150813 research outputs found
Sort by
From concept to manufacturing: evaluating vision-language models for engineering design
No full textEngineering design is undergoing a transformative shift with the advent of AI, marking a new era in how we approach product, system, and service planning. Large language models have demonstrated impressive capabilities in enabling this shift. Yet, with text as their only input modality, they cannot leverage the large body of visual artifacts that engineers have used for centuries and are accustomed to. This gap is addressed with the release of multimodal vision-language models (VLMs), such as GPT-4V, enabling AI to impact many more types of tasks. Our work presents a comprehensive evaluation of VLMs across a spectrum of engineering design tasks, categorized into four main areas: Conceptual Design, System-Level and Detailed Design, Manufacturing and Inspection, and Engineering Education Tasks. Specifically in this paper, we assess the capabilities of two VLMs, GPT-4V and LLaVA 1.6 34B, in design tasks such as sketch similarity analysis, CAD generation, topology optimization, manufacturability assessment, and engineering textbook problems. Through this structured evaluation, we not only explore VLMs’ proficiency in handling complex design challenges but also identify their limitations in complex engineering design applications. Our research establishes a foundation for future assessments of vision language models. It also contributes a set of benchmark testing datasets, with more than 1000 queries, for ongoing advancements and applications in this field
CuddleCard: Protocol for a randomized controlled trial evaluating the effect of providing financial support to low-income mothers of preterm infants on parental caregiving in the neonatal intensive care unit (NICU)
No full textBackground Preterm birth is a leading cause of childhood mortality and developmental disabilities, with persistent socioeconomic disparities in incidence and outcomes. Maternal presence during prolonged neonatal intensive care unit (NICU) hospitalization is critical for preterm infant health, enabling mothers to provide breast milk, directly breastfeed, and engage in skin-to-skin care—all of which promote infant physiological stability and neurodevelopment. Low-income mothers face significant barriers to visiting the NICU and participating in caregiving due to financial burdens and the psychological impact of financial stress. This randomized controlled trial aims to evaluate the effectiveness of financial transfers in promoting maternal caregiving behaviors that directly impact preterm infant health outcomes during NICU hospitalization. Methods We will conduct a two-arm, single-blinded randomized controlled trial with 420 Medicaid-eligible mothers of infants born between 24 weeks 0 days to 34 weeks 1 day gestation across four Level 3 NICUs in Georgia and Massachusetts. Mothers in the intervention arm will receive standard of care enhanced with weekly financial transfers and will be informed that these funds are intended to help them spend more time with their infants in the NICU. All participants will be provided with a hospital-grade breast pump and educational materials on the benefits of breast milk and skin-to-skin care. Participants will complete surveys during their infant’s hospitalization and following discharge, capturing outcomes related to maternal mental and physical health, caregiving behaviors, cognitive function, financial and socioeconomic factors, infant health and growth, and perceptions of NICU care quality. Primary outcomes are the provision of breast milk and engagement in skin-to-skin care. Secondary outcomes include infant growth and health outcomes, NICU visitation, financial and socioeconomic hardship, maternal physical and mental health measures, cognitive function, and perception of NICU care quality. Discussion This study will provide evidence of the impact of financial transfers on maternal caregiving behaviors in the NICU, addressing critical gaps in our understanding of how financial stress affects low-income mothers. Findings may inform health policy, particularly regarding Medicaid coverage of non-medical services, and contribute to understanding how to address disparities in preterm infant care. Trial registration The trial was prospectively registered with the American Economic Association Trial Registry, the primary registry for academic economists conducting policy trials, on 16 April 2024 (AEARCTR-0013256). It was also registered on ClinicalTrials.gov (NCT06362798) on 10 April 2024
BodyPrinter: Fabricating Circuits Directly on the Skin at Arbitrary Locations Using a Wearable Compact Plotter
No full textUIST ’20, October 20–23, 2020, Virtual Event, USAOn-body electronics and sensors offer the opportunity to seamlessly augment the human with computing power. Accordingly, numerous previous work investigated methods that exploit conductive materials and flexible substrates to fabricate circuits in the form of wearable devices, stretchable patches, and stickers that can be attached to the skin. For all these methods, the fabrication process involves several manual steps, such as designing the circuit in software, constructing conductive patches, and manually placing these physical patches on the body. In contrast, in this work, we propose to fabricate electronics directly on the skin. We present BodyPrinter, a wearable conductive-ink deposition machine, that prints flexible electronics directly on the body using skin-safe conductive ink. The paper describes our system in detail and, through a series of examples and a technical evaluation, we show how direct on-body fabrication of electronic circuits and sensors can further enhance the human body
Hydrogel Microparticle‐Templated Anti‐Solvent Crystallization of Small‐Molecule Drugs
