Open access to publications, including peer-reviewed articles, conference papers, working papers, reports, and other scholarly communications by University of Galway researchers.
University of Galway Research Repository
Recent Submissions
Publication Open Access Activation of self-assembled CaWO4 nanocrystals via Mn-incorporation toward nonenzymatic electrocatalytic glucose oxidation(American Chemical Society, 2025-12-28)Integrating foreign metal ions into mixed metal oxide scaffolds, such as CaWO4, by a wet chemistry coprecipitation method is an efficient approach for regulating the chemical composition, which determines the structure and properties and, in turn, offers a versatile platform for enhancements of intrinsic properties and introduction of additional attributes in the host material. In this work, a series of Ca1–xMnxWO4 was prepared via a coprecipitation process at 30 °C to exhibit sphere-like hierarchical structures assembled from individual nanocrystals. The solid solubility limit of Mn in CaWO4 is estimated to be ∼24% for this synthetic approach. The nanocrystals exhibit nonstoichiometric amorphous layers on the surface, which are expected for this type of materials synthesized under such mild conditions. The successful incorporation of Mn into the crystal lattice of CaWO4 led to an isotropic contraction of the crystal unit cells as well as the shifting and merging of several vibrational modes. Cyclic voltammetry confirmed the activity of Ca1–xMnxWO4 for the electrooxidation of glucose, which is directly linked to the presence of Mn in CaWO4. With the highest Mn content, Ca0.759Mn0.241WO4 achieved sensitivities of 3.80 and 1.00 μA·mM–1·cm–2 for glucose concentrations in the ranges of 0.002–0.2 and 0.2–1 mM, respectively, which are 237 and 60 times increases from that of CaWO4. The limit of detection and lower limit of quantitation were estimated to be 14 and 48 μM, respectively, with reasonable reproducibility and repeatability.Publication Open Access A deep learning approach to automate human-visual inspections in the medical-devices manufacturing industry(University of Galway, 2026-01-15)This thesis investigates how deep learning can be used to automate visual inspection in medical device manufacturing, a domain where extremely high accuracy is required and manual inspection remains prevalent despite its susceptibility to fatigue, variability, and error. Traditional rule-based machine vision performs well for dimensional checks but is poorly suited to attribute inspections that depend on subtle, context-dependent visual cues. To address this gap, the research focuses on methods that can be trained predominantly on defect-free samples, reflecting the rarity and cost of obtaining representative defective parts in highly regulated environments. The work makes four main contributions. First, it introduces and publicly releases a dataset of 1,200 annotated images of sterile barrier packaging with realistic seal defects, enabling reproducible benchmarking. Second, it reports a large-scale evaluation of 27 convolutional neural network architectures across multiple small and imbalanced dataset regimes, and proposes the median distance to the boundary as a robustness metric for detecting overfitting when test accuracies approach 100%. Third, it develops two attention-guided autoencoder architectures for deep anomaly detection: a structural-similarity variant tailored for lightweight, real-time inspection on production lines, and a feature-distance variant designed for lifecycle monitoring in supervisory systems; both are validated on an external dataset and shown to be competitive with state-of-the-art anomaly-detection approaches. Finally, the thesis provides a technical and regulatory analysis of the EU Artificial Intelligence Act and related medical-device frameworks, highlighting challenges in data governance, explainability, validation, and post-deployment monitoring. Overall, the findings demonstrate that deep learning models trained mainly on defect-free data can deliver scalable, high-performing and more easily qualifiable automated inspection systems, while also identifying open issues that motivate future research at the intersection of advanced manufacturing and AI regulation.Publication Embargo Risk factors and blood biomarkers for pre-clinical and clinical dementia(University of Galway, 2026-01-15)Identifying candidate plasma biomarkers associated with pre-clinical amyloid and tau PET deposition, vascular brain injury (VBI), and subsequent clinical dementia could illuminate underlying mechanistic pathways of disease and identify potential novel biomarkers for risk prediction. In addition, a greater understanding of the relationship between dietary and vascular factors, namely mid-life vitamin D and blood pressure across early to late mid-life, and preclinical dementia can inform population-based approaches to preclinical disease prediction and early disease modification. This