179566 research outputs found
Sort by
Mixtures of in-context learners
Get PDFIn-context learning (ICL) adapts LLMs by providing demonstrations without fine-tuning the model parameters; however, it is very sensitive to the choice of in-context demonstrations, and processing many demonstrations can be computationally demanding. We propose Mixtures of In-Context Learners (MoICL), a novel approach that uses subsets of demonstrations to train a set of experts via ICL and learns a weighting function to merge their output distributions via gradient-based optimisation. In our experiments, we show performance improvements on 5 out of 7 classification datasets compared to a set of strong baselines (e.g., up to +13% compared to ICL and LENS). Moreover, we improve the Pareto frontier of ICL by reducing the inference time needed to achieve the same performance with fewer demonstrations. Finally, MoICL is more robust to out-of-domain (up to +11%), imbalanced (up to +49%) and perturbed demonstrations (up to +38%). Our code is available at https://github.com/HongGiwon/Mixture-of-In-context-Learners
Contesting authoritarianism:The role of youth-led movement in shaping futures in Serbia
Get PDFThis article explores the youth-led protest movement in Serbia as a case of democratic resistance and future-making in the face of authoritarianism. Sparked by a public tragedy – the collapse of a railway station canopy that killed 16 people – the student-led uprising has developed into a powerful civic mobilisation calling for institutional accountability, justice, and systemic reform. Drawing on ethnographic observations, public statements, and student-produced materials, the article situates the movement within both local historical legacies and global democratic struggles.The protests are analysed not only as a political response to state corruption and the erosion of public institutions, but as a cultural and generational effort to reclaim democratic values, the public commons, and national identity from autocratic capture. Students have reappropriated national symbols and historical narratives to support an inclusive vision for Serbia’s future – one grounded in rule of law, participatory governance, and social solidarity. Operating through decentralised, non-violent, and self-managed structures, the movement draws on traditions of samoupravljanje (self-management) and revitalises the practice of zbor (grassroots assembly) to model alternative democratic forms.In addition to documenting successes and setbacks, the article considers four possible scenarios for Serbia’s political trajectory, ranging from democratic reform to authoritarian intensification. Ultimately, it argues that the movement has already initiated a form of irreversible change by restoring public imagination, breaking cycles of fear and silence, and affirming the liberatory potential of collective action. It thus contributes to broader debates on youth agency, resistance under autocracy, and the future of democracy in the post-Yugoslav region and beyond
About time: synchrony in onset of parental care influences cooperation and offspring performance in a burying beetle
Get PDFIn providing biparental care, each parent must coordinate its relative contribution to care with that of its partner. Previous studies have shown that each parent adjusts its contribution to care based on its partner's contribution and state. Here, the effect of timing caused by synchrony in the onset of parental care was investigated. Synchrony likely matters if the parent arriving earlier contributes more towards prehatching care, thereby altering the parents' relative contribution to posthatching care if the parent arriving later compensates for their partner's greater contribution to prehatching care. To test this idea, synchrony in the onset of care in the burying beetle Nicrophorus vespilloides was manipulated using a two-by-two factorial design with the following treatments: Both parents arrived synchronously at a fresh carcass; the parents arrived asynchronously such that the female arrived at a fresh carcass and the male arrived 24 h later; the parents arrived asynchronously such that the male arrived at a fresh carcass and the female arrived 24 h later, or both parents arrived synchronously at a 24-h-old carcass. We found that males provided more indirect care when the parents arrived asynchronously at the carcass compared with when they arrived synchronously. Despite this, larval survival to dispersal was lower in broods cared for by parents that arrived asynchronously potentially as a result of a reduction in the efficiency or quality of care being provided. Results indicate that asynchrony in the onset of parental care alters the parents' relative contribution to posthatching care possibly by disrupting synchrony in the reproductive physiology of the female and male. Thus, linking synchrony in reproductive behaviour with synchrony in reproductive physiology is encouraged in future work
Financial reporting comparability in US firms issuing debt in the US primary market
Get PDFWe propose a method of measuring the comparability of reported accounting numbers from the perspective of creditors. Our measure reflects the relationship between market variables related to default probabilities and key accounting numbers of interest to lenders. We demonstrate that our measure captures the mapping between the market variables and the accounting system and that the values are consistent with expected differences in comparability in a range of tests and settings. We employ this measure to determine the impact of comparability in the US primary bond market and find that it has a negative and significant relationship with the spread of credit rating assessments of new bond issues. We also show that new bond issues of firms with superior comparability have better credit ratings and reduced bond yields, ceteris paribus. We find that the benefits of our lender-focused, balance-sheet-based comparability measure to the public bond market are robust to controlling for the earnings-based financial statement comparability measure of De Franco et al. (2011), and independent of economic similarities between firm pairs. Our findings are commensurate with comparability reducing the information uncertainty surrounding credit risk assessments derived from a firm's financial information
Heterometallic Li/Zn, Li/Al and Li/In catalysts for rac-lactide ring-opening polymerisation: “ate” or “non-ate” pathways?
