32151 research outputs found
Sort by
Optimising resource utilisation in collaborative machine learning systems
Get PDFCollaborative Machine Learning (CML) enables multiple user devices and a central server to train models together without moving raw data off-device. In such systems, training is distributed across devices and the server, which perform their respective computations and exchange model parameters or intermediate results. This thesis optimises system efficiency while preserving the standard CML privacy boundary.
In practice, however, CML systems often have low resource utilisation due to several bottlenecks. First, limited computational capabilities on devices makes it difficult to support model training. Second, devices have heterogeneous computational capabilities, and slower ones delay overall training progress. Third, frequent data exchange between devices and the server introduces substantial communication overhead. Finally, reducing computational and communication costs often leads to lower model accuracy, especially when data distributions vary across devices.
Existing CML systems typically address only a subset of the aforementioned bottlenecks, and lack a unified solution that improves overall resource efficiency. This thesis proposes three systems that collectively addresses the abovementioned bottlenecks by reconstructing the training pipeline. The first system leverages pipeline parallelism to reduce idle time caused by the mutual waiting between devices and the server during training. Specifically, the model is split between the device and the server, and both sides process different inputs to enable concurrent execution. Second, an asynchronous training system is developed to allow devices to train models independently without waiting for the server or other devices, mitigating the impact of slow devices and further reducing idle time. Finally, a system is proposed to reduce the frequent exchanges of intermediate results between server and devices to one-shot transfer, thereby reducing communication overhead. The server model is trained on the consolidated intermediate results, which alleviates the impact of data heterogeneity and improves model accuracy.
The proposed systems demonstrate consistent improvements over state-of-the-art baselines on multiple models and datasets: lower device-side computation and idle time, reduced total communication, and higher test accuracy under heterogeneous data. Overall, the proposed pipeline reconstruction yields faster accuracy-over-time progress and provides a foundation for building more efficient and scalable systems in heterogeneous and resource-constrained environments
Memory trajectories by migration status and gender : a life-course intersectional perspective
No full textFunding: Silvia Loi was funded by the European Union (ERC Starting Grant, MigHealthGaps, 101116721).This study explores the relationships between migration, gender, and memory trajectories over the life–course within an intersectional framework. Memory is an important dimension of cognitive decline, a critical concern in aging populations, and varies significantly across different demographic groups. In this paper we address the understudied influence that the intersection of gender and migration status has on memory trajectories and evaluate the role of age at migration. Using random-effects growth curve models, our findings reveal persistent memory disparities by migration status, particularly in early older age, though differentials by migration status generally reduce after age 70. Immigrants, overall, exhibit lower baseline memory functioning levels and steeper declines. Contrary to expectations, immigrant women are not uniquely vulnerable; instead, gender differences persist across groups, with men generally faring worse. Later age at migration predicts poorer memory performance, though a late-life advantage emerges for men who migrated at ages 6–17 warranting further investigation. These results underscore the importance of considering baseline levels, longitudinal patterns, and intersectional factors in cognitive aging research by migration status.Peer reviewe
Transitions in and out of loneliness during the COVID-19 pandemic : a latent class analysis of older adults in England
No full textFunding: This study is supported by the Economic and Social Research Council (Grant ES/P000681/1) and a Natural Environment Research Council grant.Disease control measures during the COVID-19 pandemic may have intensified loneliness among older adults, though experiences varied based on individual vulnerabilities and resources. This study examines loneliness trajectories among older adults using the English Longitudinal Study of Ageing, spanning four waves: two pre-pandemic (Wave 8: 2016–2017; Wave 9: 2018–2019) and two COVID-19 substudies (June–July and November–December 2020). The sample included 4492 respondents (17,968 observations). Latent class growth analysis identified four loneliness trajectories: ‘not lonely’ (73.5%), ‘pandemic loneliness’ (12.7%), ‘transitioned out of loneliness’ (6.9%) and ‘enduring loneliness’ (6.8%). Multinomial regression analysis explored predictors of trajectory membership. Younger age (50–74), being female, depression, COVID-related worries and disrupted daily routines increased the likelihood of belonging to ‘pandemic loneliness’ rather than ‘not lonely’. Optimism and strong partner support increased the likelihood of remaining ‘not lonely’ or transitioning out of loneliness. The pandemic's unintended effects, including routine disruptions and financial concerns, heightened loneliness risks, whereas psychosocial resources provided critical resilience. To prepare for future public health crises, policies should strengthen mental health support, promote social and economic stability and enhance social connection and resilience. Addressing psychosocial factors is essential to reducing loneliness and protecting older adults' well-being during and beyond periods of crisis.Peer reviewe
