116964 research outputs found
Sort by
Textile-based supercapacitors with integrated electrode-electrolyte structure for e-textile applications
Get PDFWith the emergence and development of flexible and wearable electronics, there is a growing demand for flexible, lightweight energy storage devices. These devices can be integrated into various applications, including biomedical sensors, soft robotics, and electronic textiles (E-textiles). However, conventional batteries and supercapacitors are rigid and bulky, making them unsuitable for wearable technologies. To address this limitation, textile-based supercapacitors (SCs) were developed using three different fabrication methods: fabric/thin-film integration, force-spinning, and wet-spinning. The goal of this research was to enhance the electrochemical and mechanical properties of SCs for potential applications in wearable and flexible energy storage systems. The first approach utilized commercial carbon fiber and activated carbon (ACG), with a PVA polymer film as the electrode matrix and PVA/H₃PO₄ dry film as the solid electrolyte. The incorporation of ZnO nanoparticles improved conductivity, and the optimized SC exhibited a specific capacitance of 1.321 mF cm⁻² at 10 mV s⁻¹. The device maintained stable performance under bending deformation and was capable of powering an LED for 30 minutes after charging with a solar cell, demonstrating its feasibility for flexible energy storage applications. The second approach involved a nanofiber-based SC fabricated using the force-spinning method, where PVA/H₃PO₄ fibers were deposited onto a fabric separator and spray-coated with ACG. ZnO nanoparticles were incorporated to enhance conductivity. The performance of the SC was systematically analyzed by varying electrolyte concentration, ZnO content, and ACG loading. Increasing H₃PO₄ concentration had the most significant impact, improving specific capacitance from 0.041 to 3.827 mF cm⁻². However, limitations in proton transport due to the non-woven structure hindered further improvement, and incomplete electrode coverage through spray-coating further constrained performance. The third approach used the wet-spinning method to fabricate an electrode-electrolyte woven fabric, embedding wet-spun PCL/ACG fibers into a PVA/H₃PO₄ hydrogel film to improve interfacial contact. The highest specific capacitance achieved was 0.541 mF cm⁻², comparable to thin-film SCs, but ionic conductivity remained low. Overall, while the developed SCs demonstrated promising electrochemical and mechanical properties, further advancements in active materials and electrolyte formulations are necessary to enhance their real-world applicability.Materials Science and Engineerin
Mayhem
Get PDFThis thesis project, MAYHEM, consists of an original completed feature-length screenplay that I wrote during the final year of my fellowship and an essay that describes the creative process behind it. I chose to write this screenplay because I wanted to tell a story that addresses Native American issues and highlights what it is to be mixed Native in relation to other multicultural perspectives. It was important for me to engage with material that would foster creative growth and provide opportunity to hone my craft as a writer.Writin
Response process validity of the ASL-SDQ with deaf youth : a cognitive interview study
Get PDFDeaf youth exhibit significantly higher rates of emotional and behavioral problems compared to their hearing peers. This population is extremely heterogeneous in terms of language and communication characteristics and many deaf youth are not provided access to a fully accessible language, American Sign Language (ASL). To screen, identify, and connect deaf youth to necessary mental health services, self-report measures of emotional and behavioral functioning must be accessible and comprehensible by all deaf youth signers, ranging from those who had full access to ASL from birth to those were/are deprived of full access to language. In addition, understanding how deaf youths' unique social and communication experiences shape their understanding of and responses to items on a mental health screener must be better understood to inform appropriate test score interpretation and treatment recommendations. The Strengths and Difficulties Questionnaire (SDQ), a well-validated and widely used mental health screener, has been translated into American Sign Language (ASL-SDQ). The validity or context applicability of the ASL-SDQ has not yet been evaluated by deaf youth. This study utilizes cognitive interviewing, an essential respondent-centered procedure that provides an avenue to evaluate the performance of a translated measure with the target population, identify comprehension difficulties or response errors, and understand how social location and cultural context may shape responses