179566 research outputs found
Sort by
Geometry-aware training of factorized layers in tensor Tucker format
Get PDFReducing parameter redundancies in neural network architectures is crucial for achieving feasible computational and memory requirements during training and inference phases. Given its easy implementation and flexibility, one promising approach is layer factorization, which reshapes weight tensors into a matrix format and parameterizes them as the product of two small rank matrices. However, this approach typically requires an initial full-model warm-up phase, prior knowledge of a feasible rank, and it is sensitive to parameter initialization. In this work, we introduce a novel approach to train the factors of a Tucker decomposition of the weight tensors. Our training proposal proves to be optimal in locally approximating the original unfactorized dynamics independently of the initialization. Furthermore, the rank of each mode is dynamically updated during training. We provide a theoretical analysis of the algorithm, showing convergence, approximation and local descent guarantees. The method's performance is further illustrated through a variety of experiments, showing remarkable training compression rates and comparable or even better performance than the full baseline and alternative layer factorization strategies
Backward anaphora in L2 acquisition of English, Mandarin and Japanese
Get PDFThis study reports findings from two experiments on backward anaphora by second language (L2) learners of English, Mandarin, and Japanese, examining the effects of first language (L1) systems and whether a pronoun resolution pattern requiring syntax–discourse integration can be successfully acquired. The findings from Experiment 1 comparing English and Mandarin demonstrate that the availability of positive evidence for Mandarin-speaking learners of L2 English allows them to achieve a native-like pattern, whereas it is difficult even for advanced English-speaking learners of L2 Mandarin to acquire the syntax–discourse integration pattern that Mandarin requires. Experiment 2 compared two languages that show the syntax–discourse integration condition for backward anaphora (Japanese and Mandarin) to further investigate (1) whether L2 speakers were influenced by their L1 patterns, or (2) whether a pattern as in English, i.e. not requiring syntax–discourse integration, would emerge regardless of the L1s. The findings from Experiment 2 reveal an effect of L1 system because the Mandarin-speaking learners of L2 Japanese, the Japanese-speaking learners of L2 Mandarin, and the two groups of native speakers display similar patterns of interpretation
The W. Ormiston Roy lecture 2024:Robert Burns and patriot rhetoric
Get PDFThis is the text of the W. Ormiston Roy Fellowship Lecture delivered at the University of South Carolina on 23 May 2024 (explanatory footnotes have been added). It addresses Burns’s use, in his verse, of the rhetoric of ‘Patriot’ opposition to the British fiscal-military state, and begins by describing this rhetoric and sketching its origins. As a model for understanding Burns’s appropriation of this discourse the lecture then shows how this is done in two poems by Anna Barbauld. It argues that the value of this rhetoric for both Barbauld and Burns lay in its offering them a place from which they could address a national (British) public, despite their exclusion, by their gender and their social rank respectively, from the political class. The lecture concludes by considering two poems, unpublished in Burns’s lifetime, in which he mourns the loss of this possibility in the aftermath of the French Revolution (‘An irregular Ode for General Washington’s birth-day’), and notes the moral vanity in the enfranchised classes that it permits (‘On Glenriddel’s Fox breaking his chain’)
Flux Sampling Suggests Metabolic Signatures of High Antibody-Producing CHO Cells
Get PDFChinese hamster ovary (CHO) cells remain the industry standard for producing numerous therapeutic proteins, particularly monoclonal antibodies (mAbs). However, achieving higher recombinant protein titers remains an ongoing challenge and a fundamental understanding of the cellular mechanism driving improved bioprocess performance remains elusive. To directly address these challenges and achieve substantial improvements, a more in-depth understanding of cellular function within a bioprocess environment may be required. Over the past decade, significant advancements have been made in the building of genome-scale metabolic models (GEMs) for CHO cells, bridging the gap between high information content 'omics data and the ability to perform in silico phenotypic predictions. Here, time-course transcriptomics has been employed to constrain culture phase-specific GEMs, representing the early exponential, late exponential, and stationary/death phases of CHO cell fed-batch bioreactor culture. Temporal bioprocess data, including metabolite uptake and secretion rates, as well as growth and productivity, has been used to validate flux sampling results. Additionally, high mAb-producing solutions have been identified and the metabolic signatures associated with improved mAb production have been hypothesized. Finally, constraint-based modeling has been utilized to infer specific amino acids, cysteine, histidine, leucine, isoleucine, asparagine, and serine, which could drive increased mAb production and guide optimal media and feed formulations.</p
