263134 research outputs found
Sort by
Neurosymbolische AI unificeren en opschalen: van abstractie tot toepassing
No full textThe emulation of intelligence in all its aspects is poised to
revolutionise our way of living, working and interacting. Its
biggest success stories of the last two decades have given us
vision models that help healthcare workers with diagnosing
diseases more accurately than ever, game-playing models
that beat the best human players at the most complicated
games and language models that assist almost everyone by
answering almost any question. These successes have all
been mostly driven by the use of neural networks; machine
learning models that learn to predict from enormous amounts
of data, e.g. annotated images, self-played game sessions or
an entire internet worth of text. They have shown how far
we have come in just two decades, yet we might not be much
further in terms of actually emulating intelligence.
While learning from observations is surely an important
aspect of intelligence, it also became apparent in all the
success that learning alone has its limits, especially in terms
of generalisation and consistency. That is, if one can observe
a dense enough representation of all possibilities, then
learning works well. If that is not possible, then learning
alone can fail. In this thesis, we contribute to pushing
the boundaries of neurosymbolic AI, a field that integrates
learning with reasoning. The general thought process being
that reasoning excells at generalisation and consistency,
exactly what learning seems to miss, resulting in a naturally
symbiotic relationship.
However, how such an integration should look or how it
deals with the different kinds of concepts appearing in the
real world are still open questions. Even more, reasoning
itself in its most general form is inevitably computationally
burdened resulting in systems that have difficulties providing
answers on complex questions in a reasonable amount of
time. While tackling this computational burden of reasoning
in its full generality is infeasible, we focus on providing more
scalable algorithms for reasoning in various specific settings
from discrete-continuous domains to sequential data and
arithmetic. The algorithms are all tested on a wide range
of benchmarks to illustrate their behaviour and success.
Apart from scalability, we simultaneously propose a new
neural probabilistic logic programming language that supports
writing down background knowledge over both discrete and
continuous quantities. This new language is only one of
the many neurosymbolic system appearing in an increasingly
growing alphabet soup of systems. Hence, we will first
aim to abstract the many different approaches appearing
in neurosymbolic AI in a way that allows us to mathematically
define what a neurosymbolic model contains and how it
performs inference. This abstraction is shown to indeed unify
many different systems in a mathematically sound framework,
which further paves the way to new algorithms for scaling
joint learning and reasoning.status: Publishe
Faster Algorithms for Isogeny Computations over Extensions of Finite Fields
No full textsponsorship: We thank Lorenz Panny for helpful discussions which sparked the first algorithm. We thank Kaizhan Lin and the anonymous reviewers for helpful suggestions. Mingjie Chen and Christophe Petit are partly supported by EPSRC through grant number EP/V011324/1. (EPSRC|EP/V011324/1)status: Publishe
The Patient-Reported Outcomes Measurement Information System scale v1.2 Global Health (PROMIS-GH) is useful to assess mental and physical health in individuals with chronic musculoskeletal pain, rheumatoid arthritis, hip/knee osteoarthritis, or people undergoing physiotherapy: Results of factor and item response theory analyses.
No full textOBJECTIVE: This study aims to assess whether the 10-item Patient-Reported Outcomes Measurement Information System Scale v1.2-Global Health (PROMIS-GH) is useful to assess Global Mental Health (GMH) and Global Physical Health (GPH) in individuals with musculoskeletal disorders. METHODS: PROMIS-GH was administered to 4295 individuals (mean±SD age 56±14y, 70% female, chronic musculoskeletal pain [n=1142], rheumatoid arthritis [n=1987], hip/knee osteoarthritis [n=418], undergoing physiotherapy [n=947]). We investigated the legitimacy of calculating a GMH and a GPH subscale score (Confirmatory Factor Analyses) and of converting raw ordinal subscale scores to interval scores (checking IRT-assumptions [unidimensionality, local independence, monotonicity] and Graded Response Model fit), the ability of the subscales to discriminate different levels of health (items' discrimination parameters α), to cover relevant range of GMH and GPH (range of difficulty parameters β), their precision (item- and subscale-level information), and to compare demographical and clinical subgroups (Differential Item Functioning [DIF]). RESULTS: It is legitimate to calculate a GMH and a GPH subscale score (CFI=0.98/0.97, TLI=0.97/0.95, RMSEA=0.15/0.19, SRMR=0.05/0.07), and to convert raw scores to interval scores (IRT-assumptions met and model fit). Discrimination (items' α≥2), coverage (lowest β≤-2) and precision (reliability≥0.70 for large portions of the health domain) were adequate for all items, except item Global10, underscoring that they can distinguish individuals with different levels of GMH and GPH with precision. The subscales can be used to compare demographical and clinical subgroups (no DIF). CONCLUSION: The PROMIS-GH can be used to measure GMH and GPH in individuals with musculoskeletal disorders. Replacement or improvement of item Global10 could be considered.status: Published onlin
The ZZ Ceti Instability Strip for the Most Massive White Dwarf Pulsators
No full textsponsorship: This work is supported in part by NSF under grant AST-2508429 and NASA under grants 80NSSC22K0479, 80NSSC24K0380, and 80NSSC24K0436. M.U. gratefully acknowledges funding from the Research Foundation Flanders (FWO) by means of a junior postdoctoral fellowship (grant agreement No. 1247624N). This work was supported by PIP 112-200801-00940 grant from CONICET, grant G149 from the University of La Plata, PIP-2971 from CONICET (Argentina) and by PICT 2020-03316 from Agencia I+D+i (Argentina). This work was partially supported by the MINECO grant PID2023-148661NB-I00 and by the AGAUR/Generalitat de Catalunya grant SGR-386/2021. VSD, ULTRACAM and HiPERCAM are supported by STFC grant ST/Z000033/1. (National Aeronautics and Space Administration (NASA)|80NSSC22K0479, National Science Foundation (NSF)|AST-2508429)status: Publishe
Where is the eye of the storm of eGFR formulas?
