Open Research Exeter - University of Exeter
Not a member yet
41213 research outputs found
Sort by
A Study of Locality in Hong Kong Contemporary Literature
No full textThis thesis investigates the narratives of Hong Kong locality by Wong Bik-wan (黃碧雲) and Dung Kai-cheung (董啟章), two prominent Hong Kong writers whose fiction is set in a Hong Kong local social and historical background. Since the signing of the Sino-British Joint Declaration (Joint Declaration) in 1984, resentment against the People’s Republic of China (PRC) has been growing among Hong Kong intellectuals. To Wong and Dung, the Joint Declaration merely marked a transfer of Hong Kong from the British Empire to the control of another hegemonic power, the PRC. They preferred to see Hong Kong included in the political discussion of the sovereign transfer as an autonomous entity. The absence of such a consultation process, in their view, would result in drastic social turmoil after the formal handover in 1997. Such an understanding gradually gained popularity in Hong Kong. This thesis takes their fictional works as cultural products constructing this political ideal. Their writings expressed a dystopic image in which the vibrant and diverse society of Hong Kong during the colonial era would be assimilated by the PRC’s governance. Such a dystopic vision of the post-1997 Hong Kong society is subtly brought forward by forging Hong Kong locality in their works. The locality of Hong Kong in Wong and Dung’s writings should be read in the context of such widespread intellectual anxiety. By cross-reading Wong and Dung’s narratives on locality with archival materials, the thesis argues that the locality they constructed is rooted in their nostalgia, elitism, and resentment, and represents a small fraction of the life of the general public in Hong Kong. The working classes, in this case, are largely silenced.
The thesis contributes to the study of locality in Hong Kong contemporary literature by illustrating the complexity of its social construction. It also demonstrates the conditionality of the locality constituted by Wong and Dung. By contextualising both writers’ works and their acclaimed representation of Hong Kong locality in the post-1990s era, this thesis intends to showcase the limits of the literary imagination they created. Ultimately, this thesis suggests that such narratives of locality should be seen as products of colonisation rather than decolonisation.</p
Incentivizing Multi-Tenant Split Federated Learning for Foundation Models at the Network Edge
No full textFoundation models (FMs) such as GPT-4 exhibit exceptional generative
capabilities across diverse downstream tasks through fine-tuning. Split
Federated Learning (SFL) facilitates privacy-preserving FM fine-tuning on
resource-constrained local devices by offloading partial FM computations to
edge servers, enabling device-edge synergistic fine-tuning. Practical edge
networks often host multiple SFL tenants to support diversified downstream
tasks. However, existing research primarily focuses on single-tenant SFL
scenarios, and lacks tailored incentive mechanisms for multi-tenant settings,
which are essential to effectively coordinate self-interested local devices for
participation in various downstream tasks, ensuring that each SFL tenant's
distinct FM fine-tuning requirements (e.g., FM types, performance targets, and
fine-tuning deadlines) are met. To address this gap, we propose a novel
Price-Incentive Mechanism (PRINCE) that guides multiple SFL tenants to offer
strategic price incentives, which solicit high-quality device participation for
efficient FM fine-tuning. Specifically, we first develop a bias-resilient
global SFL model aggregation scheme to eliminate model biases caused by
independent device participation. We then derive a rigorous SFL convergence
bound to evaluate the contributions of heterogeneous devices to FM performance
improvements, guiding the incentive strategies of SFL tenants. Furthermore, we
model inter-tenant device competition as a congestion game for Stackelberg
equilibrium (SE) analysis, deriving each SFL tenant's optimal incentive
strategy. Extensive simulations involving four representative SFL tenant types
(ViT, BERT, Whisper, and LLaMA) across diverse data modalities (text, images,
and audio) demonstrate that PRINCE accelerates FM fine-tuning by up to 3.07x
compared to state-of-the-art approaches, while consistently meeting fine-tuning
performance targets.</p
Understanding implementation science in medical radiation sciences
