116964 research outputs found
Sort by
Entre Familia: Lay Summary
No full textEntre Familia is a collaborative evidenced-based service project in Hidalgo County that aims to increase HPV vaccination among Latino adolescents and young adults in order to reduce cervical cancer incidence and mortality, along with other associated health disparities, among Latinos in Hidalgo County.This lay summary outlines the goals, strategies, and significance of the Entre Familia cervical cancer prevention program funded by CPRIT. It describes the expansion of a previous promotora-led initiative to include boys and male caregivers, with outreach and education targeting Hispanic families in Hidalgo County. The summary explains the program’s approach, including group sessions, one-on-one education, clinic partnerships, and vaccine navigation support, and highlights its potential to improve HPV vaccine uptake and reduce cervical cancer disparities in the region.Liberal Art
Entre Familia: Trainings for use of Measurement Forms
No full textEntre Familia is a collaborative evidenced-based service project in Hidalgo County that aims to increase HPV vaccination among Latino adolescents and young adults in order to reduce cervical cancer incidence and mortality, along with other associated health disparities, among Latinos in Hidalgo County.This document provides detailed instructions for Promotora staff in the Entre Familia program on how to administer and complete program forms. It includes guidance for Forms A, B, C, G, D, F, G1, I, J, K, and Y, outlining each form’s purpose, use, filing procedures, and step-by-step completion methods. Some of the forms are in English and others in Spanish. The instructions ensure standardized and accurate documentation of participant eligibility, survey responses, outreach activities, session attendance, and follow-up processes throughout the program.Liberal Art
Entre Familia: Form Flowchart
No full textEntre Familia is a collaborative evidenced-based service project in Hidalgo County that aims to increase HPV vaccination among Latino adolescents and young adults in order to reduce cervical cancer incidence and mortality, along with other associated health disparities, among Latinos in Hidalgo County.This flowchart outlines the sequence of forms and procedures used in the Entre Familia program to guide Promotora staff in enrolling and following up with eligible community participants. It visually maps participant pathways based on eligibility, session attendance, and immunization navigation, and connects each step to specific forms such as A, B, C1/C2, F, G1/G2, H, and I. The chart helps staff coordinate educational sessions, track clinic appointments, and ensure consistent documentation across the program.Liberal Art
Deep-learning evaluation of nanoparticle scattering (D-LENS) with probability convolutional neural network and Debye scattering method
Get PDFSmall-angle X-ray scattering (SAXS) is a powerful technique for obtaining averaged information on the size, shape, and internal structure of self-assembled nanoscale systems. Traditionally, extracting quantitative parameters from SAXS data requires manual fitting of analytical models, a process that is time-consuming and dependent on expert input. While machine learning has recently been explored to automate this analysis, many existing approaches rely on computationally intensive simulations or algorithms not optimized for the nature of SAXS data, namely, one-dimensional arrays of spatially correlated data. In this work, we introduce a one-dimensional convolutional probabilistic neural network (1D-CPNN) designed to predict key structural features - radius, aspect ratio, and polydispersity index (PDI) - of polyelectrolyte complex micelles (PCMs) directly from SAXS profiles. The model is trained on synthetically generated SAXS data and evaluated against traditional machine learning methods without neural networks and standard SAXS analysis software. Compared to these alternatives, the 1D-CPNN demonstrates superior performance in both prediction accuracy and computational efficiency. This approach offers a scalable and accessible solution for SAXS data interpretation, particularly in small-lab environments where resources and computational power may be limited. Preliminary results also indicate the model’s potential applicability to a broader range of nanocarrier systems beyond PCMs, underscoring its versatility in nanostructure characterization.Mechanical Engineerin
The praxis of early career teachers : teaching for justice in the era of anti-woke
Get PDFPublic education in the United States is increasingly shaped by political efforts to restrict discussions of race, gender, and systemic oppression, fostering surveillance and self-censorship among teachers (Willard, 2024). This qualitative case study examines how six early career elementary teachers navigated these constraints while participating in a critical inquiry group, the New Teacher Collaborative (NTC), during the 2023–2024 school year. Grounded in critical pedagogy (Freire, 1970), critical praxis (Duncan-Andrade & Morrell, 2008), social justice curriculum reform (Banks, 1995; Picower, 2012), and ambivalent surveillance (Willard, 2024), the study draws on qualitative research including interviews, meeting transcripts, and participant artifacts to analyze how teachers engaged in cycles of critical praxis to design and implement justice-oriented K–6 social studies curricula in politically charged school environments.
