Open Research Exeter - University of Exeter
Not a member yet
41213 research outputs found
Sort by
The Politics of Technological Pathways and Developmental Alliances in the EV transition: the cases of Brazil and Mexico
No full textThis article examines variation in green industrial policies for electrified vehicles (EVs) in Brazil and Mexico. Both are middle-income democracies with significant automotive sectors, yet they have adopted distinct technological pathways under similar global decarbonization pressures. We argue that technological choices are mediated by sectoral developmental alliances whose preferences are primarily structured by the politics of national growth models.
Using a descriptive comparative analysis, we show that Brazil’s commodity-driven model and large domestic market have supported an alliance between automakers and biofuel producers, leading to the prioritization of ethanol-compatible hybrid vehicles. By contrast, Mexico’s export-led integration into North American value chains has reinforced alliances aligned with battery electric vehicles (BEVs), consistent inherent pressures of the growth model and regulatory dynamics. The comparison advances a plausible hypothesis: in peripheral economies, green technological pathways are politically negotiated outcomes shaped by the politics of developmental alliances, rather than purely efficiency-driven responses to global climate imperatives.</p
Enhancing RNA Foundation Models via Secondary Structure Modelling
No full textGenomic Foundation Models (GFMs) are reshaping our understanding of the code of life, yet their application to Ribonucleic Acid (RNA) is uniquely challenged by its functional dependence on complex structure. Modelling this intricate sequence-structure-function relationship represents a vital problem in computational biology. This thesis aims to resolve three core impediments to progress in structure-aware RNA Foundation Model (FM). First, the interpretability gap, which obscures how models leverage structural information for biological discovery. Second, the fundamental sequence-structure alignment bottleneck, which hinders the sophisticated modelling of RNA for tasks such as design. Finally, the reproducibility crisis, which impedes the rigorous evaluation of competing GFMs. To address these challenges, this thesis presents a multi-scale solution focused on advancing RNA FMs via structure-aware modelling. First, we developed PlantRNA-FM, a high-performance, interpretable foundation model pre-trained with integrated structural information for the plant domain. This model successfully identified new, experimentally validated functional RNA structural motifs, demonstrating the potential of GFMs as engines for structure-based scientific discovery. Second, we propose the OmniGenome model, which fundamentally resolves the sequence-structure alignment problem by learning a robust, bidirectional mapping. The model achieved unprecedented success on the challenging EternaV2 RNA design benchmark, demonstrating a true capability for structure-based generative tasks. Finally, to tackle systemic evaluation issues, we constructed OmniGenBench, a modular, automated benchmarking platform for the rigorous assessment of structure-aware GFMs. Integrating over 31 open-source models and 123 standardised datasets, it provides a transparent and reproducible evaluation standard for the entire field. Taking the above solutions into consideration, this thesis not only enhances the performance and functionality of RNA FMs but also facilitates their transformation from opaque predictors into trustworthy, interpretable tools in structure-aware genomics.</p
Exploring Key Social Skills Supporting the Transition of Adolescents with Autism to the Workplace in The Kingdom of Saudi Arabia
No full textThe current study aimed to provide valuable insights to specialists in the field regarding the social skills needed to help adolescents with autism transition from school to the workplace.</p
Federated Learning for Traffic Flow Prediction with Synthetic Data Augmentation
No full textDeep learning-based traffic prediction models require vast amounts of data to learn embedded spatial and temporal dependencies. The inherent privacy and commercial sensitivity of such data has encouraged a shift towards decentralised data-driven methods, such as Federated Learning (FL). Traditional machine learning models capture spatial and temporal relationships in centralised data. In reality, traffic data is likely distributed across separate data silos owned by multiple stakeholders. In this work, a cross-silo FL setting is motivated to facilitate stakeholder collaboration for optimal traffic flow prediction applications. This work introduces FedTPS, a novel framework which augments each client’s local dataset with synthetic data generated by a federated diffusion-based model. The proposed framework is evaluated on three large-scale real-world datasets and assessed against various generative and spatio-temporal prediction models, including a novel prediction model we introduce which leverages Temporal and Graph Attention mechanisms to learn embedded Spatio-Temporal dependencies. Experimental results show that FedTPS outperforms several FL baselines with respect to global model performance.</p
