154092 research outputs found
Sort by
The Material Agency of a Large Language Model
Get PDFIn the field of ethics of LLMs, agency is a highly contested topic: Can LLMs perform agency? The current discourse on the agency of LLMs is heavily shaped by a traditional, properties-based, individualistic, and anthropocentric approach. In contrast, I propose a relational and network-based approach to agency, drawing on Actor-Network Theory and New Materialism, which is open to non-human agency and a plurality of agencies. Building on this relational, non-anthropocentric perspective, I also introduce a material perspective that has been largely overlooked in the LLM discourse. While LLMs are often thought of as immaterial, I develop a framework that distinguishes various dimensions of their materiality. I refer both to their material production and conditions, as well as how they concretely materialize in our devices and the materializing effects they have on us and our bodies. This perspective highlights the connection of LLMs with culture and material practices, cultural and social values: they are part of us, our culture, our time, politics, and our society. It also shifts the focus from human users as individual agents and opens up a view of the hidden network of multiple agents – human workers, big tech actors, environmental and technological non-human agents, and more – while drawing attention to the diversity and specificities of agencies. This shift transforms the questions currently posed in the ethics of LLMs and opens up new avenues for ethical inquiry
Continuous data assimilation closure for modeling statistically steady turbulence in large-eddy simulation
Get PDFA closure model is presented for large-eddy simulation (LES) based on the three-dimensional variational data assimilation algorithm. The approach aims at reconstructing high-fidelity kinetic energy spectra in coarse numerical simulations by including feedback control to represent unresolved dynamics interactions in the flow as stochastic processes. The forcing uses statistics obtained from offline high-fidelity data and requires only a few parameters compared to the number of degrees of freedom of LES. This modeling strategy is applied to geostrophic turbulence on the sphere and enables simulating indefinitely at reduced cost. The method accurately recovers the energy spectra and the zonal velocity profiles in the coarse model for three generic situations.</p
Generalized Crowns in Linear r-Graphs
Get PDFAn r-graph H is a hypergraph consisting of a nonempty set of vertices V and a collection of r-element subsets of V we refer to as the edges of H. An r-graph H is called linear if any two edges of H intersect in at most one vertex. Let F and H be two linear r-graphs. If H contains no copy of F, then H is called F-free. The linear Turán number of F, denoted by exlinr (n, F), is the maximum number of edges in any F-free n-vertex linear r-graph. The crown C1,3 (or E4) is a linear 3-graph which is obtained from three pairwise disjoint edges by adding one edge that intersects all three of them in one vertex. In 2022, Gyárfás, Ruszinkó and Sárközy initiated the study of exlin3 (n, F) for different choices of an acyclic 3-graph F. They determined the linear Turán numbers for all acyclic linear 3-graphs with at most 4 edges, except the crown. They established lower and upper bounds for exlin3(n, C1,3). In fact, their lower bound on exlin3(n, C1,3) is essentially tight, as was shown in a recent paper by Tang, Wu, Zhang and Zheng. In this paper, we generalize the notion of a crown to linear r-graphs for r ≥ 3, and also generalize the above results to linear r-graphs.</p
World of Forms:Deformable geometric templates for one-shot surface meshing in coronary CT angiography
Get PDFDeep learning-based medical image segmentation and surface mesh generation typically involve a sequential pipeline from image to segmentation to meshes, often requiring large training datasets while making limited use of prior geometric knowledge. This may lead to topological inconsistencies and suboptimal performance in low-data regimes. To address these challenges, we propose a data-efficient deep learning method for direct 3D anatomical object surface meshing using geometric priors. Our approach employs a multi-resolution graph neural network that operates on a prior geometric template which is deformed to fit object boundaries of interest. We show how different templates may be used for the different surface meshing targets, and introduce a novel masked autoencoder pretraining strategy for 3D spherical data. The proposed method outperforms nnUNet in a one-shot setting for segmentation of the pericardium, left ventricle (LV) cavity and the LV myocardium. Similarly, the method outperforms other lumen segmentation operating on multi-planar reformatted images. Results further indicate that mesh quality is on par with or improves upon marching cubes post-processing of voxel mask predictions, while remaining flexible in the choice of mesh triangulation prior, thus paving the way for more accurate and topologically consistent 3D medical object surface meshing.</p
