143174 research outputs found
Sort by
Learning and enforcing context-sensitive control for LLMs
No full textControlling the output of Large Language Models (LLMs) through context-sensitive constraints has emerged as a promising approach to overcome the limitations of Context-Free Grammars (CFGs) in guaranteeing generation validity. However, such constraints typically require manual specification—a significant barrier demanding specialized expertise. We introduce a framework that automatically learns context-sensitive constraints from LLM interactions through a two-phase process: syntactic exploration to gather diverse outputs for constraint learning, followed by constraint exploitation to enforce these learned rules during generation. Experiments demonstrate that our method enables even small LLMs (1B parameters) to learn and generate with perfect constraint adherence, outperforming larger counterparts and state-of-the-art reasoning models. This work represents the first integration of context-sensitive grammar learning with LLM generation, eliminating manual specification while maintaining generation validity
Advanced Alexandrite lasers and technology
No full textAlexandrite has attracted great attention as a laser gain medium owing to its broad wavelength tunability in near infra-red (NIR) range and excellent properties. This work is aimed to develop new technologies of Alexandrite lasers to fulfill further needs of applications. }
Red-diode-pumped continuous-wave (CW) Alexandrite lasers are presented with different cavity designs and pumping arrangements. Laser outputs with high power, high spatial quality and tunable wavelength are demonstrated.
Deep study for pumping-induced lensing effect of Alexandrite is performed. A full theory of lensing strength and transient response is developed and shows excellent agreement with the experimental results under different conditions, for example, pumping lenses, crystal types, and temperature. This transient analysis provides a new method to separate the two lensing components, thermal and population lensing, in Alexandrite lasers.
The potential of blue diodes to be applied in Alexandrite lasers as pumping sources is explored. Record power from a blue-diode-pumped Alexandrite laser is obtained, and the analysis about the pumping-induced lensing effect is also performed. The difference in lensing performance provides a comparison with red-pumping for further understanding in the lensing mechanism.
Nonlinear optical frequency conversion of diode-pumped Alexandrite lasers has been demonstrated using both external-cavity and intra-cavity generation. Pulsed ultra-violet (UV) second harmonic generation (SHG) with beam quality improvement is conducted with the walk-off compensation technique from a diode-pumped Alexandrite laser. World-first deep-UV laser pulses from third harmonic generation (THG) are subsequently demonstrated from a diode-pumped Alexandrite laser with 30 μJ pulse energy and 19.1% total conversion efficiency. CW UV output from diode-pumped Alexandrite through intra-cavity SHG is also shown in this work. A record CW UV laser with maximum output power of 5.05 W and wide wavelength tunable range from 364 nm to 402 nm is achieved.Open Acces
Yoga outcomes get assessed in cystic fibrosis (YOGA-CF): protocol of a multi-centre interventional randomised controlled clinical trial – investigating effects of a 12-week yoga intervention for adults with cystic fibrosis
No full textIntroduction
Yoga is an emerging exercise choice for people with cystic fibrosis (CF), but evidence of its effect in this population is scarce, with a recent systematic review advocating for further research. Yoga Outcomes Get Assessed in CF (YOGA-CF) is a real-world multi-centre randomised controlled trial (RCT) investigating a bespoke CF-specific online 12-week yoga intervention, versus usual care, to determine effectiveness for adults with CF.
Methods & Analysis
A multi-centre RCT of adults with CF across the United Kingdom (UK). Participants are randomised to usual care or a 12-week online bespoke yoga programme with an expectation of two classes completed weekly. Assessments of lung function, one-minute sit-to-stand, the Cystic Fibrosis Questionnaire-Revised (CFQ-R) and other trial questionnaires are completed pre- and post-intervention (0 and 12 weeks), and after 12-weeks of follow-up (week 24). The primary outcome is difference in respiratory-related quality of life measured using the CFQ-R before and after yoga/control. Sample size was calculated based upon detecting a minimally clinically important difference (MCID) of 4 for the CFQ-R respiratory domain, with power of 80% and 5% significance level (total target, n=314).
Ethics & Dissemination
Ethics approval gained from the South-Yorkshire and Humber Research Ethics Committee (REC) (reference: 23/YH/0270, project ID 303898). Dissemination to involve direct participant feedback and lay webinar, scientific conference presentation and publication in a peer-reviewed journal.
Trial registration details
www.clinicaltrials.gov.uk reference: NCT0612046
Evaluating spoken language as a biomarker for automated screening of cognitive impairment
No full textBackground
Timely and accurate assessment of cognitive impairment remains a major unmet need. Speech biomarkers offer a scalable, non-invasive, cost-effective solution for automated screening. However, the clinical utility of machine learning (ML) remains limited by interpretability and generalisability to real-world speech datasets.