to items. This study conducted cognitive interviews with a community (n = 11) and clinical (n = 9) sample of deaf youth to evaluate the clarity and ease of responding to the ASL-SDQ, explore the experiences behind deaf youth's responses to the items, and identify additional questions deaf youth feel are important to include when assessing deaf youth's mental health. Results from the cognitive interviews suggest several modifications to be made to increase the clarity of the ASL- SDQ with deaf youth and inform assessment practices for language deprived deaf youth. Deaf youth also recommended assessing the impact of COVID-19, suicidal thoughts, and deaf-specific factors such as communication access at home, communication fatigue, and being treated differently by hearing peers. Finally, overall implications for measure translation for deaf youth are discussed.Educational Psycholog
Diffusion models : training, sampling, and reconstruction
Get PDFDiffusion models [Hyv05, SDWMG15, SE19] have recently emerged as the standard approach to generative modeling of images and video, largely due to their performance and reliability. Despite their popularity, several aspects of these models remain poorly understood. In this thesis, we make progress towards better understanding diffusion models from a fundamental perspective. We show: 1. The first polynomial sample complexity bounds for training a diffusion model, assuming access to a sufficiently expressive neural network. 2. A faster algorithm for sampling from diffusion models with provable guarantees, and the first theoretical guarantees for parallel sampling. 3. The first computational intractability results for posterior sampling from diffusion models, which captures the task of performing reconstruction, given a noisy linear measurement of a sample.Computer Scienc
Constrained pressure residual implementation In UTCOMP reservoir simulator
Get PDFIn the reservoir simulation community, the main focus of research related to Numerical Linear Algebra, is to develop a physics-based preconditioner that can be efficiently coupled to a well-known high performing Krylov-type iterative solver. This is due to the fact that the performance of these solvers can significantly depend on the preconditioning of the matrix. Usually, preconditioning is a mixture of experience, art and science and requires a domain specific knowledge of the problem being solved. The industry-standard, state-of-the-art preconditioner is the Constrained Pressure Residual. This report presents results related to the implementation of CPR on UTCOMP-RS∗.Petroleum and Geosystems Engineerin
Impact of Demographics and Resource Availability on Rural Life Expectancy and Healthcare
Get PDFThis study utilizes a mixed-methods approach, identifying the characteristics of rural residents and the impact on their life expectancy (LE). It also takes the observations and experiences of rural healthcare providers to better understand the benefits and challenges of practicing rurally. County-level data were analyzed using bivariate and ordinary least squares (OLS) regression methods to identify the impact of different variables on life expectancy. The qualitative data was obtained through a series of interviews with healthcare providers in which they were asked about their experience working rurally.
The study indicates a correlation between high smoking and inactivity rates with lower LE. It also found that LE is negatively associated with child poverty rate and median income. These findings point to the impact of health behaviors and socioeconomic status, and how this affects LE. There is a clear relationship between healthcare deserts and LE as well, emphasizing the importance of healthcare access. Through the interviews with healthcare providers, it is clear that rural areas lack specialty services. Healthcare providers often have to be creative in rural areas because of a lack of resources and the unique situations they face in rural hospitals. The findings of this thesis suggest a need for more healthcare resources in rural areas and reveal the challenges rural residents face.Health and Societ
Automating PCR with the UFACTORY xArm 6 Lite
Get PDFPolymerase Chain Reaction (PCR) is a fundamental technique in molecular biology, extensively employed in genetic analysis, disease diagnostics, and forensic science. Despite its critical role, the preparation of PCR reactions remains a labor-intensive process that relies on precise pipetting of reagents, a task susceptible to human error and variability. Manual pipetting can lead to inconsistencies in sample preparation, impacting the accuracy and reproducibility of experimental results. To address these challenges, computer automation has gained significant attention, aiming to improve efficiency, precision, and scalability in laboratory workflows.