No full textObjectives: Radiography, like many allied health professions, faces persistent challenges in translating evidence and innovation into routine clinical practice. Despite a strong foundation in evidence-based practice, the adoption of new technologies, protocols, and models of care is often inconsistent, delayed, or un-sustained. This paper introduces Implementation science, which offers a key, yet underutilised approach for advancing radiographic practice by focusing on how evidence-based interventions are adopted, integrated, and sustained in real-world settings. Key findings: We present a conceptual overview of implementation science frameworks with particular relevance to radiography e.g. technology, devices and service improvement. Key frameworks considered include the Consolidated Framework for Implementation Research (CFIR), Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE-AIM), the Non-adoption, Abandonment, Scale-up, Spread, and Sustainability (NASSS) framework, the Theoretical Domains Framework (TDF), and Normalisation Process Theory (NPT). Each is appraised for its focus, strengths, and applicability to common implementation challenges in radiography. The frameworks highlight different but complementary perspectives, for example CFIR and TDF emphasise multilevel determinants of behaviour and RE-AIM structures evaluation of implementation outcomes. Applied examples from radiography and allied health illustrate how these approaches can be used to diagnose barriers, design strategies, and evaluate implementation efforts. Conclusion: Implementation science provides a rich methodological and theoretical toolkit for strengthening radiography research. By applying these frameworks, studies can move beyond questions of clinical efficacy to address the practical realities of translation, adoption and sustainability. Implications for practice: Embedding implementation science within radiographic research, practice, and education can support more systematic and context-sensitive translation. This shift enables the profession to progress from demonstrating clinical potential to delivering sustained improvements in service delivery, patient safety, and professional practice.</p
GraphRAG-Rad: Concept-Aware Radiology Report Generation via Latent Visual-Semantic Retrieval
No full textRadiology report generation involves translating visual signals from pixels into precise clinical language. Existing encoder-decoder models often suffer from hallucinations, generating plausible but incorrect medical findings. We propose GraphRAG-Rad, a novel architecture that integrates biomedical knowledge through a novel Latent Visual-Semantic Retrieval (VSR). Unlike traditional Retrieval-Augmented Generation (RAG) methods that rely on textual queries, our approach aligns visual embeddings with the latent space of the Knowledge Graph, PrimeKG. The retrieved sub-graph guides the Visual Encoder and the Multi-Hop Reasoning Module. The reasoning module simulates clinical deduction paths (Ground-Glass Opacity → Viral Pneumonia → COVID-19) before it combines the information with visual features in a Graph-Gated Cross-Modal Decoder. Experiments on the COV-CTR dataset demonstrate that GraphRAG-Rad achieves competitive performance with strong results across multiple metrics. Furthermore, ablation studies show that integrating latent retrieval and reasoning improves performance significantly compared to a visual-only baseline. Qualitative analysis further reveals interpretable attention maps. These maps explicitly link visual regions to symbolic medical concepts, effectively bridging the modality gap between vision and language.</p
Time to belong: Why management education needs a pedagogy of contemporaneity
No full textCalls to ‘transform’ management education often presume a linear temporal trajectory, from a deficient present towards a better future, while leaving the present itself unexamined. Drawing on philosophical accounts of contemporaneity as a conjunctively disjunctive historical condition, we argue that transformation must be grounded in how the present is lived and shared, not merely measured or projected beyond. Through auto-ethnographic vignettes of life under late Communism and subsequent migration to Britain, we show how ostensibly progressive narratives can reproduce exclusionary temporal politics, creating experiences of temporal displacement even among those chronologically ‘present’ in academic communities. We then propose a pedagogy of contemporaneity for management education – an adaptive scaffolding organised around three principles: commitment to the layered present (refusing nostalgic or utopian escape routes); collaboration across different temporal trajectories (not only across perspectives or disciplines) and contextualisation of learning within situated historical, social and political timescales. Rather than offering a blueprint for the future, this pedagogy equips educators and students to recognise and navigate temporal multiplicity as the condition of belonging in time. Our contribution is to recast transformation not as an endpoint but as the means of working with the temporal complexity that already constitutes our classrooms and institutions.</p
Greenland’s ‘green mining’ row highlights the key tensions in the energy transition
No full textNo abstract</p
Audio deepfake detection at the first greeting: “Hi!”
