Search for: All records

Abstract While woody root structures, such as bald cypress (Taxodium distichum) “knees,” can act as conduits of methane (CH₄), little has been done to explain variation from this flux pathway. We captured spatial (i.e., across knee surface, within sites, between sites) and temporal dynamics of CH₄from knees, and built empirical models to predict the contribution of knees to net CH₄fluxes. Knee and soil CH₄fluxes were measured across seasons within the lower Mississippi Alluvial Valley in a main channel (semi‐permanently flooded), side channel (seasonally flooded), and a reservoir edge (artificially flooded). Knees were a net source of CH₄across all seasons, even during periods of soil CH₄uptake. During periods of high knee CH₄efflux, fluxes varied across the knee surface, decreasing with height from the ground. Knee CH₄fluxes at the main and side channels decreased during a severe drought and increased ∼ ten‐fold in summer and two‐fold in winter following flooding events. At the reservoir edge, knee fluxes differed between the controlled draw up and draw down at the same water level, likely due to differences in temperature and oxygen availability. Knee CH₄fluxes were positively correlated with water level (measured from subsurface wells, above ∼−70 cm in the soil profile) and subsurface temperature, but the strength of the relationships differed across geomorphic positions. Cypress knees appear to be an important contributor to wetland CH₄efflux and accounting for the density of knees is needed to upscale their fluxes and better understand their ecosystem contribution. 

Layered “mosaic” metal-halide perovskite materials display a wide-variety of microstructures that span the order–disorder spectrum and can be tuned via the composition of their constituent B-site octahedral species. Such materials are typically modeled using computationally expensive ab initio methods, but these approaches are greatly limited to small sample sizes. Here, we develop a highly efficient hard-particle packing algorithm to model large samples of these layered complex alloys that enables an accurate determination of the geometrical and topological properties of the B-site arrangements within the plane of the inorganic layers across length scales. Our results are in good agreement with various experiments and, therefore, our algorithm bypasses the need for full-blown ab initio calculations. The accurate predictive power of our algorithm demonstrates how our minimalist hard-particle model effectively captures complex interactions and dynamics like incoherent thermal motion, out-of-plane octahedral tilting, and bond compression/stretching. We specifically show that the composition-dependent miscibility predicted by our algorithm for certain silver–iron and copper–indium layered alloys is consistent with previous experimental observations. We further quantify the degree of mixing in the simulated structures across length scales using our recently developed sensitive “mixing” metric. The large structural snapshots provided by our algorithm also shed light on previous experimentally measured magnetic properties of a copper–indium system. The generalization of our algorithm to model 3D perovskite alloys is also discussed. In summary, our packing model and mixing metric enable one to accurately explore the enormous space of hypothetical layered mosaic alloy compositions and identify materials with potentially desirable optoelectronic and magnetic properties. 

As rendering engines become increasingly important in film and television, with their use in virtual production (VP), some underlying issues become more apparent. This paper aims to investigate how we can improve asset color matching of VP elements with real-life objects found on sets. Experiments were conducted in which objects were exposed to various types of lighting setups, and digital twins were rendered using both RGB methods and spectral methods, with data reduction techniques also employed. The renderings were then filmed, alongside their real-life counterparts. Color difference metrics were used to determine whether spectral rendering and data reduction techniques offered advantages over RGB renderings. The conclusion illustrates that spectral rendering offers advantages, including higher accuracy in rendering the colours of materials. 

Atmospheric water vapor is an abundant and renewable resource that can alleviate growing water scarcity. Hybrid hydrogel desiccants composed of hygroscopic salts hold significant promise for atmospheric water harvesting (AWH) due to their increased capacity for water uptake. Thus far, many efforts in fabricating these desiccants require multistep processes, where the salt impregnation is achieved post-hydrogel fabrication. Here, we develop a scalable wet spinning methodology using aramid nanofibers (ANFs) to template and coagulate hydroxypropyl cellulose (HPC) into filaments in a coagulation bath consisting of water and lithium chloride (LiCl). HPC serves as the matrix to retain the captured water vapor, and later releases it upon heating. ANFs serve as the physical cross-linker between HPC, allowing for wet spinning at a speed up to 61 m h–1. The composite filaments achieve up to 0.55 g g–1 water uptake at 30% relative humidity (RH) and 21 °C, reaching 80% saturation in 40 min. With a lower critical solution temperature of 39 °C, the desiccant filaments can release up to 72% of the captured water at 60 °C after 30 min. In an AWH chamber, the filaments can achieve daily water production of 5.21 L kg–1 day–1 at 30% RH and 21 °C. 

In safety-critical robotic systems, perception is tasked with representing the environment to effectively guide decision-making and plays a crucial role in ensuring that the overall system meets its requirements. To quantitatively assess the impact of object detection and classification errors on system-level performance, we present a rigorous formalism for a model of detection error, and probabilistically reason about the satisfaction of regular-safety temporal logic requirements at the system level. We also show how standard evaluation metrics for object detection, such as confusion matrices, can be represented as models of detection error, which enables the computation of probabilistic satisfaction of system-level specifications. However, traditional confusion matrices treat all detections equally, without considering their relevance to the system-level task. To address this limitation, we propose novel evaluation metrics for object detection that are informed by both the system-level task and the downstream control logic, enabling a more context-appropriate evaluation of detection models. We identify logic-based formulas relevant to the downstream control and system-level specifications and use these formulas to define a logic-based evaluation metric for object detection and classification. These logic-based metrics result in less conservative assessments of system-level performance. Finally, we demonstrate our approach on a car-pedestrian example with a leaderboard PointPillars model evaluated on the nuScenes dataset, and validate probabilistic system-level guarantees in simulation. 

AI education is rapidly becoming essential as artificial intelligence transforms industries, yet students with disabilities often encounter significant barriers to learning and engagement. This paper examines accessibility challenges encountered by learners with visual, cognitive, and physical disabilities when using foundational tools for AI development. Using HuggingFace, an influential open-source platform, as a case study, we analyze barriers such as insufficient screen reader support, complex interfaces, and information overload. We propose design recommendations to promote equity and inclusivity in AI tools, aiming to empower diverse learners to thrive in AI education. Our work highlights the importance of inclusive design for CS educators, researchers, and policymakers. 

AI education is rapidly becoming essential as artificial intelligence transforms industries, yet students with disabilities often encounter significant barriers to learning and engagement. This paper examines accessibility challenges encountered by learners with visual, cognitive, and physical disabilities when using foundational tools for AI development. Using HuggingFace, an influential open-source platform, as a case study, we analyze barriers such as insufficient screen reader support, complex interfaces, and information overload. We propose design recommendations to promote equity and inclusivity in AI tools, aiming to empower diverse learners to thrive in AI education. Our work highlights the importance of inclusive design for CS educators, researchers, and policymakers. 

Concerns over foreign and domestic interference have raised questions about the legitimacy of U.S. elections. While research has explored election administration and public views on electronic voting, little attention has been given to election administrators’ perspectives. This study addresses that gap by examining how Georgia election officials perceive the use of electronic pollbooks (e-pollbooks) for voter check-in. The research hypothesizes that administrators view e-pollbooks as enhancing democratic legitimacy and election security. To test this, the paper presents findings from an NSF-funded online survey conducted two months before the 2024 general election. The survey was distributed to all 159 Georgia county election administrators and received IRB approval. It asked respondents to evaluate the security, reliability, ease of use, and fairness of various voter check-in systems, along with broader characteristics of elections in their counties. The results offer insight into how those managing elections assess the tools that support electoral integrity.