Get PDFThe ring-opening polymerisation (ROP) of lactide (LA) is an attractive route to produce aliphatic polyesters, with bimetallic catalysts displaying some of the highest catalyst activities to date. While a range of heterometallic catalysts have been reported to outperform their homometallic analogues, the origins of cooperativity are not always well understood. Previous studies indicate that the reaction pathways may differ for different metal heterocombinations, especially when an alkali metal is combined with zinc or aluminium. Here, a series of homo- and hetero-metallic complexes combining Li with Al, Zn or In, supported by an asymmetric methyl-ester substituted salen ligand (H2L), have been synthesised and characterised by single-crystal X-ray diffraction, to probe potential differences. The heterobimetallic LLiZnCl, LLiAlCl2, and LLiInCl2 complexes were all active for rac-LA ROP in the presence of an epoxide initiator, with LLiInCl2 offering the most efficient polymerisation while homobimetallic LLi2 was inactive. Investigations into the roles of the different metals through X-ray diffraction and DFT structural studies suggest that oxophilicity, Lewis acidity, and electronegativity difference between the two metals all play a role, with the high oxophilicity and Lewis acidity of Al overriding the “ate” pathway
Detection of 85 new active subglacial lakes in Antarctica from a decade of CryoSat-2 data
Get PDFSubglacial lake activity influences ice sheet flow, grounding line discharge and ice shelf basal melting. Although 146 active subglacial lakes have been detected in Antarctica via ice surface elevation change associated with their activity, only 36 fill-drain cycles have been observed worldwide, and knowledge of these mechanisms is limited. Here, we use a decade of CryoSat-2 radar altimetry to detect 85 active subglacial lakes in Antarctica, documenting 37 and 34 complete draining and filling events respectively. We delineate time-varying boundaries of subglacial lake activity and investigate their variability over time. Our observations increase the number of known active subglacial lakes in Antarctica by 58%, with six of these located within 8 km of the grounding zone. We observe five subglacial lake networks, with concurrent upstream drainage and downstream filling, and 25 clusters of lakes, improving our knowledge of interconnected subglacial hydrological pathways
The role of general practitioner empathy on patient-reported outcomes in telephone consultations:a cross-sectional study with validation of the Consultation and Relational Empathy (CARE) Measure for measuring empathy in telephone consultations
Get PDFBackgroundThe Consultation and Relational Empathy (CARE) Measure is a validated measure of GP empathy in face-to-face consultations (FTFC) and higher CARE scores predict better patient outcomes. However, the validity of CARE and its importance in telephone consultations (TC) is unknown.Aims1.To determine the validity and reliability of CARE in TCs2.To determine the associations between CARE and outcomes in TC compared with FTFC.Design and settingCross-sectional survey of 1023 patients who had a TC (n=369) or FTFC (n=654) with a GP in Scotland within the preceding 30 days.MethodsValidity and reliability testing of CARE using standard methods. Associations between low, medium, and high CARE scores and three patient outcomes (enablement, symptom change, and satisfaction) were determined using multilevel binary logistic regression. ResultsCARE was valid and reliable. Compared with the low CARE group, high CARE was positively associated with enablement (adj.OR 6.4; 95%CI 3.48 to 11.84), symptom improvement (adj.OR 5.7; 95%CI 2.70 to 11.93) and satisfaction (adj.OR 20.1; 95%CI 8.93 to 45.35) in TC. Findings were similar in FTFC (enablement adj.OR 6.21; 95%CI 3.90 to 9.89, symptom improvement adj.OR 4.26; 95% CI 2.53 to 7.16, and satisfaction adj.OR 36.29; 96%CI 17.85 to 73.78) , and were not influenced by patient or consultations characteristics.ConclusionThe CARE Measure is valid and reliable in TC and predicts better patient outcomes in TC, similar to FTFC. Given the common use of TC in primary care, strategies to support empathic communication are essential irrespective of consultation modality.<br/
The descriptive epidemiology of age at colorectal cancer diagnosis in Africa: a systematic review and meta-analysis
