154092 research outputs found
Sort by
The HAICu Project (WP2): Continual Machine Learning and Humans in the Loop
No full textThe HAICu (digital Humanities Artificial Intelligence and Cultural heritage) project is an interdisciplinary research initiative at the intersection of digital humanities, AI, and cultural heritage preservation. Work Package 2 (WP2) focuses on researching and developing machine-learning algorithms for the cultural heritage sector promotingcontinual, "life-long" learning, enabling multi-modal data mining of historical documents
The Effect of Crime Type and Knowledge about Autism Spectrum Disorder (ASD) on the Judgement of Defendants with ASD
No full textResearch indicates autistic defendants are treated more leniently than non-autistic defendants. However, it is unclear whether this is caused by positive perceptions of autistic people, or because autism is a relevant mitigating factor. Across two studies we tested how mock jurors evaluated autistic defendants in autism relevant and irrelevant criminal cases to better understand when and why autistic defendants receive leniency. Further, we provided some of the participants with educational information about ASD allowing these participants to recognise autism as a relevant mitigation for criminal cases for which autistic traits can be a risk factor (assault and stalking cases, but not burglary cases). Defendants with ASD only received lower sentences when autism was a relevant mitigation (assault and stalking), but not when autism was irrelevant (burglary). Thus, participants did not simply give preferential treatment to autistic defendants, but considered how ASD was related to the offense. However, providing information about autism weakened these effects, indicating educating people about autism may not always benefit autistic defendants. Our study highlights the importance of nuanced messaging about autism. While explaining how autistic characteristics can explain criminal behaviour may reduce perceptions of criminal responsibility, it may also imply greater dangerousness or risk of reoffending
Generative 3D reconstruction of Ti-6Al-4V basketweave microstructures by optimization of differentiable microstructural descriptors
No full textWe present a methodology for the generative reconstruction of 3D microstructures from 2D cross-sectional electron backscatter diffraction micrographs. The method is applied to Ti-6Al-4V processed by laser powder bed fusion, where a high amount of basketweave morphology is observed, which arises from the solid-state β→α-transition upon cooling. Prior-β-grain reconstruction is performed and the out-of-plane orientation of the observed grains is obtained leveraging Burgers orientation relationship. Microstructural descriptors related to convolutional neural networks are extracted from the 2D micrographs, and used for cross-section-based optimization of pixel values in a 3D volume. In order to reconstruct crystallographic orientations, the orientation distribution of the basketweave microstructure is reduced to a discrete set of characteristic orientations, which are sequentially reconstructed as separate components. Our reconstructions capture the characteristic lath morphology that is typically observed in powder bed fusion-processed Ti-6Al-4V and perform well in comparisons of chord length, as well as grain size, aspect ratio, and axis orientation distributions.</p
Dual-energy x-ray absorptiometry (DXA) in practice:a clinical centre survey endorsed by the European Association of Nuclear Medicine (EANM)
No full text[<sup>68</sup>Ga]FAPI PET/CT reveals increased pulmonary fibroblast activation protein expression in long COVID patients after ICU discharge
No full textPurpose: Post-acute sequelae of COVID-19 (PASC) has emerged as a major healthcare problem. A comprehensive mechanism of disease remains to be elucidated. In this study we aimed to explore pulmonary and muscle fibroblast activation protein (FAP) activity in former critical COVID-19 patients with persistent dyspnea, using [68Ga]FAPI-46 PET/CT.Methods: In this single center prospective observational study we included former critical COVID-19 patients reporting complaints of dyspnea > 3 months after hospital discharge. A [68Ga]FAPI PET/CT scan was performed including a high-resolution CT scan, lung function test, EQ-5D questionnaire, 6 min walking test and inflammatory markers. Age and sex-matched subjects, without pulmonary pathology, served as controls. The [68Ga]FAPI uptake was corrected for lean body mass and the target-to-background ratio (TBR) was calculated.Results: Eighteen PASC patients and 15 controls (median age 59 and 63 years and BMI of 34.6 