TWISTER (Twente water injection system for turbulence experimental research):a jet array in the Twente water tunnel for generating strong turbulence using four-dimensional gradient noise
Get PDFWe present a newly constructed jet array with a novel driving scheme for turbulence generation in a vertical water tunnel and measurements of the turbulent flow this jet array establishes. The design of the array allows us to control the mean background flow and the turbulence intensity independently of each other. The array consists of a rectangular arrangement of 112 individually computer-controlled water jets that are aligned streamwise to the measurement section of our 8-m tall vertically recirculating water tunnel. Using solenoid valves, individual jets are activated following predefined protocols that can be tailored to obtain different turbulence statistics within the measurement section. The protocols are based on four-dimensional OpenSimplex noise, a type of gradient noise that features spatial and temporal coherence. Details of the mechanical and electrical designs are presented, together with a detailed description of the protocol generation. We show that the resulting turbulence is near homogeneous and isotropic, with a turbulence intensity of Order 1, an energy dissipation rate of order 10-1m2/s3 and Reλ≈1400. Additionally, we present experiments that show the effects that various system and protocol parameters have on the created flow conditions and address the streamwise development, as well as the homogeneity and isotropy of the flow.</p
An analytical predictor machine learning corrector scheme for modeling lateral flow in hot strip rolling
Get PDFRolling is a metal forming process where slabs are passed through rollers to produce strips with specific dimensions and mechanical properties. This process is performed in hot or cold formats. In hot rolling, the workpiece is initially heated above its recrystallization temperature. During the hot rolling process, plastic deformation occurs as the material’s thickness decreases and elongation takes place along the longitudinal axis of the workpiece. Due to the incompressibility of plastic deformation, the material also expands in the transverse direction, a phenomenon known as spread or lateral flow. Modeling spread is crucial for sustainability considerations and meeting customer expectations regarding the quality of the final product. Current prediction methodologies, such as the accurate but slow Finite Element (FE) method or the fast but inaccurate analytical metal forming analysis, are impractical for optimal control. To tackle these challenges, hybrid frameworks have emerged as a promising alternative. The present work aims to develop a fast and accurate model for predicting spread in hot rolling. Specifically, machine learning improves analytical models by leveraging data from a high-fidelity FE model. Initially, we review analytical models for spread, which address key aspects of the problem’s physics. To generate the ground truth (GT) space, an automated FE model for hot strip rolling is created. Moreover, the model’s sensitivity to both process and material parameters is investigated. In the Analytical Predictor Machine Learning Corrector scheme, the analytical models generate initial predictions of GT. In the correction step, a data-driven machine learning model is used to refine these predictions by compensating for deviations from high-fidelity FE simulations. The proposed hybrid framework improves the accuracy of the existing analytical models while preserving their computational efficiency.</p
Gigahertz thermoelastic acousto-optic modulation in lithium niobate integrated photonic device
Get PDFTailoring the interaction between acoustic waves and optical waves in integrated photonic circuits has emerged as a promising avenue for quantum optics and high-speed information processing. Among various approaches, on-chip acousto-optic modulation (AOM) has been extensively explored in the past decade and successfully demonstrated across many integrated photonics platforms. In this paper, we design and fabricate metallic gratings with periods varying from 0.4 μm to 3 μm deposited on a half-etched thin film lithium niobate (TFLN) device cladded by 1-μm thick silica. By illuminating the metallic grating with an intensity-modulated pump beam, we can generate thermoelastically driven SAWs with frequencies up to 6.58 GHz, and a linewidth of 1.8 MHz at a center frequency of 1.41 GHz. This thermoelastic approach eliminates reliance on a piezoelectric material response, offering broader compatibility for integrated photonics applications.</p