No full textThe estimation of glomerular filtration rate (eGFR) is central to nephrology, yet the growing number of equations often raises concern. However, only a few creatinine-based equations have gained widespread validation: Cockcroft-Gault (CG), Modification of Diet in Renal Disease (MDRD), Chronic Kidney Disease Epidemiology (CKD-EPI), and European Kidney Function Consortium (EKFC). The CG equation, introduced in 1976, was simple but imprecise. In 1999, the MDRD study equation improved GFR estimation and enabled automatic reporting by laboratories by removing weight from the formula. However, its development in patients with CKD led to GFR underestimation in healthy individuals. In 2009, the CKD-EPI equation corrected this, offering improved accuracy and less bias, except in hyperfiltration. In 2021, the CKD-EPI equation was updated to exclude race, mainly for societal reasons, though at the cost of slightly worse performance. In the same year, the EKFC equation was introduced based on European cohorts. It brought two innovations: normalization via the creatinine-to-Q ratio (Q being the median creatinine in healthy individuals), and refined age modeling. These enhancements reduced bias across both the GFR and age ranges and allowed the equation use in both adults and children. At the individual level, all equations remain estimations with similar precision, unless an additional biomarker like cystatin C is used. However, at the population level, EKFC improves bias and accuracy, especially in young adults. Switching from CKD-EPI to EKFC increases CKD prevalence moderately, as shown in European cohorts, but likely reflects better detection. Prognostic performance, particularly for cardiovascular outcomes, is maintained or improved among those reclassified to more severe CKD stages with EKFC. In summary, the EKFC equation represents a scientifically grounded progression in eGFR estimation, an evolution rather than a revolution, offering better performance and flexibility for broader clinical implementation in Europe.status: Published onlin
Rethinking Compensation for High-Voltage Electricity Transmission Line Impacts on Residential Property in Belgium
No full textsponsorship: This study received financial support from Elia Group. (Elia Group)status: Accepte
Quality of Life in Adults With Transposition of the Great Arteries With a Systemic Right or Left Ventricle.
No full textBACKGROUND: Advances in medical and surgical care have improved survival in patients with transposition of the great arteries (TGA), shifting focus toward quality of life (QoL). In this study we evaluate QoL in adults with TGA, including congenitally corrected TGA (ccTGA) and patients with dextro-TGA (d-TGA), by comparing patients with a systemic right ventricle (sRV) and systemic left ventricle (sLV), while identifying mediating factors. METHODS: This cross-sectional study, part of the APPROACH-IS II trial, included 798 adults with TGA from 42 centres worldwide. QoL was assessed using a linear analogue scale (0-100). Regression models identified variables associated with QoL, and mediation analysis assessed the effect of sRV on QoL. RESULTS: Among participants (median age 34 years, 44.9% women), 504 (63.2%) had an sRV (ccTGA or d-TGA with atrial switch) and 294 (36.8%) had an sLV (ccTGA with double-switch operation or Rastelli and d-TGA with arterial switch or Rastelli). Patients with an sRV reported lower QoL (median 75, interquartile range 60-85) compared with those with an sLV (median 80, interquartile range 70-90; P < 0.001). The negative effect of sRV on QoL was mediated for 59% of patients by ventricular dysfunction (B = -2.37, 95% confidence interval -3.38 to -1.21; P < 0.001). Poorer QoL was independently associated with Asian race, employment status (job seeking, unemployed, or disabled), less social support, New York Heart Association functional class ≥ II, ventricular dysfunction, more interventional catheterizations, and depression/anxiety. CONCLUSIONS: TGA patients with an sRV experience a lower QoL than those with an sLV, mediated mainly by ventricular dysfunction. CLINICAL TRIAL REGISTRATION: NCT04902768.sponsorship: Research Foundation Flanders, Heart and Lung Foundation|20190525, National Institute for Health and Care Research, Netherlands Heart Foundationstatus: Publishe
Use of p53 immunohistochemistry can improve diagnostic agreement for differentiated vulvar intraepithelial neoplasia (dVIN): an international reproducibility study