Collision-free robot scheduling
No full textThe authors thank the Leverhulme Trust for funding this research via the Leverhulme Research Centre for Functional Material Design. The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Nathan Flaherty reports financial support was provided by University of Liverpool Leverhulme Research Centre for Functional Materials Design (RC-2015-036).In this paper, we investigate the problem of designing schedules for completing a set of tasks at fixed locations with multiple robots in a laboratory. We represent the laboratory as a graph with tasks placed on fixed vertices and robots represented as agents, with the constraint that no two robots may occupy the same vertex at any given timestep. Each schedule is partitioned into a set of timesteps, corresponding to a walk through the graph (allowing for a robot to wait at a vertex to complete a task), with each timestep taking time equal to the time for a robot to move from one vertex to another and each task taking some given number of timesteps during the completion of which a robot must stay at the vertex containing the task. The goal is to determine a set of schedules, with one schedule for each robot, minimising the number of timesteps taken by the schedule taking the greatest number of timesteps within the set of schedules. We show that this problem is NP-complete for both star graphs (for k≥2 robots), and planar graphs (for any number of robots). Finally, we provide positive results for path, cycle, and tadpole graphs, showing that we can find an optimal set of schedules for k robots completing m tasks of equal duration of a path of length n in O(kmn), O(kmn2) time, and O(k3m4n) time respectively.Peer reviewe
Rare earth and niobium mobility at the magmatic-hydrothermal transition : Motzfeldt Sø Centre, Greenland
No full textFunding: CJWR acknowledges the generous support of Elemental Rare Metals in the present study. Samples were collected in several field seasons in 2001, 2005, 2006, 2017, and 2023. Fieldwork in 2017 was supported by the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 689909 to AAF. Fieldwork in 2023 was supported by Elemental Rare Metals. WH is funded by a UKRI Future Leaders Fellowship (MR/S033505/1). Drone photography was provided by Jan Richard Heinicke during the 2023 field season. Additional thanks are made to Anna Szreter, Ninni Jeramiassen and Ole Christiansen for their support and assistance in the field.Alkaline igneous complexes host the world’s largest deposits of rare earth elements (REE) and high field strength elements (HFSE) (e.g. Nb, Ta). The genetic models for these metals are underdeveloped, and the role hydrothermal fluids play in modifying the deposit is unclear. The Motzfeldt Sø Centre, Greenland offers an exceptional 3D slice through a collapsed roof-zone in which REE and HFSE mineralisation is found within pyrochlore-group minerals. We use textural and chemical analyses of pyrochlore to interrogate hydrothermal activity in the roof-zone, revealing a two-step model to form an REE-HFSE enriched deposit. First, magma stalls in the roof of the intrusion, accompanied by an aggressive, F-rich fluid from the cooling pluton below. Intensive hydrothermal alteration converts fluorcalciopyrochlore ( P1 ) by adding to A-site vacancy (□) to create an Fe/□-dominated pyrochlore ( P2 ). This liberates REEs, alkalis, and F from the pyrochlore A and Y sites. The elements form a secondary coating on the pyrochlore surfaces. Further alteration then dissolves the coating, enriching the fluid in REE, Nb, and Fe. This process forms mineralised veins containing columbite and REE-fluorcarbonates, which escape as “chimneys” in the upper-roof zone. Late-stage fluids play a key role in determining the nature of critical metal mineralisation, through reworking primary REE and Nb/Ta mineral assemblages into secondary hydrothermal phases, both proximal and distal to the primary deposit. This expands the genetic model for this deposit type, providing a robust framework for future resource exploration in alkaline mineralised roof-zone systems.Peer reviewe
Addressing epistemic violence and methodological nationalism through a meta-analytical review on intergroup contact and conflict studies in Turkey
No full textThis paper provides an examination of intergroup contact research in the context of the Turkish-Kurdish ‘conflict’ through a meta-analytical scoping review. By doing so, we discuss how dominant biases in the literature such as methodological nationalism and epistemic violence influence research practices and knowledge-production systems in conflict and contact studies conducted under colonial rule. Through a thematic investigation, we found four variables that were commonly tested in intergroup contact studies: i) attitudes towards outgroup, ii) support for minority rights or multiculturalism, iii) perception of discrimination against the minority group, and iv) ethnic identity. We conducted four meta-analyses using a pool of 28 studies conducted in Turkey to unpack the associations of intergroup contact with: i) outgroup attitudes (r = .42; n = 5,624), ii) support for minority rights and multiculturalism (r = -.06ns; n = 1,567), iii) perception of discrimination against Kurds (r = .22; n = 2,431), and iv) ethnic identity (r = -.13; n = 4,636). The results demonstrated a moderate to strong relationship between intergroup contact and positive outgroup attitudes, while the findings were less robust for support for minority rights, perception of discrimination against Kurds, and ethnic identity. Furthermore, the group status and sample characteristics of the studies moderate the relationships between variables. Based on the findings and a critical analysis of the current literature, we discuss the limitations of the existing research and critical points for future studies.Peer reviewe
Spinal circuit mechanisms constrain therapeutic windows for ALS intervention : a computational modeling study