Light-dependent switching of circling handedness in microswimmer navigation
No full textMany swimming microorganisms navigate their environment by modulating the curvature of their swimming trajectories in response to external cues. Here, we show that the biflagellate alga Chlamydomonas reinhardtii swims in circles and actively switches its trajectory handedness in response to orthogonal illumination: the cell swims counterclockwise at low light intensities yet clockwise at high light intensities. This handedness switching arises from light-dependent modulation of flagellar beating, including rapid and reversible changes in beat extension, phase, and-crucially-beat plane orientation. Using high-speed imaging and hydrodynamic modeling, we reveal that this beat plane reorientation is critical for Chlamydomonas to swim orthogonally to light as well as to dynamically modulate its trajectory curvature, enabling transitions between global exploration and localized searching in spatially structured light fields. Our results establish beat plane reorientation as a novel mechanism for curvature control in microswimmer navigation.</p
Detection of Solar Prominences using Automated Methods and Analysis of their Dynamic Pre-Eruptive Behaviour
No full textSolar prominences, or filaments when viewed on the disk, are cool, dense plasma structures in the solar corona. Prominences are of great interest for space weather and can be seen as a natural laboratory to explore concepts of plasma physics and magnetism. The eruption of prominences can have a significant influence on the solar-terrestrial environment. However, accurately predicting these eruptions remains a challenge. We first explore the oscillatory dynamics of prominences in the Extreme Ultraviolet (EUV), and how these may change in the build-up to their eruption. Two events are studied. Firstly, an eruptive filament is analysed using data from the Solar and Heliospheric Observatory (SoHO), revealing similar ultra-long period oscillations in both the 304 Angstrom irradiance and the intensity timeseries from an image sequence in 195 Angstrom. For the second event, we develop and apply automated detection methods for EUV prominences observed by the twin spacecraft from the Solar Terrestrial Relations Observatory (STEREO) mission and the Solar Dynamics Observatory (SDO) near Earth. During March 2011, when each STEREO spacecraft is in quadrature with respect to the Earth, for two time ranges, we obtain longitudinal height profiles as a function of time. We also track the corresponding EUV filaments across the solar disk, which reveal the emergence of ultra-long-period oscillations in the EUV filament channels. Our analysis shows a correlation between the prominence's increasing height and the oscillation periods, suggesting a potential link to the subsequent eruption observed by the STEREO spacecraft off-limb. These findings offer new insights into prominence dynamics and may pave the way for improved eruption prediction. We also present new methods for detecting prominences, using both a multi-wavelength approach and a single-wavelength detection with machine learning. First, we analyse the distribution of prominences over a duration greater than one full solar cycle. We present a single-wavelength detection in 304 Angstrom, which uses a type of neural network known as a Mask Region-based Convolutional Neural Network (MRCNN). This reveals the latitudinal evolution of prominences across the solar cycle, while we find the prominence statistics lag behind the sunspot count by approximately 3 months. These results are consistent with observations and theory surrounding prominence formation. Several open questions remain regarding prominences, their dynamics and how we may optimise their detection. Discussions arising from this research may present opportunities for future projects and investigations into this field.</p
Evolving classifiers for early detection of diabetic retinopathy from retinal images
No full textBackground and Aims: This research investigates the use of evolutionary computation to enhance automated classification of diabetic retinopathy (DR) from retinal fundus photographs. The primary aim is to improve diagnostic precision and streamline screening processes for early-stage DR in clinical ophthalmology. Methods: The study employed a publicly available fundus image dataset sourced from India. All images were captured with Eidon technology, which uses confocal wide-field Scanning Laser Ophthalmoscopy (SLO) for high-resolution, non-invasive retinal imaging. Two experienced clinicians manually reviewed each image to label the presence and severity of DR, assigning cases to one of three categories: No Diabetic Retinopathy, Mild DR, or Moderate Non-Proliferative Diabetic Retinopathy (NPDR). An evolutionary algorithm was applied to train classification models, refining both the structure and weights of neural networks over multiple generations. For model evaluation, a balanced subset of 40 images was used, including 20 showing signs of DR and 20 without. Training spanned 400 generations, resulting in a peak classification accuracy of 62.5% and a minimum cost value of 0.3. Additional short-term training over 7 generations was performed to analyze early-stage learning behavior. Results: Diabetic retinopathy remains a major cause of vision loss worldwide, underscoring the need for early screening solutions. This study demonstrates that evolutionary learning techniques can contribute to automated DR grading systems. Conclusions: Specifically, neuro-evolution approaches combining neural networks with genetic optimization present promising opportunities for enhancing diagnostic tools, improving both efficiency and classification performance in ophthalmic care.</p