No full textAIMS: Differentiated or HPV-independent vulvar intraepithelial neoplasia (dVIN) can progress rapidly to invasive cancer and accurate pathological diagnosis is essential to facilitate appropriate interventions. Histological similarities of dVIN with non-neoplastic lesions, however, often make the diagnosis less reproducible. We investigated among a diverse group of pathologists whether the diagnostic agreement improves with the use of p53 immunohistochemistry (IHC) interpreted using the pattern-based schema. METHODS AND RESULTS: Fifty haematoxylin-eosin (HE) stained archival slides (30 dVIN and 20 non-dysplastic vulvar lesions) were selected and p53-IHC was performed. Twenty-four board-certified pathologists from eight countries first assessed the HE slides alone, and after a washout period, re-evaluated them alongside the p53-IHC slides. During both rounds, slides were diagnosed as dVIN, favour dVIN, favour no-VIN or no-VIN. p53-IHC was scored as wild-type or mutant (diffuse, basal, cytoplasmic or null). Kappa (κ) statistics and McNemar's test were used for statistical analyses. Overall diagnostic agreement for dVIN saw a significant increase in the Kappa value (κ = 0.6 vs. κ = 0.4, P = 0.002) when HE and p53-IHC slides were assessed together compared with histology assessment alone, although the level of agreement remained moderate. For p53-IHC assessment, overall agreement was substantial (κ = 0.7). Diagnoses changing from no-VIN/favour no-VIN to dVIN correlated significantly with the identification of a p53-mutant pattern (P < 0.001). CONCLUSIONS: Our findings indicate that p53-IHC is a robust ancillary tool that can be reproducibly interpreted by pathologists with varying experience levels and supports the routine use of p53-IHC in cases where dVIN is considered in the differential diagnosis.status: Accepte
Progression of the estimated glomerular filtration rate in asphyxiated neonates undergoing therapeutic hypothermia during the first 10 days of life
No full textBACKGROUND: Serum creatinine (Scr) centile values were recently described in a cohort of 1136 (near)-term neonates that underwent therapeutic hypothermia (TH) because of moderate to severe hypoxic-ischemic encephalopathy. Recent methodological progress enables conversion of these Scr centiles to estimated glomerular filtration rate (eGFR) values. METHODS: Scr centiles in the TH dataset during the first 10 days of life were converted to eGFR values, using the Schwartz formula, with the Smeets k-value (0.31) and fixed body length (50 cm) to generate postnatal reference eGFR values, centiles, and an equation for median eGFRs. These findings were compared to published eGFR data in term controls. RESULTS: A polynomial function was estimated: eGFR mL min ∙ 1.73 m 2 = 9.1667 + 7.1173 - 0.3439 x 2 , ( x = days ) for eGFR in TH neonates. The median eGFR increases 2- to threefold over the first week (day 1: 16.1; day 2: 19.4; day 7: 41.2 mL/min∙1.73 m2), while the polynomial function does not fully reflect the interindividual variability in eGFR values (intra-day variability is also 2- to threefold). Patterns in acute kidney injury (AKI) TH cases differ significantly from non-AKI TH cases. Based on pooling of published eGFR data, this was compared to a function in healthy term neonates: eGFR mL min ∙ 1.73 m 2 = 14.2167 + 6.7644 - 0.3901 x 2 ( x = days ) (day 1: 20; day 2: 26; day 7: 42 mL/min/1.73 m2). CONCLUSIONS: Based on a pooled dataset in TH cases, we converted Scr centiles to eGFR centiles. Based on median values, this resulted in a polynomial function in TH cases, compared to healthy term neonates. This eGFR function enables precision pharmacotherapy for GFR-cleared drugs in this vulnerable population.sponsorship: This research was funded by a senior research grant from the Research Scientific Foundation-Flanders (FWO)-G0D0520N, I-PREDICT: Innovative Physiology-based pharmacokinetic model to pREdict Drug exposure In neonates undergoing Cooling Therapy, and by a KU CELSA research project (Central Europe Leuven Strategic Alliance, CELSA/24/022). The research activities of A. Smits are supported by a Senior Clinical Investigatorship of the Research Foundation-Flanders (FWO) (18E2H24N). (Onderzoeksraad, KU Leuven)status: Accepte
Ground Vibration Attenuation Induced by Gabion and Concrete Mats
No full textsponsorship: The research leading to this paper has been funded in part by the Sao Paulo Research Foundation (Fapesp) through Grant 2022/02753-5 and by the National Council for Scientific and Technological Development (CNPq) through Grant 420314/2023-7. (Research Foundation Flanders)status: Accepte