Findings highlight the tensions between teachers’ commitments to social justice and the pressures of anti-woke policies. Teachers with transformative understandings of social justice engaged in counter-conduct pedagogies (Willard, 2024), while those influenced by additive multiculturalism (Banks, 1995) were more likely to equate diversity with justice, rarely pushing their students to engage in issues of justice beyond the scope of the district-provided curriculum. The study underscores the necessity of curiosity, criticality, and asset-based pedagogies in fostering critical praxis, demonstrating that teachers who viewed students, families, and colleagues as collaborators were better positioned to resist restrictive policies. While the NTC provided a space for resistance and mutual support, first-year teachers faced significant structural barriers, including unfamiliarity with curricula and limited collaboration. This study expands scholarship on critical inquiry groups (Navarro, 2018; Picower, 2007; 2011), illustrating their potential to mitigate isolation and sustain justice-oriented teaching in politically conservative contexts. Implications call for teacher education programs to distinguish between diversity and justice, ensuring new teachers develop transformative approaches to curriculum design. In an era of escalating censorship and privatization (Maharaj et al., 2024), sustaining justice-oriented teaching requires institutional advocacy and robust professional networks that challenge the constraints of ambivalent surveillance.Curriculum and Instructio
Critical practices without guarantees : strengthening the practice-theory connection of social studies preservice teachers
Get PDFIn this set of critical qualitative case studies, I examined how well pre-service social studies teachers were connecting the theory they are learning in their teacher education with their practices at their student teaching placement sites. Utilizing cultural historical activity theory (CHAT) and the core practice approach, I explored how thick the practice-theory connection was for social studies pre-service teachers and the program’s role for that connection. This was done through an investigation of how pre-service teachers understood the communities and rules of the two-worlds of the university and schools, pre-service teachers rationales for enacting critical pedagogy at their placements, and how the pre-service teachers were applying broader critical theoretical constructs they were learning in their program (i.e. humanizing pedagogy) in their student teaching pedagogy. The studies found that pre-service teachers understood the two-worlds of the university and the school as having a lot of overlap, especially in regards to their tools (historical inquiry) and rules (critical theory). The critical theory manifested itself in critical pedagogy, where they honed their political and ideological clarity in both sites to deepen their understanding of critical constructs. Specifically, the practice of critical historical inquiry was better grounded in critical theory, furthering their understanding of the purpose of the theory and allowing for more authentic, agentic enactments at their placement sites.Curriculum and Instructio
Efficient and robust (hybridized) discontinuous Galerkin methods with the application to magnetohydrodynamics
No full textThis dissertation develops efficient and robust (hybridized) discontinuous Galerkin (DG) methods for solving a broad class of partial differential equations (PDEs), with a particular focus on both compressible and incompressible visco-resistive magnetohydrodynamics (MHD). The proposed methods address key computational challenges, including resolving sharp gradients, reducing computational costs, and overcoming numerical stiffness while ensuring scalability on modern high-performance computing architectures. First, I introduce a unified hp-adaptive hybridized DG (HDG) framework for Friedrichs-type PDEs to enhance both efficiency and robustness. The framework leverages HDG’s built-in mortars to deal with non-conforming interfaces, enables an easy implementation of an upwind-based numerical flux and results in a parameter-free scheme. The well-posedness is established for both one-field and two-field systems and the framework is validated through numerical experiments on elliptic, hyperbolic, and mixed-type PDEs. Furthermore, I explore two hp-adaptivity strategies, one based on an error indicator and another on an adjoint-based estimate, demonstrating their effectiveness in handling solutions with strong gradients, discontinuities, and singularities. Next, I propose an embedded-HDG (E-HDG) method that significantly reduces computational costs while preserving the divergence-free and H(div)-conforming properties of the velocity and magnetic fields. By using continuous facet unknowns for vector fields and discontinuous facet unknowns for scalar variables, E-HDG achieves substantial reductions in degrees of freedom compared to conventional HDG, making it particularly advantageous for high-order three-dimensional problems. The well-posedness is established for linearized MHD equations, and the method is extended to nonlinear MHD via a Picard iteration, ensuring that the divergence-free property is maintained throughout. Numerical experiments confirm the accuracy, robustness, and efficiency of the method, with divergence errors that reach machine precision. Finally, to address numerical stiffness in compressible MHD simulations, an exponential-DG approach is developed that couples spatial discretization of DG with an exponential time integrator. By decomposing the governing equations into linear and nonlinear components, the stiffness of the system can be properly tackled by the method in the sense that high-order accuracy is maintained and the stability constraint is relaxed. Compared to the explicit RK4-DG method, our approach demonstrates enhanced robustness in enforcing the generalized Lagrange multiplier (GLM) divergence-cleaning technique, which is essential for controlling divergence errors that can compromise numerical stability. On top of that, the approach achieves comparable computational costs while allowing significantly larger time steps. Furthermore, leveraging Krylov-subspace matrix-free exponential time integration and DG’s compact communication stencil, the method demonstrates both strong and weak scalability on modern supercomputers. Through these contributions, this dissertation advances the state of the art in DG-based numerical methods, particularly for MHD simulations, by providing a suite of efficient, robust, and scalable computational techniques for solving complex partial differential equations in scientific and engineering applications.Aerospace Engineerin
From diversity to adaptivity : effective multitask learning and continual learning neural architectures
Get PDFDespite remarkable successes from recent advances in deep learning, there remain many challenges. This dissertation focuses on two specific challenges: effectively optimizing a linear combination of multiple loss functions and enabling continual learning for deep neural networks.