Comparative Analysis of Model Predictive Control (MPC) Algorithms for Optimizing Blood Glucose in Fully Closed-Loop (FCL) Systems
No full textObjective: Model Predictive Control (MPC) is gaining traction in fully closed-loop (FCL) systems as a promising approach for automating glucose regulation in individuals with Type 1 Diabetes. This review evaluates the clinical effectiveness of current FCL systems and explores opportunities for future optimization by comparing recent technological advancements. Method: A comprehensive literature search was conducted across PubMed, TRIP Pro, and Web of Science databases to identify relevant studies. Keywords included “Fully Closed Loop,” “Type 1 Diabetes,” “Model Predictive Control,” “Continuous Glucose Monitoring,” and “Artificial Pancreas.” Boolean operators were applied to refine the search: “AND” ensured relevance by linking key terms, while “OR” broadened the scope to capture studies exploring various algorithmic approaches to FCL systems. Result: Current evidence indicates that MPC-based FCL systems outperform hybrid closed-loop (HCL) models using Proportional-Integral-Derivative (PID) control, with higher time-in-range (TIR) values (74.4% vs. 63.7%, P = 0.020) and improved postprandial glucose control. However, most systems still fall short of consistently exceeding the clinical TIR target (>70%). Persistent challenges include postprandial hyperglycemia and delays in insulin absorption. Recent developments in FCL design include nonlinear MPC (NMPC) for dual-hormone systems incorporating glucagon to reduce hypoglycemia, λ-Policy Iteration (λ-PI) using adaptive reinforcement learning, and pulse-modulated artificial pancreas (PMCL) systems designed to replicate natural insulin secretion patterns. Conclusion: While these innovations show promise in simulation studies, clinical validation remains limited. Key obstacles include glucagon stability issues, continuous glucose monitoring inaccuracies, high system costs, and variable patient adherence. Future research should focus on long-term clinical trials that account for real-world factors such as physical activity and dietary variability. By integrating predictive control, adaptive learning algorithms, and dual- hormone delivery, FCL systems have the potential to revolutionize diabetes management and move closer to fully autonomous glucose regulation.</p
Role of peer-tutors with dementia in Recovery College dementia courses: an ethnographic account
No full textBackground and objectives: Receiving a diagnosis of dementia impacts life plans and can lead to feelings of hopelessness and social disengagement. Post-diagnostic support can help people adjust to and assimilate a changing identity. Recovery Colleges in the UK offer a specific form of post-diagnostic peer-led support. This paper aims to provide a rich account of ‘stand out’ moments where the key tenets of recovery-focused post-diagnostic support were enacted. Research design and methods: Using ethnographic observations and interview data from the anonymized Study, a realist evaluation of Recovery College dementia courses, we examined data to specify activities of peer-tutors and the mechanisms which shaped outcomes for people with dementia. Results: Five Recovery College dementia courses were observed across four NHS mental health services in England. Post-course interviews were undertaken with 13 tutors (3 peer-tutors with dementia) and 32 attendees (8 people with dementia). We found that through co-facilitation of recovery-focused content by peer-tutors who have well developed facilitation skills, attendees appeared to mediate self-stigma, manage emotional uncertainty and make meaningful social connections in ways which engendered hope for their future. Discussion and implications: Identifying the activity between peer-tutors with dementia and course attendees foregrounds key strengths and limitations of this distinctive form of post-diagnostic support. Future work should evaluate longer term outcomes for people with dementia attending recovery courses before potentially expanding this form of post-diagnostic support.</p
Cryptococcal Antigenemia in South African Children Living With HIV.
No full textBackground: Cryptococcal meningitis (CM) is associated with high mortality and neurodevelopmental sequelae and predominantly affects people living with HIV. Screening and pre-emptive treatment for cryptococcal antigen (CrAg) in blood in adults with CD4 Methods: Children (Results: Prevalence of cryptococcal antigenemia in South African children Conclusions: CrAg prevalence in children with CD4 <100 cells/µL is comparable to adults (4.7% and 5.8%, respectively). CrAg screening guidelines should be extended to include all children to improve outcomes.</p