement is often restricted to elementary teachers, and unfortunately, the research on supporting middle and high school students is mainly limited to small-group interventions. The minimal research on secondary literacy, the fact that less than one-third of eighth graders are proficient readers in the country, according to the National Assessment of Educational Progress, and the lack of training required for content-area teachers to teach reading indicate a need for a better understanding of supporting struggling readers in middle and high school classrooms. Purpose: This qualitative study with descriptive statistics investigated content-area middle and high school teachers (grades 6-12) who have gone through state-approved science of reading professional development to learn about their takeaways and what they are doing to support foundational reading skills in their classrooms. This study examined common practices that secondary teachers are implementing in their classrooms after professional development. The following question was analyzed: How are secondary teachers who have gone through state-approved training in foundational reading skills implementing reading instruction in their Tier 1 classrooms to support literacy? A second question analyzed was what factors enable a secondary teacher to implement foundational skills in their classrooms. Method: This study occurred in two phases. The first phase was a nationwide survey to gather information on secondary teachers. In total, 22 teachers were surveyed, and they had to provide a certificate of completion from a state-approved professional development program before taking the survey. The survey included required demographic questions, yes-or-no questions, Likert scale questions, and optional open-ended questions. The survey was analyzed to identify changes in teaching practices and perspectives before, during, and after professional development. The second phase of the study was a focus group with five teachers who volunteered to participate. Questions were asked to learn how teachers understand phonological awareness, word recognition, and fluency, and what changes occurred in their teaching practices because of the professional development. Using a thematic analysis approach, the survey results and focus group responses were collected but analyzed separately to confirm emerging themes. Results: The following four themes were identified in response to the research questions: (1) secondary teachers have students identify syllables in vocabulary words, (2) morphology instruction is included regularly in content-area classrooms, (3) teachers prioritize read alouds, partner reading, and repeated reading, and (4) teachers need to believe that they are responsible for teaching reading to incorporate foundational reading skills into their instruction. Conclusion: Secondary teachers can incorporate foundational reading skills in Tier 1 classrooms. However, to implement what is taught in state-approved science of reading professional development, teachers need to believe that they can help students learn to read. Without that belief, many teachers may complete a required professional development course but never change their instructional practices
Understanding Perceptions and Conversations with Augmented Reality Avatars
No full textThis study explores how individuals perceive and interact with spatially embodied augmented reality (AR) chatbots, focusing on their realism, social presence, and communication dynamics. Leveraging theories of social presence, conversational analysis and human-ai interaction. Participants, recruited from undergraduate Digital Media courses, interacted with an AR chatbot and participated in a semi-structured interview and quantitative survey. Data collection included Likert-scale responses and open-ended questions addressing perceptions of spatial presence, connectedness, and plausibility illusion.Information Science Technology, Department ofHonors Colleg
Effects of Pork Protein Ingestion Prior to and Following Performing the Army Combat Fitness Test on Markers of Catabolism, Inflammation, and Recovery
No full textTactical athletes and military personnel engaged in intense exercise need to consume enough quality protein in their diet to maintain protein balance and promote recovery. Plant-based protein sources contain fewer essential amino acids (EAAs), while pork loin contains a higher concentration of EAAs and creatine than most other animal protein sources. This study aimed to determine whether the ingestion of plant-based or pork-based military-style meals ready-to-eat (MREs) affects recovery from and subsequent Army Combat Fitness Test (ACFT) performance. <b>Methods:</b> Twenty-three (<i>n</i> = 23) University Corps of Cadets members participated in a randomized, double-blind, placebo-controlled, and crossover-designed study. Diets were prepared by a dietitian, food scientist, and chef to have similar taste, appearance, texture, and macronutrient content. The chef also labeled the meals for double-blind administration. Participants refrained from intense exercise for 48 h before