thesis explores the association between candidate plasma proteins (including markers of vascular dysfunction and neurodegeneration), vitamin D and blood pressure trajectories in the mid-life period with neuroimaging markers of preclinical dementia as well as clinical dementia in the multi-generational, community-based, Framingham Heart Study. In the relatively young (mean age 39, standard deviation [SD] 8) Framingham Generation 3 cohort, vitamin D, blood pressure at mid-life, nerve growth factor (NGF) and candidate plasma proteins (n=125) were related to positron emission tomography (PET) and magnetic resonance imaging (MRI) measures performed approximately 13-16 years later. Following this, candidate biomarkers (n=55) identified as demonstrating significant associations with neuroimaging markers of preclinical dementia were validated for the outcomes of clinical all- cause and AD dementia in the older Framingham Generation 2 (Offspring) cohort. In this series of chapters, a steep decline in diastolic blood pressure (DBP) from early to late midlife was associated with greater tau-PET deposition in the entorhinal cortex, one of the earliest cortical regions affected in AD, highlighting how maintaining normotension across the ~15-16y mid-life period may confer protection against subsequent pathological tau accumulation in the brain. Higher vitamin D in mid-life was associated with lower subsequent tau-PET deposition on average 16 years later, but was not associated with amyloid burden in the brain, highlighting how mid-life vitamin D may represent a potentially modifiable target to mitigate the risk of neuroimaging markers of preclinical dementia. Higher nerve growth factor (NGF), a marker of neuroprotection and neural repair, was associated with a reduced risk of VBI, namely clinical stroke and covert brain infarcts, indicating its potential as a predictor of ischemic stroke risk. In the exploratory, ‘development’ chapter to identify potential candidate plasma protein biomarkers for neuroimaging markers of pre-clinical dementia, we identified 55 candidate proteins (many markers of systemic vascular dysfunction) associated with 19 Risk Factors and Blood Biomarkers for Pre-Clinical Dementia neuroimaging markers of preclinical dementia. Proteins reflecting endothelial dysfunction and fibrinolysis (such as growth differentiation factor-15 [GDF-15], plasminogen activator inhibitor-1 [PAI-1], leptin, and proinsulin) were associated with white-matter vascular brain injury (VBI), while markers of inflammation and blood–brain barrier disruption (including tau, butyrylcholinesterase [BChE], insulin-like growth factor binding protein-3 [IGFBP-3], bikunin, melanoma cell adhesion molecule [MCAM], and soluble glycoprotein 130 [sGP130]) were associated with amyloid deposition. Proteins involved in perfusion, purine metabolism, and excitability (such as N-terminal pro-B-type natriuretic peptide [NT- proBNP], purine nucleoside phosphorylase [PNP] and insulin-like growth factor-1 [IGF-1]) were associated with tau burden in the entorhinal cortex and medial temporal lobe. These findings suggest that systemic vascular dysfunction may contribute to both white- and gray-matter pathologies, highlighting candidate biomarkers for VBI that could strengthen dementia risk prediction and support the vascular ‘V’ component expansion of the ATNIVS+ (amyloid [A], tau [T], neurodegeneration [N], inflammation [I], vascular [V], and synaptic dysfunction [S]) classification framework. In the final chapter (validation study), the 55 biomarkers that were previously associated with neuroimaging outcomes were validated in the older Framingham Offspring cohort with clinical dementia outcomes. A total of 14 midlife plasma proteins were associated with future dementia outcomes; higher levels of insulin-like growth factor binding protein-2 (IGFBP-2), stromal cell-derived factor-1 (SDF-1), PAI-1, matrix metalloproteinase-8 (MMP- 8), intercellular adhesion molecule-1 (ICAM-1), and total tau were associated with an increased dementia risk, while IGFBP-3 and amyloid-beta 42 (Aβ42) were associated with decreased dementia risk. These results highlight systemic pathways linking vascular dysfunction and subsequent neurodegeneration with clinical dementia. This thesis offers an integrated, ‘mid to late-life’ lifespan perspective on the role of select dietary, vascular and candidate blood biomarkers in the risk of preclinical and clinical dementia. Across two generations and four decades of follow-up, midlife systemic changes in DBP, vitamin D, NGF, and plasma protein biomarkers were associated with downstream amyloid and tau deposition, white matter injury, and ultimately clinical dementia.Publication Open Access Rank distributions of graphs over the field of two elements(University of Galway, 2026-01-14)A square matrix M represents a graph Γ if its nonzero off-diagonal entries encode the adjacencies of Γ according to a fixed vertex ordering. Over