and 25.2 kg/m2) were included. The interval between hospital discharge and study visit was 30 months. Increased pulmonary FAP expression was observed in PASC, (TBR 0.79 ± 0.23) compared to controls (TBR 0.40 ± 0.13, P < 0.001). Increased FAP expression was also observed in the paravertebral muscles (PASC: TBR 1.17 and controls TBR 1.00, P = 0.03). Forced expiratory volume and forced vital capacity showed moderate negative correlation with the pulmonary TBR, while the percentage of ground glass opacities showed a moderate positive correlation.Conclusion: [68Ga]FAPI PET/CT demonstrated elevated FAP expression in PASC. These findings provide insight into possible pathophysiological mechanisms of PASC and a potential new diagnostic modality.</p
An electronic gear concept for optimized efficiency operation of automotive converters
No full textIn this article, the concept of an Adjustable Hybrid Switch (AHS) converter is proposed and its benefits in improving the energy efficiency of an electric vehicle (EV) over the whole range of the driving profile is investigated. The AHS transforms a standard electric vehicle (EV) inverter into an advanced “Electronic Gear (E-Gear)” system by introducing a Cross-Switch (XS) hybrid semiconductor. Replacing the single transistor in a conventional EV inverter with a unique combination of unipolar Silicon Carbide (SiC) MOSFETs and bipolar Silicon (Si) IGBTs in parallel, the AHS allows for varying ratios of SiC MOSFETs to Si IGBTs, adapting to the EVs shifting load profile. Specifically, the converter controller selectively activates each transistor using dedicated gate units allowing it to choose the most efficient area ratio of SiC MOSFETs to Si IGBTs in the specific operating point of the vehicle's real-time torque-speed profile. A detailed analysis of the AHS converter featuring the E-Gear mechanism is performed in this paper. A water-cooled power train prototype is designed to verify the advantages of the system studied.</p
Detection Range Enhanced Highly Sensitive Niobate Nanosheet Pressure Sensors by Optimization of Substrate and Electrode Materials
No full textWith the widespread application of flexible pressure sensors in various fields, the demand for developing high-performance flexible pressure sensors is increasing. Two-dimensional oxide pressure sensors (2DOPSs) hold immense potential due to their ultra-high sensitivity. However, further improving the overall performance of the device remains a challenge for 2DOPSs, especially to widen the pressure sensing range of the sensor. In this article, the effect of substrate and electrode materials on the performance of the Ca2Nb3O10 nanosheet pressure sensors is investigated. It is found that the lower the Rockwell hardness of the electrodes of the device, the easier the substrate material deforms under pressure, and the lower the device pressure threshold, the higher the sensitivity. The optimization of the substrates and electrodes yields Ca2Nb3O10 nanosheet pressure sensors with sensitivity up to 105 kPa−1 in a large detection range, fast response time (83/165 ms) and good stability. This work provides guideline for further development of 2DOPSs.</p
Inline Mapping of Amorphous Silicon Layer Thickness of Heterojunction Precursors Using Multispectral Imaging
No full textIn this paper, we present an inline characterization technique to determine spatially resolved thickness maps of ultra-thin layers on textured silicon substrates. The technique is based on multispectral imaging and optical modelling of discrete spectral reflectance data using rigorous polarization ray tracing and the transfer matrix method. The study demonstrates that quantitative inspection of ultra-thin amorphous silicon (a-Si) layers on textured silicon substrates requires an extension of the standard RGB illumination by two additional LED wavelengths in the near-UV. As the required five images are measured in less than a second, the tool is a suitable candidate for inline applications. The optical modelling requires reflectance-calibrated images which are obtained via linear calibration functions and allows the a-Si thickness to be determined at each pixel. The thin-film thickness can be determined either by a direct modelling of the measured reflectance spectra or by a differential approach using the reflectance spectra before and after coating to eliminate effects from non-idealities due to scattering as well as instrumental errors. The a-Si thickness extracted from the reflection data at the five chosen LED wavelengths shows good quantitative agreement with reference values from spectrally-resolved differential reflectance data. Evaluating a test sample with an intentional a-Si thickness variation, we compared the results from the multispectral thickness map and reference values from spectroscopic ellipsometry. We found good quantitative agreement for a-Si thicknesses above 10 nm and a slight overestimation of about 1.5 nm for thinner layers. Overall, the multispectral approach based on only five different channels proves to allow quantitative thickness maps with reasonable accuracy at inline speed