reporting to the lab in a fasted condition with a 24 h urine sample. Participants donated a blood sample, completed questionnaires and cognitive function tests, and consumed a pre-exercise meal. After four hours, participants performed the ACFT according to military standards. Participants were fed three MREs daily while returning to the lab in a fasted condition at 0600 with 24 h urine samples after 24, 48, and 72 h of recovery. On day 3, participants repeated the ACFT four hours after consuming an MRE for breakfast. Participants resumed normal training and returned to the lab after 2&ndash;3 weeks to repeat the experiment while consuming the alternate diet. Data were analyzed using general linear model statistics with repeated measures and percent changes from baseline with 95% confidence intervals. <b>Results:</b> Results revealed that 3 days were sufficient for participants to replicate ACFT performance. However, those consuming the pork-based diet experienced less muscle soreness, urinary urea excretion, cortisol, inflammation, and depression scores while experiencing a higher testosterone/cortisol ratio and appetite satisfaction. There was also evidence of more favorable changes in red and white blood cells. Conversely, blood lipid profiles were more favorably changed when following a plant-based diet. <b>Conclusions:</b> These findings suggest that protein quality and the availability of creatine in the diet can affect recovery from intense military-style exercise. Minimally, plant-based MREs should include 6&ndash;10 g/d of EAA and 2&ndash;3 g/d of creatine monohydrate to offset dietary deficiencies, particularly in military personnel following a vegetarian diet. Registered clinical trial #ISRCTN47322504
Determination of TCR-Specific Fingerprints Through Structural Modeling
No full textT-cells are a crucial part of our adaptive immune system, responsible for identifying and fighting off foreign antigens. The activation of T-cells via their transmembrane T-cell receptor (TCR) can trigger the cellular immunity. These receptors can only recognize peptides presented by Major Histocompatibility Complex (MHC) molecules. Each TCR has a unique specificity for a particular peptide-MHC (pMHC) complex to better target the pathogens and avoid autoimmunity. However, in some cases, a TCR can recognize multiple pMHCs, a phenomenon known as T-cell cross-reactivity. Therefore, understanding the molecular interactions determining T-cell specificity is crucial for diagnosing autoimmune diseases and developing immunotherapies. Our proposal is that computational modeling of TCRpMHC structures can help us derive TCR-specific fingerprints and support the development of safer T-cell-based immunotherapies. There are several methods to predict the 3D structure of TCRpMHC complexes, including ImmuneScape, TCRpMHCmodels, and TCRmodel2. Our analyses show that there is room for improvement in this field, particularly regarding TCR docking pose prediction. We used molecular dynamics simulation (300 ns) to define the TCR fingerprints, which are the interaction profiles between the complementarity-determining regions (CDRs) of the TCRs and peptides. This experiment was performed for a dataset of 70 TCRpMHC complexes obtained from the Protein Data Bank (PDB). Our findings indicate that positions 4, 5, and 7 of the peptide are typically the key residues in TCRpMHC structures, but subtle variations on the importance of each position can be observed across complexes. We leveraged this information to predict the cross-reactivity of 17 cancer peptides in the dataset using our in-house developed tool, CrossDome. Our ultimate goal is to use these protocols to enable computer-aided evaluation and design of TCR-based immunotherapies for cancers and viral infections in the future
Integrating Color Theory into a Smart Desk Lighting System for Enhanced Productivity and Cognitive Function in Work Environments
No full textIn the age of remote and hybrid work, the quality of the personal workspace has a significant impact on cognitive performance, productivity, and overall well-being. Among the most influential yet often overlooked environmental factors is lighting. While studies have shown that lighting affects focus, mood, circadian rhythms, and mental clarity, most commercially available workspace lighting solutions remain static, one-size-fits-all, and disconnected from user needs. This thesis investigates how lighting can be intentionally designed to support focus, reduce stress, and enhance work experience through a human-centered design approach. Drawing from interdisciplinary research in lighting ergonomics, color theory, environmental psychology, and UX/UI design, the study explores how adaptable lighting conditions can influence user behavior, cognitive function, and workspace satisfaction. The project follows a comprehensive research process involving user surveys, environmental analysis, and performance-based testing to uncover lighting challenges and preferences in real-world desk environments. These findings will inform the design and development of a smart desk lighting system tailored to the needs of modern workers. The final product will integrate dynamic control, customizable lighting profiles, and aesthetic versatility to create a system that aligns with both functional and emotional workspace demands. This thesis contributes to the field of industrial design by demonstrating how research-driven, human-centered solutions can elevate everyday products and foster healthier, more productive environments in an evolving world of work