the field of two el ements, we study the distribution of ranks within the affine space of all matrices representing a particular graph. The motivating question is which graphs of or der n are represented by more matrices of rank n − 1 than of rank n. This reflects the fact that the most frequently occurring rank is not n but n − 1 in the space of all n × n matrices over F2, a property which is exceptional to F2. This thesis focuses on connected graphs that have a path or cycle as a subgraph induced on all but one vertex (called the extra vertex). The path graph Pn serves as the starting point of this study. The path graph is fundamental in the related and widely studied minimum rank problem, and provides a foundation for our later analysis of the set G P of graphs containing an induced path on all but one vertex. A main result is a characterisation of all such graphs that are represented by more matrices of rank n − 1 than rank n over F2. This is achieved by first examining the vectors in the nullspace of each matrix representing Pn. An expression for the difference α(Γ ) between rank n − 1 and rank n representations of a given graph Γ ∈ G P is determined in terms of these nullspace vectors. A recurrence is then established, expressing α(Γ ) in terms of α for graphs in G P for which the extra vertex has lower degree than in Γ . We classify all Γ ∈ G P satisfying α(Γ ) < 0 by first classifying those for which the extra vertex has degree 1, then using that to simplify and classify the degree 2 case, and continuing like this until it is shown that no such graphs exist for degree ⩾ 6. We then turn to the analogous problem for the cycle graph. We show that half of all F2-matrices representing Cn have rank n − 1, approximately one-third have rank n, and approximately one-sixth have rank n − 2 . We then investigate the set of graphs containing an induced cycle on all but one vertex, denoted by G C . Our analysis reveals essential structural contrasts between the classes G P and G C : while the degree of the extra vertex is bounded in the path case, it can be arbitrarily large in the cycle case. An infinite family of graphs called alternat ing wheel graphs demonstrate this contrast, as there exists an alternating wheel graph Γ ∈ G C with an extra vertex of any even degree d ⩾ 4 satisfying α(Γ ) < 0.Publication Open Access Principles of Lipschitz continuity in neural networks(University of Galway, 2026-01-14)Deep learning has achieved remarkable success across a wide range of domains, significantly expanding the frontiers of what is achievable in artificial intelligence. Yet, despite these advances, critical challenges remain --- most notably, ensuring robustness to small input perturbations and generalization to out-of-distribution data. These critical challenges underscore the need to understand the underlying fundamental principles that govern robustness and generalization. This understanding is indispensable for establishing deep learning systems that are: reliable --- performing consistently under expected conditions on in-distribution data; resilient --- capable of recovering from unexpected conditions such as noise or adversarial attacks; and trustworthy --- behaving transparently, ethically, in alignment with intended use, and technically robust, particularly in safety-critical applications. Among the theoretical tools available, Lipschitz continuity plays a pivotal role in governing the fundamental properties of neural networks related to robustness and generalization. It quantifies the worst-case sensitivity of network's outputs to small input perturbations. While its importance is widely acknowledged, prior research has predominantly focused on empirical regularization approaches based on Lipschitz constraints, leaving the underlying principles less explored. This thesis seeks to advance a principled understanding of the principles of Lipschitz continuity in neural networks within the paradigm of machine learning, examined from two complementary perspectives: an internal perspective --- focusing on the temporal evolution of Lipschitz continuity in neural networks during training (i.e., training dynamics); and an external perspective --- investigating how Lipschitz continuity modulates the behavior of neural networks with respect to features in the input data, particularly its role in governing frequency signal propagation (i.e., modulation of frequency signal propagation). Guided by these perspectives, the thesis formulates three primary research questions: (RQ1) State of Knowledge --- what is the state of knowledge of Lipschitz continuity in neural networks? (RQ2) Training Dynamics --- how does Lipschitz continuity in neural networks evolve during the training process? (RQ3) Modulation of Frequency Signal Propagation --- how does Lipschitz continuity modulate frequency signal propagation in neural networks?
Communities in Research Repository
Select a community to browse its collections.