This work presents an innovative robotic system designed to automate the pipetting process for PCR preparation using the xArm 6 Lite robotic arm. Our approach combines computer vision for precise object localization and volume verification, a custom 3D-printed gripper for micropipette manipulation, and motorized actuation controlled by an Arduino-based system. By leveraging the Robot Operating System (ROS) and Python, along with deep learning-based image analysis techniques, we enhance the precision and repeatability of the PCR process.Computer Scienc
Learning vision language models from a messy and unstructured world
Get PDFToday’s recipe for computer vision has been quite simple: collect data, annotate the data and fit your model to the data. However, this paradigm is slowly running out of steam. As models scale, they demand substantially more annotated data, with finer granularity and richer semantic concepts. The cost of annotation is rising, yet even our largest vision systems today recognize only a fraction of the concepts humans can effortlessly perceive. While there are millions of objects, attributes, and interactions in the real world, our best computer vision models remain constrained to well-defined categories in clean, highly structured datasets. At the same time, the Internet provides a vast reservoir of visual data—images, videos, and multimodal content—accompanied by rich but unstructured and noisy metadata. Unlike carefully labeled datasets, this data is abundant but messy: captions are incomplete, annotations are sparse, and labels vary in granularity. Yet, humans learn from such ambiguous information effortlessly, making sense of the world by integrating context, prior knowledge, and weak supervision signals. The goal of this thesis is to develop general-purpose vision-language models by fully exploiting the Internet data. We cover learning from unlabeled, weakly-labeled, heterogeneous, and finally from the raw Internet data. First, we begin with leveraging large amounts of unlabeled images to learn part segmentation for over 10,000 object categories. Starting with pre-trained object instance segmentation, we use a clustering-based unsupervised learning technique to obtain part-level pseudo masks for different object categories. With dedicated self-training, we achieve high-quality part segmentation for wide range of object classes with zero part-level mask annotations. Second, we use weakly-labeled and long-tail object detection data to train a state-of-the-arts zeroshot large-vocabulary object detector. This detector can readily work on any set of arbitrary concepts and can be trained 10-100 times more data-efficiently. The key idea is to first train a specialized object detector that only focuses on prompted visual concepts, which drastically improves pseudo-labeling quality. Training an open-vocabulary detector with this pseudo-labels pushes the state-of-the-arts of zero-shot, open-vocabulary, and large-vocabulary object detection. Next, we collect a long list of heterogeneous datasets for various different tasks, and re-purpose them to train a multimodal large language model (MLLM). More specifically, we train a MLLM that can reason in both 2D and 3D space for object grounding and complex reasoning tasks from image input. The result model can do free-form 3D grounding for indoor and outdoor scenes with limited amount of 3D data. Notably, we impose no architectural changes or training objectives, which makes it easy to integrate the new 3D capability into any MLLM. Finally, we build frontier-class VLMs from ground zero, starting from raw internet-scale images and videos with our data engine. We develop a robust data engine to annotate natural images, chart, document, diagrams, and videos, resulting in 65 millions of training data. Furthermore, we extend this data engine to construct human annotations of challenging video tasks such as fine-grained activity understanding and spatio-temporally grounded reasoning. We release the entire synthetic and human annotations, models, and training code for the communityComputer Scienc
The aestheticization of ugliness : the Pulcinella narrative in Giandomenico Tiepolo's Villa Zianigo
Get PDFWhat does one make of beautiful depictions of ugliness? In an attempt to answer this question, this paper used Giandomenico Tiepolo’s Pulcinella fresco series in Villa Zianigo from the late eighteenth century as a case study. The frescos foreground Pulcinella, a Commedia dell’arte theatrical character known for his ugly appearance— hunchback, hooked nose, and protruding belly— and actions on stage. In this series, the artist displaces the character from the stage and onto the walls of a quaint villa, in joyous scenes within the beautiful country landscape. Though the ugliness of his character is maintained, I claim that his recontextualization changes his relationship to the viewer, making him their stand-in. This recontextualization I consider to be a part of the aestheticization of ugliness. Pushing back against the widely purported understanding of aestheticization as beautification, this paper explores how the recontextualization of an ugly subject into a beautiful, artistic context strengthens the relationship between the subject and the audience. As a complement, this paper also argues for the possibility of a narrative within the seemingly disjunctive group of frescos. While at first glance each scene of village life seems isolated from the next, this paper, using comic narrative theory and close looking, builds a loose narrative to tie together the scenes. I liken the ambiguous and unpredictable nature of the fresco narrative to the improvisation technique that was the cornerstone of the Commedia theatrical tradition and made the audience active agents in the formation of the play. In focusing on the aestheticization of ugliness and the Commedia-esque fresco narrative, this paper becomes an exploration of the possibilities within viewer/subject relationship.Art Histor