Perturbed Degrade-and-Fire Oscillations via Bernoulli Distributed Delay
No full textWe build on the results of a paper by Mather et al. on degrade-and-fire oscillations in two ways: we prove the existence of asymptotically stable periodic solutions using the Hopf bifurcation theorem for delay differential equations, and we consider the effect of a small perturbation via the inclusion of symmetric discrete time delays in the production rate. Proving the existence of stable periodic solutions to the original, unperturbed system was straightforward. We were able to calculate the bifurcating time delay exactly. The perturbed bifurcation values, on the other hand, could not be expressed analytically. Hence, we used regular perturbation theory to approximate them to the second-order. We found these correction terms to be quite small, implying that the qualitative dynamics are approximately the same in the perturbed system.Mathematics, Department ofHonors Colleg
Regulation of CBP Protein Stability by CARM1 Activity
No full textLymphoma is a type of blood cancer, originating from rapidly dividing white blood cells - an important component of the immune system. It is divided into two subtypes: Hodgkin and much more prevalent non-Hodgkin lymphoma (NHL). In contrasts to Hodgkin, NHL can arise from a more diverse group of lymph nodes, can spread unpredictably, and generally has a worse prognosis. Amongst all NHLs, Diffuse Large B-Cell lymphoma (DLBCL) contains these characteristics and is the most common accounting for 20 to 50% of cases depending on the country (Wang, 2023). The most common treatment for this aggressive lymphoma is a chemoimmunotherapy regimen called R-CHOP, consisting of a CD-20 antibody Rituximab and common chemotherapy drugs. Despite this regimen being quite successful in treating DLBCL, its nonspecific chemo components can cause other long-term toxicities such as cardiac issues and hormone imbalances (Watson, 2022). Therefore, more research is required to develop a more specific treatment for DLBCL to avoid this. To achieve this goal, I focused on CREB-Binding Protein (CBP), which is responsible for regulating access to many genes, including ones essential for cell growth, proliferation and apoptosis (Santos, 2023). To switch on CBP, Coactivator-Associated Arginine Methyltransferase 1 (CARM1) methylates it at specific Arginine residues, which means that a selective CARM 1 inhibitor will effectively turn off CBP function. This is particularly relevant in DLBCL, which often harbors CBP mutations that weaken its functions. Thus, a CARM1 inhibitor would turn off residual CBP activity, cutting tumor cells off from vital functions and killing it (Veazey, 2020). Further consultations with my mentors revealed that inhibiting CARM1 will lead to a decrease in CBP. However, it is yet unknown through which mechanism CBP stability can be controlled. At M.D Anderson, I set out with my mentors, Dr. Margarida Santos and Dr. Jee Won Hwang, to elucidate this mechanism with the hope that it will lay the foundation for a more specific and less toxic lymphoma treatment. [This project was completed with the contributions from Margarida Albuquerque Almeida Santos and Jee Won Hwang from UT MD Anderson Cancer Center.]Honors Colleg
Modeling Downstream Effects of Medium-Chain acyl-CoA Dehydrogenase Deficiency on Fatty Acid β-Oxidation and the Tricarboxylic Acid Cycle
No full textFatty acid β-oxidation (FAO) and the tricarboxylic acid cycle (TCA) are vital for cellular energy (Voet, 2016). Medium-Chain Acyl-CoA Dehydrogenase Deficiency (MCADD) disrupts fatty acid breakdown. If left untreated, individuals with MCADD can experience seizures, vomiting, coma, or death (Chang, 2000). Utilizing MATLAB, a computational model was built to simulate FAO and the TCA cycle in healthy and MCADD individuals. Based on Michaelis-Menten kinetics and enzyme data, the model investigates how reduced MCAD enzyme levels in MCADD affects substrate dynamics. The results show that in the FAO pathway of MCADD individuals, the concentration of intermediates is depleted rapidly, reducing the rate of consumption of the initial substrate, octanoyl-CoA, and reducing the rate of production of acetyl-CoA, which is detrimental to the subsequent TCA cycle. This model highlights the crucial role of the MCAD enzyme in maintaining metabolic homeostasis and displays the potential of computational models in understanding metabolic disorders. Future research could focus on investigating the compensation efforts made by other metabolic pathways to counteract the decrease in acetyl-CoA production in individuals with MCADD.Biomedical Engineering, Department ofHonors Colleg