No full textFunding: This work was funded by the Lundbeck Foundation, Denmark; Grant no. R426-2023-158 (IA, BS) and Aage og Johanne Louis-Hansens Fond, Denmark; Grant no. 23-2B-14112 (IA, BS), the School of Psychology and Neuroscience at University of St Andrews (IA), UK; the MRC UKRI, UK; Grant no. MR/Y014901/1 (IA) and the National Science Foundation NSF CRCNS/DARE, USA Grant no. 2113069 (JA).Amyotrophic Lateral Sclerosis (ALS) is a fatal neurodegenerative disease characterized by progressive breakdown of neural circuits which leads to motoneuron death. Earlier work from our lab showed that dysregulation of inhibitory V1 interneurons precedes the degeneration of excitatory V2a interneurons and motoneurons and that stabilizing V1-motoneuron connections improved motor function and saved motoneurons in the SOD1G93A ALS mouse model. However, the optimal timing for this intervention remains unclear. To address this, we developed a spiking neural network model of spinal locomotor circuits to simulate healthy and ALS-like conditions. By modeling changes in network connectivity and synaptic dynamics, we predict that V1 dysregulation induces an imbalance in motoneuron output which results in flexor-biased activity, leading to the disruption of flexor-extensor coordination, and potentially contributing to selective vulnerability of flexor motoneurons. Stabilizing V1 synapses preserved motor output even after motoneuron loss, suggesting that therapeutic benefit is possible into symptomatic stages. However, model predictions also highlighted that after sustained synaptic loss and the development of slower synaptic dynamics within the network, synaptic stabilization leads to maladaptive extensor-biased activity, suggesting that excitatory/inhibitory balance impacts treatment effectiveness. Finally, the model indicated that V1 stabilization could lead to rescue of the V2a excitatory interneurons, a finding that we were able to confirm experimentally in the SOD1G93A ALS mouse model. By exploring different scenarios of synaptic loss and cell dysregulation during synaptic stabilization, our models provide a framework for predicting candidate time windows for spinal circuit interventions, which may guide future preclinical investigations.Peer reviewe
Plasmid streamlining drives the extinction of antibiotic resistance plasmids under selection for horizontal transmission
No full textFunding: T.D. acknowledges funding support from the Royal Society (University Research Fellowship URF\R1\231740) https://royalsociety.org/. The funders played no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.Conjugative plasmids carrying antimicrobial resistance (AMR) genes are critical for the spread of AMR, due to their ability to transmit horizontally between bacterial hosts. We previously observed that during experimental evolution in the presence of abundant susceptible Escherichia coli hosts, the AMR plasmid R1 rapidly evolves variants with increased horizontal transmission due to mutations causing increased plasmid copy number. Yet AMR was progressively lost from the evolving populations. Here, we show that AMR loss was associated with evolution of streamlined plasmids in which the AMR region is spontaneously deleted, making plasmid carriage undetectable by plating on selective antibiotic-containing media. These plasmids transmit both vertically and horizontally more efficiently than the ancestral AMR plasmid, driving AMR extinction in bacterial populations and effectively acting as an intrinsic defence against AMR plasmids. A simple model of plasmid competition further shows that any horizontal or vertical transmission advantage conferred by plasmid streamlining would be enough to drive the displacement of competing AMR plasmids, with a given horizontal transmission advantage leading to faster replacement in conditions favoring horizontal transmission. Our results suggest that within-host plasmid evolution or engineered streamlined plasmids could be exploited to limit the spread of AMR in natural populations of bacteria.Peer reviewe
The work of the British School at Athens 2024-2025
No full textThis article, based on an oral presentation by the author at the BSA’s annual general meeting in February 2025, summarizes the activities of the British School at Athens with a focus on the calendar year 2024. It gives us great pleasure to present the innovative and varied work of BSA-sponsored field and research projects, the Fitch Laboratory, Knossos Research Centre, Archive, and Library as well as the inspiring work of the School students, post docs, and fellows.Peer reviewe
From plastic waste to pharmaceutical precursors : PET upcycling through ruthenium catalyzed semi-hydrogenation
No full textFunding: This research is funded by a UKRI Future Leaders Fellowship (MR/W007460/1).We report here the upcycling of PET (polyethylene terephthalate) waste via semihydrogenation to make ethyl 4-(hydroxymethyl)benzoate. The reaction is catalyzed by a ruthenium pincer catalyst at 80 °C in bioderived solvents – a combination of 2-methyl THF and ethanol. A detailed mechanistic investigation through organometallic and kinetic studies, as well as chemical exchange saturation transfer (CEST) NMR spectroscopy, provides insights into the nature of active species and factors that promote and inhibit the catalytic hydrogenation of PET. Using this mechanistic knowledge, a record high turnover number of >30 000 was achieved for the hydrogenative depolymerization of end-of-life PET waste (e.g., bottles and textiles). The semihydrogenation product, ethyl 4-(hydroxymethyl)benzoate, was utilized to make precursors of various known pharmaceutical drugs, an agrochemical, as well as a new and recyclable polyester. A cradle-to-gate life cycle assessment demonstrated that using PET waste as a feedstock for EHMB production significantly reduces the environmental footprint compared to the conventional route from p-toluic acid.Peer reviewe
Access Repository Dashboard
Do you manage St Andrews Research Repository? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!