922017 research outputs found
Sort by
Hybrid Autoregressive Inference for Scalable Multi-hop Explanation Regeneration
No full textRegenerating natural language explanations in the scientific domain has been proposed as a benchmark to evaluate complex multi-hop and explainable inference. In this context, large language models can achieve state-of-the-art performance when employed as cross-encoder architectures and fine-tuned on human-annotated explanations. However, while much attention has been devoted to the quality of the explanations, the problem of performing inference efficiently is largely under-studied. Cross-encoders, in fact, are intrinsically not scalable, possessing limited applicability to real world scenarios that require inference on massive facts banks.To enable complex multi-hop reasoning at scale, this paper focuses on bi-encoder architectures, investigating the problem of scientific explanation regeneration at the intersection of dense and sparse models. Specifically, we present SCAR (for Scalable Autoregressive Inference), a hybrid framework that iteratively combines a Transformer-based bi-encoder with a sparse model of explanatory power, designed to leverage explicit inference patterns in the explanations. Our experiments demonstrate that the hybrid framework significantly outperforms previous sparse models, achieving performance comparable with that of state-of-the-art cross-encoders while being ≈ 50 times faster and scalable to corpora of millions of facts. Further analyses on semantic drift and multi-hop question answering reveal that the proposed hybridisation boosts the quality of the most challenging explanations, contributing to improved performance on downstream inference tasks
Numerical Stability of Algorithms at Extreme Scale and Low Precisions*
No full textThe largest dense linear systems that are being solved today are of order 𝑛 = 107. Single precision arithmetic, which has a unit roundoff 𝑢 ≈ 10-8, is widely used in scientific computing, and half precision arithmetic, with 𝑢 ≈ 10−4, is increasingly being exploited as it becomes more readily available in hardware. Standard rounding error bounds for numerical linear algebra algorithms are proportional to 𝑝(𝑛)𝑢, with 𝑝 growing at least linearly with 𝑛. Therefore we are at the stage where these rounding error bounds are not able to guarantee any accuracy or stability in the computed results for some extreme-scale or low-accuracy computations. We explain how rounding error bounds with much smaller constants can be obtained. Blocked algorithms, which break the data into blocks of size 𝑏, lead to a reduction in the error constants by a factor 𝑏 or more. Two architectural features also reduce the error constants: extended precision registers and fused multiply–add operations, either at the scalar level or in mixed precision block form. We also discuss a new probabilistic approach to rounding error analysis that provides error constants that are the square roots of those of the worst-case bounds. Combining these different considerations provides new understanding of the numerical stability of extreme scale and low precision computations in numerical linear algebra.<br/
A genotype-first approach in carriers of CDH1 rare germline variants: a European Reference Network on Genetic Tumour Risk Syndromes (ERN-GENTURIS) multicentre study
No full textBackgroundTruncating CDH1 pathogenic/likely pathogenic variants (PV/LPV) cause Hereditary Diffuse Gastric Cancer (HDGC), predisposing to diffuse gastric (DGC) and to lobular breast cancer (LBC). Rare CDH1 missense variants often classify as variants of unknown significance (VUS). We conducted a genotype-phenotype analysis in CDH1 rare-variant carrier families, comparing the cancer landscape of PV/LPV and missense-VUS, assessing frequency of LBC-centered families among PV/LPV carrier families, and testing performance of LBC-expanded criteria for CDH1 testing.MethodsThis genotype-first study used retrospective diagnostic and clinical data from 854 carriers of 398 CDH1 rare-variants and 1021 relatives, irrespective of HDGC clinical criteria, from ten European Reference Network on Tumour Risk Syndromes (ERN-GENTURIS) countries. Variants were classified for molecular type and clinical actionability with ACMG/AMP CDH1 guidelines v2. Families were categorized for compliance with 2015 and 2020 HDGC clinical criteria. From 1971 phenotypes (probands and relatives aged 1-93 years), 460 had gastric and breast cancer histology available. Genotype-phenotype associations were analysed by student’s t-test, Kruskal Wallis, chi-square and multivariable logistic regression models. Performance of HDGC clinical criteria sets were tested with equivalence test, Youden index, and the Area Under the Curve (AUC) of the Receiver operating characteristic (ROC) curves were compared with Z-test.FindingsCDH1-truncating PV/LPV occurred in 176/854 (21%) and missense-VUS in 169/854 (20%) families. Multivariable logistic regression comparing phenotypes presented by PV/LPVs and Missense-VUS carrier families showed that LBC has the greatest positive-association with presence of PV/LPVs (OR=12·4, 95% CI [2·66-57·7], p=0·0014), followed by DGC (8·00, 95%; [2·18-29·4], p=0·0017) and gastric cancer (GC) (7·81, 95%, [2·03-30·0], p=0·0027). 