Methods
We evaluate explainable ML for screening of Alzheimer’s disease and related dementias (ADRD) and severity prediction using benchmark DementiaBank speech (N = 291, 64% female, 69.8 ± 8.6 years). We validate generalisability on pilot data collected in-residence (N = 22, 59% female, 76.2 ± 8.0 years). To enhance clinical utility, we stratify risk for actionable triage and assess linguistic feature importance.
Results
We show that a Random Forest trained on linguistic features for ADRD detection achieves a mean sensitivity of 69.4% (95% confidence interval (CI) = 66.4–72.5) and specificity of 83.3% (78.0–88.7). On pilot data, this model yields a mean sensitivity of 70.0% (58.0–82.0) and specificity of 52.5% (39.3–65.7). For prediction of Mini-Mental State Examination (MMSE) scores, a Random Forest Regressor achieves a mean absolute MMSE error of 3.7 (3.7–3.8), with comparable performance of 3.3 (3.1–3.5) on pilot data. Risk stratification improves specificity by 13% on the test set, offering a pathway for clinical triage. Linguistic features associated with ADRD include increased use of pronouns and adverbs, greater disfluency, reduced analytical thinking, lower lexical diversity, and fewer words that reflect a psychological state of completion.
Conclusions
Our predictive modelling shows promise for integration with conversational technology at home to monitor cognitive health and triage higher-risk individuals, enabling early screening and intervention
Increasing heavy rainfall and extreme flood heights in a warming climate threaten densely populated regions across Sri Lanka and the Malacca Strait
No full textCite as Kew, S. et al., (2025): Increasing heavy rainfall and extreme flood heights in a warming climate threaten densely populated regions across Sri Lanka and the Malacca Strait (WWA scientific report No. 78) World Weather Attributio
Temporal link prediction in the wild
No full textIn a data-driven world, knowledge graphs become essential in organising and navigating the web of explicit and implicit relationships within data. Knowledge graphs store factual information on real-world objects and concepts and the relationship between entities, offering us a holistic view and understanding of the domain of interest. Despite their potential to revolutionise information management and enhance various industrial applications, traditional knowledge graphs often lack the temporal information necessary to capture the dynamic nature of real-world data. Additionally, many existing methodologies for encoding knowledge graphs are developed to handle traditional static knowledge graphs. As such, they are ineffective in handling an ever-evolving temporal knowledge graph with new emerging entities and relations over time. This reduces the applicability of knowledge graphs in real-world scenarios.
This limitation highlights the critical need for methodologies to effectively encode temporal knowledge graphs, enabling reliable integration of evolving knowledge graphs into industrial applications. Addressing this important challenge serves as the overarching goal of this PhD thesis.
This thesis aims to develop a novel method that can accurately encode unseen entities and relations in ever-evolving temporal knowledge graphs such that knowledge graphs can reliably infuse up-to-date information into industrial applications
Cross-herpesvirus immunity of the cytomegalovirus gB/MF59 vaccine response
No full textVaccination against human cytomegalovirus (HCMV) to protect transplant recipients and prevent congenital infection remains highest priority. Follow-up analyses of a vaccine directed against the fusion protein glycoprotein B (gB/MF59) identified a vaccine-specific response (AD-6) that correlated with protection. Subsequently, it was demonstrated that AD-6 antibodies are anti-viral by preventing cell-associated spread. Here we now demonstrate AD-6 antibodies limit HCMV reactivation – an event critical for pathogenesis via hematogenic spread in vivo. To better understand the AD-6 immunogen, we use structural homology to identify putative AD-6 regions in related herpesviruses and show, despite limited sequence similarity, they share key physico-chemical properties. Of note was that AD-6 mapped to a region under high molecular frustration within gB – arguing AD-6 antibodies inhibit gB function by targeting activity dependent on conserved conformational changes. Consistent with structural conformation being crucial, we observe that both rabbit and human HCMV AD-6 antibodies recognise other herpesvirus AD-6s and that AD-6 antibodies are potently antiviral against HSV-1. Thus, a combinatorial in silico, biochemical and immunological approach reveals conformational epitopes within AD-6 are critical components of the gB/MF59 vaccine, represent crucial conserved elements of AD-6 in gB structure and function which makes it an attractive target of multiple herpesviruses
Enhancing the quality of systematic reviews and meta-analyses
No full textSystematic reviews and meta-analyses are often considered the highest level in evidence hierarchies, and therefore are often drawn upon when considering changes in policy. Despite journals implementing measures aiming to enhance the quality of systematic reviews they publish, the authorship raise concerns about the quality of existing and ongoing systematic reviews, particularly relating to transparency and bias minimisation. Building on the current guidelines, standards and tools, we suggest a ‘meta checklist’ which aims to maximise methodologically sound, unbiased and reproducible reviews of the best scientific quality while considering feasibility throughout the process