Get PDFBackground: The optimal age for introducing colorectal cancer (CRC) screening remains controversial especially in low-and middle- income countries, where evidence is limited, and the CRC incidence appears to be gradually increasing. In this systematic review we aimed to summarise the age profile of CRC cases at diagnosis on the African continent to inform the most appropriate age for CRC screening initiation. Methods: Four electronic databases: MEDLINE, Embase, Scopus and Web of Science were searched using relevant subject headings and keywords from 1 January 2012 to 22 September 2023. The third volume of the African Cancer Registry Network (AFCRN) report was also included. Studies of African populations reporting the age profiles of CRC patient populations were eligible for inclusion. Random effects meta-analysis was used to pool mean age at CRC diagnosis, and the proportions of early-onset CRC (EOCRC); <40 and <50 years old. Age specific incidence rates and proportions of EOCRC were calculated from the AFCRN report. The study was registered with PROSPERO, CRD42020187538.Findings: Eighty-seven (87) studies reporting on 72,204 CRC cases from 24 African countries were included. The pooled mean age at CRC diagnosis was 53·5 years, with significant regional variations observed. The pooled proportion of EOCRC cases diagnosed <40 years old was 21% (95% confidence interval (CI): 18%-25%; n=44 studies) and 38% for those <50 years old (95% CI 35%-41%; n=41 studies). Estimated proportions of EOCRC from published studies and the AFCRN report were comparable. Interpretation:Whereas the incidence of CRC is still low in Africa, there is younger age at diagnosis and an overall high proportion of EOCRC. Future CRC screening programs on the continent may need to consider a lower CRC screening age compared to that used in Western countries where EOCRC appears to be less common.<br/
Ostwald ripening in evaporating respiratory breath figures
Get PDFUnderstanding collective evaporation of micron-sized breath droplets can enhance modelling of clouds and inform strategies to limit virus transmission. In this work we analyse thousands of micron-sized droplets within human and pure water breath figures, identifying two regimes. Firstly, during Coarsening whilst the entire array is evaporating, larger droplets grow at the expense of smaller ones, driven by differences in equilibrium vapour concentration as low as 0.02%, arising from varying interfacial curvature. In this regime the temporal distribution of droplet radii behaves according to classic Ostwald ripening theory, a mechanism which is usually considered only for sub-micron droplets. We then use a mean field model to connect droplet evaporation theory with Ostwald ripening, and predict individual droplet dynamics, showing how the accuracy depends on droplet coverage fraction. Finally, we show how the coarsening regime is terminated by an Edge regime, in which all remaining droplets evaporate rapidly
Sensor Placement Optimisation for FastBlade Reaction Frame Health Monitoring Using Machine Learning Approaches
Get PDFFastBlade, a leading centre for testing composite materials and structures, specialises in conducting experimental fatigue testing on tidal turbine blades and composite structures under various load conditions. These experimental fatigue campaigns can sometimes run for months and hence generate vast datasets from numerous sensors used in testing. Traditional grid-based sensor deployment often results in redundant data, complicating data processing and analysis. Hence, optimising sensor placement is crucial for efficient data collection and cost-effectiveness. This paper addresses the challenge of optimising sensor placement on the FastBlade reaction frame, essential for full-scale tidal turbine testing, by integrating finite element (FE) simulation data with machine learning (ML) algorithms. Our approach leverages validated finite element (FE) models to identify critical stress regions and simulate mechanical responses, generating high-resolution strain datasets. ML algorithms then determine optimal sensor locations, focusing on areas with high strain gradients. The optimised sensor network is validated using calibrated FE models, which provides continuous deformation data across the frame. Results show a significant reduction in the quantity of stain gauges or sensors while maintaining accurate monitoring of critical areas, demonstrating the effectiveness and cost-efficiency of our innovative sensor placement strategy
Access Repository Dashboard
Do you manage Edinburgh Research Explorer? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!