马里兰大学

Carbon capture remains a crucial and growing area of research, driven by market demand for innovative and scalable solutions in CO2 abatement and related industrial reuse. Primary and secondary amines are essential for CO2 binding. This research investigates the radiation-induced graft polymerization of allylamine onto Fluorinated Ethylene Propylene (FEP) and Ethylene tetrafluoroethylene (ETFE) films. Analyses revealed minimal surface modifications and low nitrogen incorporation in the films. Because optimizing RIGP parameters is complex, a generative machine-learning framework is proposed to optimize the grafting process, supported by experimental data from other amine containing monomers grafted on to FEP. A dataset involving the grafting of 4-vinylpyridine onto FEP was selected due to data availability for model training. The ML framework is designed to be a customizable generative model in which the grafting process can be modeled by the algorithm of choice and optimized through iterative cycles of Gaussian Processes and Bayesian Optimization. This approach provided flexible learning and demonstrated strong predictive ability in identifying the optimal conditions for achieving a higher degree of grafting while minimizing experimental costs and resource use.

Pre-exposure prophylaxis (PrEP) is a highly effective intervention for preventing HIV transmission, but its high cost and uneven uptake raise key challenges for allocating resources efficiently. While spillover effects-wherein PrEP use in one group reduces infections in others-are known to occur, they remain poorly quantified and rarely guide policy. We provide a comprehensive modeling study, backed by data, for PrEP spillover effects in HIV risk populations, and develop both analytic and numerical tools for its quantification. Specifically, we first develop a novel compartmental model for HIV transmission that stratifies the total population into four interacting subpopulations: heterosexual males (HETM), high-risk heterosexual females (HETF-hi), low-risk heterosexual females (HETF-lo) and men who have sex with men (MSM). The asymptotic stability of the disease-free equilibrium of the model is analyzed. The spillover effect is directly quantified for this model by deriving an expression for the spillover-adjusted number needed to treat (NNT), a measure of the population-level impact of PrEP uptake in one group on disease incidence in others. Simulations show that PrEP delivery to MSM yields substantial indirect benefits, particularly for HETF-lo, where the spillover effect exceeds the direct effect by a factor of five. Furthermore, we show that targeting HETF-hi outperforms direct PrEP delivery to HETM, emphasizing the importance of intra-group heterogeneity. To evaluate whether these results hold under more detailed assumptions, we embed our framework into the national HOPE model maintained by the Centers for Disease Control and Prevention (CDC) and conduct global sensitivity analysis using Sobol indices with Polynomial Chaos Expansion. This approach extends our analytical insights and quantifies how uncertainty in PrEP allocation strategies propagates through complex epidemic dynamics. Further, this framework provides a numerical procedure for quantifying spillover effects in settings where direct mathematical analysis is impractical (or impossible). Our results demonstrate that spillover effects are a central driver of PrEP dynamics and that failing to account for them risks mis-allocating of control resources. This study provide both analytic and numerical methods for realistically quantifying PrEP spillover effects across models of differing complexity, bridging the gap between interpretable theoretical insights and high-dimensional national-scale simulations.

Sustainable construction requires effective control over both economic efficiency and environmental performance, particularly in terms of construction costs and greenhouse gas (GHG) emissions. However, current practices often lack real-time, data-driven frameworks to dynamically assess the economic and environmental impacts of ongoing construction activities. This study proposes a vision-based system that enables real-time monitoring, assessment, and management of construction costs and GHG emissions, contributing to low-carbon and resource-efficient construction practices. Three core contributions are presented: (1) a comprehensive real-world construction resource dataset covering 12 object categories is developed to address the scarcity of domain-specific training data; (2) a modular visual analysis framework is designed to automatically translate surveillance video data into quantitative cost and GHG emission metrics in real time; and (3) an earned value management-based deviation diagnosis mechanism is integrated to benchmark actual performance against budgeted expectations at the resource and activity level, enabling targeted interventions during ongoing construction. Experimental validation on a real-world construction project demonstrates that the system achieves high detection accuracy (mAP@0.5 = 92.5%), rapid processing (714 FPS), and near-instantaneous output (within 5 s), significantly reducing manual effort while enhancing decision-making. The proposed method represents a practical step toward integrating intelligent visual sensing with sustainable construction management, offering a practical tool for smart and green transformation of the construction industry.