Implementation insights from the PEACE pathway across UK eating disorder services
Get PDFBackground/Objectives: Autistic individuals with eating disorders (ED) face socio-emotional, sensory, and communication difficulties that influence engagement and treatment outcomes. We examined how the PEACE Pathway—an autism-informed approach to ED treatment—addresses these challenges through tailored adaptations in clinical care. Methods: A qualitative multiple case studies design was employed, drawing data from clinical documentation, stakeholder feedback, and service evaluations. Results: We identified eight core domains essential for implementation: pathway knowledge, assessment and planning, psychological interventions, sensory management, nutritional care, lived-experience feedback, family/community engagement, and staff training. These domains informed the development of the PEACE Self-Assessment Checklist to support the wider adoption of the pathway. Conclusions: The PEACE Pathway offers a structured approach to adapting ED treatment for autistic individuals. The checklist provides practical guidance for implementing autism-friendly adaptions
Hide and seek with falsified medicines: Current challenges and physico-chemical and biological approaches for tracing the origin of trafficked products
Get PDFThe criminal trafficking of falsified medical products is a worldwide, yet still largely overlooked, public health problem. A falsified medicine fraudulently misrepresents its identity, composition and/or source, often being ineffective or toxic for patients. Although techniques have been developed to detect falsified medicines, it remains a challenge to trace where- and by whom- the products are manufactured. We aim to discuss plausible biological and physico-chemical analytical techniques that could reveal information about the origin of medical falsifications. We first provide a brief overview on the prevalence, criminal activities, health impacts and (bio)chemical features of falsified medical products. We then explore diverse laboratory approaches, that are used in food fraud, illicit drug and wildlife trafficking investigations, and discuss how they could be combined and redirected towards tracing falsified medicine origin and hence empowering enforcement to counter this pernicious but neglected global health problem.</p
A population-based cross-sectional investigation of COVID-19 hospitalizations and mortality among autistic people
Get PDFCurrent evidence suggests the possibility that autistic people may be at more risk of COVID-19 infection, hospitalisation, and mortality than the general population. Previous studies, however, are either limited in scale or do not investigate potential risk factors. Research into risk factors focused on general population samples. The current study aims to investigate these risk factors in the autistic population. Using data-linkage and a whole-country population, this study modelled associations between autism and COVID-19 hospitalisation and mortality risk in adults, investigating a multitude of clinical and demographic risk factors. Autistic adults had higher rates of hospitalisation, Standardised Incident Ratio 1.6 in 2020 and 1.3 in 2021, and mortality, Standardised Mortality Ratio 1.52 in 2020 and 1.34 in 2021, due to COVID-19 than the general population. In both populations, age, complex multimorbidity and vaccination status were the most significant predictors of COVID-19 hospitalisation and mortality. Effects of psychotropic medication varied by class. Although similar factors exhibited a positive association with heightened risk of severe COVID-19 in both the autistic and general populations, with comparable effect sizes, mortality rates were elevated among the autistic population compared to the general population. Specifically, complex multimorbidity and classification of prescribed medications may emerge as particularly significant predictors of severe COVID-19 among individuals within the autistic population due to higher prevalence of complex multimorbidity in the autistic population and variability in the association between medication classes and severe COVID-19 between both populations, though further research is needed
Characteristics of influenza, SARS-CoV-2, and RSV surveillance systems that utilise ICD-coded data: A systematic review