The first part of the thesis addresses the challenge of optimizing loss functions composed of heavily conflicting components. In deep learning, it is common practice to optimize a combination of several loss functions to satisfy multiple desiderata. However, standard optimization techniques can lead to poor local minima for these problems, as different loss functions often conflict with one another. A few loss functions with dominating gradients may tend to be disproportionately optimized, thereby dictating the entire optimization trajectory. To mitigate this issue, we propose a method to quantify the local conflict and design algorithms that minimize the overall loss following trajectories that achieve a more balanced descent of the individual sub-losses. Empirical results demonstrate that these methods produce better local minima.
The second focus of the dissertation is on continual learning, enabling deep networks to autonomously adapt. Traditional deep learning models become static after training. We present two methods to overcome this limitation: dynamically expanding the network’s architecture in response to new data, and designing architectures that inherently support online learning. These innovations allow networks to autonomously update and adapt over time, reducing the need for frequent retraining.
Together, these efforts take steps to enable deep learning models to learn from diverse loss functions and adapt continually to new data and problems.Computer Scienc
Bayesian nonparametric methods for heterogeneous treatment and mediation effect estimation
No full textThis dissertation develops Bayesian nonparametric methods for estimating heterogeneous causal mediation effects and treatment effects with complex outcomes. We introduce Bayesian Causal Mediation Forests (BCMF), a varying coefficient model based on Bayesian additive regression trees that estimates and carefully regularizes causal mediation effects. This framework is then extended to accommodate ordinal mediators, heteroskedastic variances, zero-inflated outcomes, and continuous treatments to enable more accurate modeling of real-world relationships. We also develop Bayesian nonparametric quasi-likelihood methods for estimating heterogeneous treatment effects with non-Gaussian outcomes, providing robust inference and reliable uncertainty quantification while relaxing restrictive distributional assumptions. The applicability of our proposed methods is demonstrated through comprehensive simulation studies and applications to real-world datasets, including the Medical Expenditures Panel Survey (MEPS), National Medical Expenditure Survey (NMES), RAND Health Insurance Experiment (HIE), and National Health and Nutrition Examination Survey (NHANES).Statistic
Error bounds for stochastic user equilibrium traffic assignment
No full textIn stochastic user equilibrium traffic assignment, we derive bounds on the distance between a given feasible solution and the equilibrium solution. The intent is to provide guidance on termination criteria to reduce runtimes, which is important because the traffic assignment problem is often a subproblem of a more complex bilevel optimization. These mathematical bounds complement existing rules of thumb drawn empirically from numerical case studies. Our approach is based on Taylor's theorem, applied to the fixed-point formulation of the stochastic user equilibrium assignment, and provides upper bounds on differences in both aggregate metrics (average link travel cost, total system travel time) and disaggregate metrics (link flows, path flows). We demonstrate that the proposed path flow, link flow, and travel cost bounds are tight and cannot be further improved without additional restrictions on the network topology or problem instance. Results on city-level networks indicate that the bounds on path flows are reasonably tight, with the absolute difference between the derived bound and actual value often in the order of 10⁻¹. We study the tradeoff between the distance from equilibrium—where the bound becomes applicable—and its tightness. Detailed results guide the choice of a parameter that yields the tightest possible bound for a given distance from equilibrium. Empirical evidence shows that the bound on the distance between any feasible path flow and the equilibrium flow exhibits a near-linear rate of convergence close to equilibrium. This convergence rate is found to be close to 0.9, consistent across all tested networks. Finally, we demonstrate the practical utility of the derived bounds in a simple network design problem to find optimal link closures. By enumerating the complete feasible set, we show a 36% improvement in computation time on the Sioux Falls network when using the derived bounds, compared to not using them.Civil, Architectural, and Environmental Engineerin