136/176 (77%) PV/LPV carrying families fulfilled HDGC 2015 criteria and 40/176 (23%) did not. 11/40 fulfilled 2020 HDGC criteria, and 18/40 presented LBC-only or LBC+GC, but no criteria. No specific CDH1 variant seemed predisposing specifically to LBC, although 7% (12/176) of all PV/LPV-carrier families were LBC-only. Addition of three new LBC-centered criteria improved testing sensitivity while retaining high specificity. Probability of finding CDH1 PV/LPV in patients fulfilling LBC-expanded criteria compared to former criteria increased significantly (AUC: 0·92 vs. 0·88; z-score=3·536; p<0·0001).InterpretationCDH1 PV/LPV are positively associated with HDGC-related phenotypes (LBC, DGC and GC), and no evidence for a positive association was found for CDH1 missense-VUS. CDH1 PV/LPV occurred often in LBC-enriched families lacking 2020 HDGC criteria, supporting the expansion of LBC-centered criteria
IDENTIFYING VULNERABLE LINES CONSIDERING UNCERTAIN HEAT ELECTRIFICATION IN INTEGRATED GAS AND ELECTRICITY NETWORKS
No full textThe installation of heat pumps, powered from clean electricity sources, will play a key role in heat decarbonisation. However, the resulting surge in electricity demand will put the existing transmission lines under enormous strain, increasing the risk of line overloading, potential failure and even loss of load events. This paper presents a probabilistic method for assessing line overload in integrated gas and electricity networks with the consideration of heat electrification. The established model fully considers the uncertainties associated with intermittent generation and fluctuating loads. The effectiveness of the proposed method is verified via its application in an integrated 9-bus electricity and 8-node gas network. Numerical results show that the overload probability of power lines and gas pipelines increases as a result of heat electrification. The overload probability further rises when the correlations of renewable generation and loads are considered, confirming the significance of modelling correlation for an accurate estimation of power and gas flows. The overload probability of some power lines and gas pipelines soar as they become disproportionately loaded, when considering an uneven distribution of heat loads (i.e., clustering of lowcarbon technologies) in the electricity network.<br/
Nurses’ adoption and use of digital technology during the COVID-19 pandemic
No full textAim: to explore nurses’ adoption and use of digital technologies during the COVID-19 pandemicMethods: online survey to obtain nurses’ feedback and ratings of technologiesimplemented and used to support patient care during the COVID-19 pandemic. Survey questions comprised fixed response and free-text questions and a rating of system usability (quantitative and qualitative data). Participants were recruited via nursing networks and social media.Results: 55 respondents provided details on 85 separate technologies. The majority of technologies were used for patient monitoring/data sharing (n=39; 46%), online communication (n=22; 26%) virtual appointments (n=15; 18%). Other technologies included electronic patient records (n=5, 6%), e-Prescribing (n=3; 3%), and for PPE (n=1, 1%). Usability of systems varied significantly across types of technology. Barriers to successful use included accessibility to effective infrastructure.Conclusion: there was a range and breadth of digital technologies adopted and used by nurses during the pandemic. There are ongoing issues with the availability of digital infrastructure to enable effective digital working, and poor usability of some systems that have been implemente
Meta-Analysis of the Effect of Colchicine on C-Reactive Protein in Patients with Acute and Chronic Coronary Syndromes
No full textObjective: The anti-inflammatory drug Colchicine has recently shown benefits in the prevention of major adverse cardiovascular events (MACE) in patients with acute coronary syndrome (ACS) and chronic coronary syndromes (CCS). This meta-analysis focuses on understanding Colchicine's effects on the high sensitivity C-reactive protein (hs-CRP) to provide mechanistic insight to explain its clinical event reduction.Methods: A computerized search of MEDLINE was conducted to retrieve journal articles with studies performed on humans from January 1, 2005, to January 1, 2022, using keywords: “Colchicine AND Coronary”, “Colchicine AND CRP”, and “Colchicine AND Coronary Artery Disease”. Studies were included if they measured hs-CRP changes from baseline, and Colchicine or placebo were given to patients with ACS or CCS. Results: Thirteen studies with a biomarker subgroup population of 1636 patients were included in the hs-CRP meta-analysis. Of those 13 studies, 8 studies with a total population of 6016 reported clinical events defined as myocardial infarction (MI), stroke, cardiovascular death, periprocedural MI, repeat angina after PCI, and repeat revascularization. . A multivariate analysis revealed a weak negative correlation of -0.1056 (p= 0.805) between change in CRP and clinical events. Overall, colchicine treatment resulted in greater reduction in hs-CRP levels compared with placebo (standardized mean difference [MD-1.59 [95% CI: -2.40, -0.79], p=0.0001) and clinical events (Odds Ratio: 0.78 [0.64, 0.95], p=0.01).Conclusion: Colchicine therapy is associated with a reduction in hs-CRP and clinical events in patients with ACS and CCS. This finding supports colchicine’s anti-inflammatory efficacy via CRP reduction to explain its clinical benefit.<br/