PDE-Constrained Machine Learning with Gaussian Processes towards Digital Twins
No full textThe development of trustworthy digital twins for complex physical systems, often governed by high-dimensional partial differential equations (PDEs), demands predictive models that are both highly accurate and provide reliable uncertainty quantification (UQ). This thesis addresses the complementary weaknesses of two leading machine learning paradigms: the powerful expressiveness of deep neural networks (NNs), which often lack robust UQ, and the principled probabilistic nature of Gaussian processes (GPs), which struggle with the curse of dimensionality.This work presents a unified framework that synergizes these approaches through physics-constrained deep kernel learning (DKL). In this hybrid architecture, an NN learns a low-dimensional, feature representation, which in turn defines the kernel of a GP. This design leverages the representation learning power of NNs to make GP inference tractable and effective in high-dimensional settings.The core contributions are twofold. For forward problems, we introduce a PDE-constrained DKL model that effectively mitigates the curse of dimensionality. By embedding physical laws into deep kernel, the framework learns physically consistent solutions and provides reliable uncertainty estimates even from sparse data. For inverse problems, such as estimating unknown PDE parameters, we propose a novel two-stage Bayesian inference strategy to overcome computational intractability. An initial physics-informed pretraining stage optimizes the high-dimensional NN weights and provides robust point estimates. In the second stage, these weights are fixed, enabling efficient Hamiltonian Monte Carlo (HMC) sampling of the posterior distribution for only the low-dimensional PDE parameters and kernel hyperparameters.Collectively, this dissertation delivers a cohesive and robust framework for scientific machine learning. By systematically addressing architectural foundations, forward modeling in high dimensions, and tractable Bayesian inference for inverse problems, this work provides a significant step towards building trustworthy, uncertainty-aware digital twins for complex systems in science and engineering
Lifestyle behaviour change of patients following cardiac rehabilitation: the BENEFIT intervention study with one-year follow-up
No full textAims: The majority of people with cardiovascular disease do not maintain a healthy lifestyle. To help patients implement behaviour change at home, the BENEFIT programme was developed as an addition to cardiac rehabilitation (CR) care.Methods and results: Using a cluster non-randomized controlled trial design involving seven CR centres, we examined whether intervention group patients (n = 587) showed increased improvements in health behaviour change compared with control group patients (n = 298) who (only) received a multidisciplinary, comprehensive CR programme. Physical activity, smoking, alcohol use, diet, stress, and sleep were assessed at the start and after finishing CR (short-term) and at 1-year follow-up (long-term). Core of the intervention was access to an advanced eHealth platform consisting of functionality for daily goal monitoring, access to lifestyle interventions, personal coaching and a reward programme.Findings: The standard CR programme improved most lifestyle behaviours, while the intervention led to additional short-term changes in vegetable intake (t = 2.00, P = 0.023), work-related stress (z = −2.97, P = 0.002), and sleep hours (t = 2.57, P = 0.005). Finally, in contrast to the control group (t = 1.88, P = 0.415), the intervention group significantly increased its physical activity long-term (t = 5.04, P <0.001) exercising 42 min more per week, yet this group-interaction effect showed only a trend (t = 1.55, P = 0.061).Conclusion: While comprehensive CR care led to improvements in most lifestyle behaviours, the BENEFIT programme demonstrated additional benefits, particularly in physical exercise, dietary habits, stress reduction, and sleep, across a diverse CR-patient population. These findings underscore the potential of integrating eHealth solutions as an effective supplement to traditional CR care