No full textConventional formulation strategies for hydrophobic small‐molecule drug products frequently include mechanical milling to decrease active pharmaceutical ingredient (API) crystal size and subsequent granulation processes to produce an easily handled powder. A hydrogel‐templated anti‐solvent crystallization method is presented for the facile fabrication of microparticles containing dispersed nanocrystals of poorly soluble API. Direct crystallization within a porous hydrogel particle template yields core–shell structures in which the hydrogel core containing API nanocrystals is encased by a crystalline API shell. The process of controllable loading (up to 64% w/w) is demonstrated, and tailored dissolution profiles are achieved by simply altering the template particle size. API release is well described by a shrinking core model. Overall, the approach is a simple, scalable and potentially generalizable method that enables novel means of independently controlling both API crystallization and excipient characteristics, offering a “designer” drug particle system
Cryptographic Censorship
No full textWe formulate and take two large strides towards proving a quantum version of the weak cosmic censorship conjecture. We first prove “Cryptographic Censorship”: a theorem showing that when the time evolution operator of a holographic CFT is approximately pseudorandom (or Haar random) on some code subspace, then there must be an event horizon in the corresponding bulk dual. This result provides a general condition that guarantees (in finite time) event horizon formation, with minimal assumptions about the global spacetime structure. Our theorem relies on an extension of a recent quantum learning no-go theorem and is proved using new techniques of pseudorandom measure concentration. To apply this result to cosmic censorship, we separate singularities into classical, semi-Planckian, and Planckian types. We illustrate that classical and semi-Planckian singularities are compatible with approximately pseudorandom CFT time evolution; thus, if such singularities are indeed approximately pseudorandom, by Cryptographic Censorship, they cannot exist in the absence of event horizons. This result provides a sufficient condition guaranteeing that seminal holographic results on quantum chaos and thermalization, whose general applicability relies on typicality of horizons, will not be invalidated by the formation of naked singularities in AdS/CFT
Highly Efficient Carbon Dioxide Electroreduction via DNA-Directed Catalyst Immobilization
No full textElectrochemical reduction of carbon dioxide (CO2) is a promising route to up-convert this industrial byproduct. However, to perform this reaction with a small-molecule catalyst, the catalyst must be proximal to an electrode surface. Efforts to immobilize molecular catalysts on electrodes have been stymied by the need to optimize the immobilization chemistries on a case-by-case basis. Taking inspiration from nature, we applied DNA as a molecular-scale "Velcro" to investigate the tethering of three porphyrin-based catalysts to electrodes. This tethering strategy improved both the stability of the catalysts and their Faradaic efficiencies (FEs). DNA-catalyst conjugates were immobilized on screen-printed carbon and carbon paper electrodes via DNA hybridization with nearly 100% efficiency. Following immobilization, a higher catalyst stability at relevant potentials is observed. Additionally, lower overpotentials are required for the generation of carbon monoxide (CO). Finally, high FE for CO generation was observed with the DNA-immobilized catalysts as compared to the unmodified small-molecule systems, as high as 79.1% FE for CO at -0.95 V vs SHE using a DNA-tethered catalyst. This work demonstrates the potential of DNA "Velcro" as a powerful strategy for catalyst immobilization. Here, we demonstrated improved catalytic characteristics of molecular catalysts for CO2 valorization, but this strategy is anticipated to be generalizable to any reaction that proceeds in aqueous solutions
Neural scaling of deep chemical models
No full textMassive scale, in terms of both data availability and computation, enables important breakthroughs in key application areas of deep learning such as natural language processing and computer vision. There is emerging evidence that scale may be a key ingredient in scientific deep learning, but the importance of physical priors in scientific domains makes the strategies and benefits of scaling uncertain. Here we investigate neural-scaling behaviour in large chemical models by varying model and dataset sizes over many orders of magnitude, studying models with over one billion parameters, pre-trained on datasets of up to ten million datapoints. We consider large language models for generative chemistry and graph neural networks for machine-learned interatomic potentials. We investigate the interplay between physical priors and scale and discover empirical neural-scaling relations for language models in chemistry with a scaling exponent of 0.17 for the largest dataset size considered, and a scaling exponent of 0.26 for equivariant graph neural network interatomic potentials
Radiatively Cooled Magnetic Reconnection Experiments Driven by Pulsed Power
No full textMagnetic reconnection is a ubiquitous process in astrophysical plasmas, responsible for the explosive conversion of magnetic energy into thermal and kinetic energy. In extreme astrophysical systems, such as black hole coronae and neutron star magnetospheres, radiative cooling modifies the energy partition by rapidly removing internal energy. In this thesis, we perform experimental and computational studies of magnetic reconnection in a radiatively cooled regime, previously unexplored in reconnection studies. The Magnetic Reconnection on Z (MARZ) experiments consist of a dual exploding wire array, driven by a 20 MA peak, 300ns rise time current generated by the Z pulsed-power machine (Sandia National Labs). The load generates oppositely-directed supersonic, super-Alfvénic, collisional plasma flows with anti-parallel magnetic fields, that generate a reconnection layer (Lundquist number SL ∼ 100), in which the total cooling rate far exceeds the Alfvénic transit rate [mathematical notation].