Formation-containment tracking and scaling for multiple quadcopters with an application to choke-point navigation
No full textThis paper investigates the cooperative control problem of choke-point navigation for multiple quadcopters when only their subgroup is equipped with obstacle detecting sensors. We define a quadcopter as a leader if it is equipped with an obstacle detecting sensor; otherwise, it is a follower. In addition, we introduce a virtual leader agent to create the group motion. First, we apply the leader-follower approach and propose a formation-containment tracking controller for multiple quadcopters to track the time-varying velocity of the virtual leader agent. At the same time, the leader quadcopters form the prescribed formation while the follower quadcopters converge inside a safe region, which is the convex hull spanned by those leaders. Then, we introduce a scaling vector into the displacement-based formation constraints. When the leader quadcopters identify the choke-point via their obstacle detecting sensors, they update the scaling variable to adjust the size of the formation (i.e. the safe region) and guide all quadcopters to safely pass through the choke-point. The proposed cooperative controllers are distributed because each quadcopter’s control command only relies on the information states from its neighbours. Finally, two autonomous flight experiments, including formation-containment tracking and choke-point navigation, are provided to validate the effectiveness of the proposed cooperative control laws
Young children rely on gossip when jointly reasoning about whom to believe
No full textPeople rely on reputational information communicated via gossip when deciding about with whom to cooperate, whom to believe, and whom to trust. In two studies, we investigated whether 5 and 7-yearold children trust in gossip when determining a course of action. In Study 1, 5 and 7-year-old Germanspeaking peer dyads (N = 64 dyads, 32 female dyads) were presented with a collaborative problem-solving task (e.g., deciding together what a creature eats). Each child individually received conflicting information about the solution from a different informant (e.g., one proposed rocks; the other proposed sand). Each child additionally heard gossip about the informant’s reputation: one informant had a good reputation; the other had a bad reputation. In the experimental condition, the reputation was relevant to the task (honesty), whereas it was irrelevant in the control condition (tidiness). Seven-year-old dyads, and 5-year-old dyads to a lesser extent, settled on the items suggested by the informant with good reputation in the experimental but not in the control condition. Only 7-year-old children explicitly referred to the information conveyed via gossip, engaging in metatalk about the reputations of the informants. In Study 2, we replicated these findings in a more controlled experiment in which 5 and 7-year-old American English-speaking children (N = 48, 27 girls) tried to convince an adult partner who proposed the item suggested by the informant with bad reputation. Thus, starting around age 5, and more reliably at age 7, children selectively rely on gossip in identifying trustworthy individuals in their collaborative reasoning with partners.</p
Review of 'Harnessing Public Research for Innovation in the 21st Century: An International Assessment of Knowledge Transfer Policies. Edited by: Anthony Arundel, Suma Athreye and Sacha Wunsch-Vincent Cambridge and the World Intellectual Property Organization. Cambridge, United Kingdom. 2021. 502 pp. ISBN 9781108842792'
No full textControlling desolvation through polymer-assisted grinding
No full textWe demonstrate the ability to controllably desolvate a crystal-solvate system in step-wise fashion through polymer-assisted grinding by varying the type and proportion of polymer agent used. A plausible mechanistic explanation is proposed based on a combination of experimental evidence and computational analysis. Specifically, Raman spectroscopy, total scattering pair distribution function analysis and computed reaction energies suggest that the desolvation process is associated with preferred interactions between the solvent molecules and specific polymers. This approach could potentially be extended to any type of material, including heat-sensitive materials, where classical desolvation by thermal processes is not possible, and provides a new potential technique for formulation processing