Get PDFBackgroundSome surveillance systems for influenza, severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), and respiratory syncytial virus (RSV) utilise International Classification of Diseases (ICD)-coded data and are useful for analysing trends and enhancing quick, evidence-based decisions against the epidemic potential that threatens global health security. With variations in the design of systems globally, the World Health Organization requested a systematic review to identify key characteristics of influenza, SARS-CoV-2, and RSV surveillance systems that utilise ICD-coded data, and to assess their performance.MethodsWe searched EMBASE, MEDLINE, and Global Health to identify relevant studies reporting on influenza, SARS-CoV-2, and RSV surveillance systems that use ICD-coded data. We independently assessed studies for the ICD codes used, their statistical estimates and limitations. We appraised included studies using Joana Briggs Institute’s critical appraisal tools and synthesised using narrative synthesis.ResultsWe identified 77 studies, reporting on 71 surveillance systems – 33 systems recorded surveillance data only, 15 systems recorded burden of disease data only, and 23 systems recorded both surveillance and burden of disease data. Surveillance systems utilised ICD-10 codes (75%), ICD-9 codes (22%), or both (3%). ICD-10 codes J09 and J10 were frequently used for influenza, U07.1 for COVID-19, and B97.4, J12.1, J20.5, and J21.0 for RSV. ICD-9 codes 487 and 488 were mostly used for influenza, and codes 466.11 and 480.1 for RSV. ICD-10 codes had low-to-moderate sensitivity (6.60–79.87%) and high specificity (97.40–99.72%) for influenza, low-to-high sensitivity (30.00–98.4%) and specificity (39.50–99.80%) for COVID-19, and low-to-high sensitivity (6.00–99.80%) and specificity (12.10–100.00%) for RSV. ICD-9 codes had low sensitivity (45.60%) and high specificity (97.90%) for influenza. Underestimation of infections or mortality attributable to influenza, SARS-CoV-2, or RSV is a major limitation to using ICD-coded data across surveillance systems.ConclusionsThe performance of ICD codes for syndromic- or disease-specific surveillance remains inconclusive, although using only ICD-coded data within these systems may underestimate influenza, SARS-CoV-2, or RSV-attributable morbidity and mortality. Future studies should assess the accuracy of ICD code combinations for surveillance of influenza, SARS-CoV-2, and RSV.<br/
Interactions between tone and voice quality in San Sebastián del Monte Mixtec modal & rearticulated vowels
No full textSan Sebastián del Monte Mixtec (henceforth SSM), also known as Tò’on Ndà’vi, is a language of the Mixtecan family, Otomanguean stock. SSM has lexical tones that are orthogonal to rearticulation on vowels. The aim of this production study was to examine both long modal and rearticulated vowels to gain insight into the SSM tonal system, contrastive voice quality, and any potential interactions between voice quality and f0. Rearticulated vowels are described as having a glottal gesture between two vowels of the same quality (V]ˀV), while modal vowels have no such gesture (VV). To this end, we examined the phonetic realization of the lexical tones in long modal vowels in terms of f0. All tones are distinguished by f0; f0 patterns largely as expected given previously ascribed labels, with minor deviations. Secondly, the phasing and degree of glottalization in rearticulated vowels was measured using “strength of excitation” (SoE); generally the glottal gesture was vowel medial with a dip in SoE at the beginning of the glottal gesture and a rise in SoE following the glottal gesture. However, there was a large degree of interspeaker variation in the production of rearticulated vowels. Additionally, lexical tone category was found to have an impact on the phasing and degree of glottal gesture in rearticulated vowels, and on voice quality in long modal vowels. This supports the idea that voice quality is an additional correlate of lexical tone in SSM
Simulating endogenous institutional behaviour and policy implementation pathways within the land system
No full textPolicy interventions have substantial effects on land use change, providing key levers for multiple objectives, including mitigating climate change and biodiversity loss, and maintaining food security. Policy effects are often complicated, conflicting, and subject to regular change. Despite this, land system models typically treat policies as simple, exogenous modifications to models. To better represent the dynamic nature of policy-making, we develop an endogenous institutional model that can be embedded within land system models, here exemplified by an agent-based model. Numerical experiments are conducted to examine an institution with two policies targeting the production of ecosystem services. We find a clear scope for simulation-based exploration of policy-making, with emergent processes including the marginal diminishing effect of economic policy interventions, asymmetric spill-over effects for different ecosystem services, and trade-offs between policy goals. The endogenous institutional model demonstrates the potential to reveal various emergent patterns with important consequences for land systems.</p
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!