Ultrafast Light-Driven Electronic and Structural Changes in LaFeO<sub>3</sub> Perovskites Probed by Femtosecond X-Ray Absorption Spectroscopy
Get PDFConducting real-time, element-specific studies of photo-excited systems is a long-standing challenge. The development of X-ray free-electron lasers (XFELs) has paved the way for the emergence of a promising technique: femtosecond X-ray absorption spectroscopy (fs-XAS). This powerful technique reveals electronic and geometric characteristics, providing unprecedented insight into their dynamic interactions under nonequilibrium conditions. Herein, the fs-XAS technique is employed at PAL-XFEL to unravel light-driven ultrafast electronic and structural changes in epitaxial lanthanum iron oxide (LaFeO3) thin films. Density functional theory (DFT) and multiplet calculations are utilized to expound on the experimental results. The analyses reveal that photoexcitation initially induces high- and intermediate-spin Fe2+ states through ligand-to-metal charge transfer (LMCT), followed by polaron formation. It is demonstrated that the reduced overlap between the oxygen 2p and iron 3d orbitals accounts for all experimental observations, including 1) the XAS shifts to lower energies, 2) the decrease in the crystal field splitting, and 3) the relatively larger shifts observed in the oxygen 1s XAS.</p
Reduction of thermal conductivity by nanopillar inclusion in thermoelectric vertically aligned nanocomposites
Get PDFThe phonon-glass electron-crystal paradigm has guided thermoelectric research in recent years. However, the inherent conflict between atomic disorder reducing phonon conduction, and the order required to maintain high electron mobility, creates a significant challenge in material design, which has driven innovation in nanostructuring and composite materials. Here, vertically aligned nanocomposites (VANs) composed of self-assembled metallic La0.7Sr0.3MnO3 (LSMO) nanopillars in a surrounding ZnO matrix are investigated for controllable thermal conductivity. Tuning of the crystal orientation of the substrate controls the epitaxial alignment of the LSMO and ZnO phases along the horizontal and vertical interfaces. The VAN films on (111)-oriented STO substrates exhibit an increased power factor of 0.52 μW·cm−1·K−2 at 600 °C beyond ZnO films of 0.15 μW·cm−1·K−2. Detailed characterization and modeling of the thermal conductivity demonstrates a reduction of about 75% as well as anisotropic behavior for the VAN films with out-of-plane and in-plane thermal conductivities of respectively 9.2 and 1.5 W·m−1·K−1, in strong contrast to the isotropic behavior in ZnO films with a thermal conductivity of 38 W·m−1·K−1. These results show the promising strategy of VAN thin films with a nanopillar-matrix architecture to scatter phonons and to enhance the thermoelectric performance.</p
Spectral characterization of plant diversity in a biodiversity‐enriched oil palm plantation
Get PDFAssessing plant diversity using remote sensing, including airborne imaging spectroscopy, shows promise for large-scale biodiversity monitoring in landscape restoration and conservation. Enriching plantations with native trees is a key restoration strategy to enhance biodiversity and ecosystem functions in agricultural lands. In this study, we tested how well imaging spectroscopy characterizes plant diversity in 37 experimental plots of varying sizes and planted diversity levels in a biodiversity-enriched oil palm plantation in Sumatra, Indonesia. Six years after establishing the plots, we acquired airborne imaging spectroscopy data comprising 160 spectral bands (400–1000 nm, at ~3.7 nm bandwidth) at 0.3 m spatial resolution. We calculated spectral diversity as the variance among image pixels and partitioned spectral diversity into alpha and beta diversity components. After controlling for differences in sampling area through rarefaction, we found no significant relationship between spectral and plant alpha diversity. Further, the relationships between the local contribution of spectral beta diversity and plant beta diversity revealed no significant trends. Spectral variability within plots was substantially higher than among plots (spectral alpha diversity ~82%–87%, spectral beta diversity ~11%–18%). These discrepancies are likely due to the structural dominance of oil palm crowns, which absorbed most of the light, while most of the plant diversity occurring below the oil palm canopy was not detectable by airborne spectroscopy. Our study highlights that remote sensing of plant diversity in ecosystems with strong vertical stratification and high understory diversity, such as agroforests, would benefit from combining data from passive with data from active sensors, such as LiDAR, to capture structural diversity