Two- and three-dimensional simulations of the MARZ experiments are performed in GORGON, an Eulerian resistive magnetohydrodynamic code. The simulations demonstrate the generation of a reconnection layer, which radiatively collapses, exhibiting a rapid fall in temperature, strong compression, and an increased reconnection rate consistent with theoretical predictions. The reconnection layer is unstable to the plasmoid instability, generating secondary current sheets separated by magnetic islands. High energy X-ray emission is generated predominantly by the plasmoids. The plasmoids also collapse radiatively, and the reconnection layer recovers a laminar large aspect ratio structure, which does not exhibit further plasmoid generation, indicating stabilization of the original plasmoid instability of the current sheet.
The experiments confirm numerical predictions by providing evidence of plasmoid formation and strong radiative cooling. Experimental diagnostics directly measure the spatial, temporal, and spectral properties of radiative emission from the reconnecting system. The reconnection layer generates a transient burst of >1 keV X-ray emission, consistent with the formation and subsequent rapid cooling of the layer. Time-gated X-ray images show fast-moving (up to 50 km s−1) hotspots in the layer, consistent with the presence of plasmoids in 3-D resistive magnetohydrodynamic simulations. X-ray spectroscopy shows that these hotspots generate the majority of Al K-shell emission (around 1.6 keV), and exhibit temperatures (170 eV) much greater than that of the plasma inflows and the rest of the reconnection layer.
The findings in this thesis are of particular relevance to the generation of radiative emission from reconnection-driven astrophysical events, and to the global dynamics of reconnection in strongly cooled systems. The MARZ experiments also provide a novel platform for investigating radiative effects in high-energy-density and laboratory astrophysics experiments, and for validation of radiation magnetohydrodynamic and atomic spectroscopy codes.Ph.D
Marsh restoration in front of seawalls is an economically justified nature-based solution for coastal protection
No full textA marsh-fronted seawall is a hybrid nature-based coastal protection solution because it attenuates wave energy, reduces erosion, and provides ecosystem services. However, we still have a limited understanding of how to quantify the marsh wave attenuation benefits for economic analysis. Here, we incorporate a prediction of wave attenuation that accounts for species-specific morphology and structural stiffness into a 1-D wave model and validate it with field measurements. Our results show that the wave attenuation varies by a factor of two across different vegetation species. Further, we performed a benefit-cost analysis, in which the economic benefits represent the environmental services value and avoided seawall heightening cost that would otherwise be required to deliver the same overtopping rate without vegetation. We applied the model to a real-world, marsh-fronted seawall design at Juniper Cove, Massachusetts. Although the benefit of marsh-fronted seawalls is sensitive to discount rate, they have benefit-cost ratios greater than one, indicating that it is an economically justified nature-based solution. Further, we found that wave attenuation and benefit-cost ratio are more sensitive to water depth than wave height. Our study demonstrates the importance of considering the coastal protection of marshes and economic benefits in one framework
Deep learning initialized compressed sensing (Deli-CS) in volumetric spatio-temporal subspace reconstruction
No full textObject Spatio-temporal MRI methods offer rapid whole-brain multi-parametric mapping, yet they are often hindered by prolonged reconstruction times or prohibitively burdensome hardware requirements. The aim of this project is to reduce reconstruction time using deep learning. Materials and methods This study focuses on accelerating the reconstruction of volumetric multi-axis spiral projection MRF, aiming for whole-brain T1 and T2 mapping, while ensuring a streamlined approach compatible with clinical requirements. To optimize reconstruction time, the traditional method is first revamped with a memory-efficient GPU implementation. Deep Learning Initialized Compressed Sensing (Deli-CS) is then introduced, which initiates iterative reconstruction with a DL-generated seed point, reducing the number of iterations needed for convergence. Results The full reconstruction process for volumetric multi-axis spiral projection MRF is completed in just 20 min compared to over 2 h for the previously published implementation. Comparative analysis demonstrates Deli-CS’s efficiency in expediting iterative reconstruction while maintaining high-quality results. Discussion By offering a rapid warm start to the iterative reconstruction algorithm, this method substantially reduces processing time while preserving reconstruction quality. Its successful implementation paves the way for advanced spatio-temporal MRI techniques, addressing the challenge of extensive reconstruction times and ensuring efficient, high-quality imaging in a streamlined manner