No full textThis paper focuses on audio deepfake detection under real-world communication degradations, with an emphasis on ultra-short inputs (0.5–2.0s), targeting the capability to detect synthetic speech at a conversation opening, e.g., when a scammer says “Hi.” We propose Short-MGAA (S-MGAA), a novel lightweight extension of Multi-Granularity Adaptive Time–Frequency Attention, designed to enhance discriminative representation learning for short, degraded in?puts subjected to communication processing and perturbations. The S-MGAA integrates two tailored modules: a Pixel–Channel Enhanced Module (PCEM) that amplifies fine-grained time–frequency saliency, and a Frequency Compensation Enhanced Module (FCEM) to supplement limited temporal evidence via multi-scale frequency modeling and adaptive frequency–temporal interaction. Extensive experi?ments demonstrate that S-MGAA consistently surpasses nine state-of-the-art baselines while achieving strong robustness to degradations and favorable efficiency–accuracy trade-offs, including low RTF, competitive GFLOPs, compact parameters, and reduced training cost, highlighting its strong potential for real-time deployment in communication systems and edge devices</p
Understanding the pattern, shifts and environmental impacts of dynamic plant communities in the Himalayan alpine zone (>4100 m.a.s.l.) under climate change
No full textThis PhD research aims to advance academic understanding of the ecological systems in the Himalayan Alpine zone (HAZ) with a specific focus on alpine plant community distribution, their ecohydrological impacts, and elevation-driven vegetation shifts under climate change. This study initially concentrated on Khumbu district in northeastern Nepal before expanding to six study regions across the Himalayan mountains. The HAZ was chosen as the research region because unlike areas at lower elevation it has been relatively neglected in ecological studies, despite its hypothesised importance in regulating the downstream water supply for billions of people across the region (Ives and Messerli 1990). The HAZ is also noted as being sensitive and vulnerable to climate change (Golovatch and Martens 2018). Despite its likely importance, its remoteness, high altitude, and lack of scientific attention have hindered comprehensive research efforts. Currently, the limitations of available remote sensing products constrain our understanding of long-term vegetation dynamics, while the lack of in-situ measurements prevents a detailed examination of interactions between alpine plant communities and environmental metrics. These knowledge gaps that centre on the HAZ are prohibiting understanding of the potential environmental implications of future climate shifts, which have been shown in modelling studies and in a few limited remote sensing investigations to lead to a greener mountain system at higher altitude.
This PhD uses remotely sensed data from multiple scales, coupled with in-situ measurements and observations in the alpine regions within Khumbu, to create the first land cover map of Sagarmatha National Park at a plant community level. The work then uses a field experiment in the Gokyo valley to understand ecological influences on land surface temperature, simulating the impacts on processes of potential evapotranspiration for different dominant plant species (Rhododendron spp. and Juniperus spp.). Large-scale cloud-based analysis of Landsat series data were then used to identify and track over time the changing upper limits of vegetation across 6 regions (organised longitudinally according to a major climatic gradient from west to east) in the Himalayan mountains. Additionally, field data from community-maintained trail cameras were used to monitor vegetation dynamics at two alpine villages, akin to a simple low-cost PhenoCam but using different spectral principles (Red, Green and Blue bands).
The major results of the study were as follows:
• A systematic review of alpine and Arctic studies revealed that there are likely to be profound hydrological implications arising from a greening HAZ (published in the journal Ambio).
• Dwarf shrubs belonging to the Rhododendron and Juniperus families dominate the ecology of the HAZ above 4100 m.a.s.l (published in the journal Arctic, Antarctic, and Alpine Research).
• Altitude and aspect are dominant drivers of vegetation community distribution (published in the journal AAAR).
• The presence of plants not surprisingly reduces maximum temperature (with 9.92-30.09 °C difference during the daytime) and increases the minimum temperature under shrub canopy compared to open areas (journal article in second review for Journal of Mountain Science).
• Juniperus plants were found to have stronger impacts on temperature than the Rhododendron plants, but the potential evapotranspiration of Rhododendron plants are higher than Juniperus plants (journal article in second review for JMS).
• in-situ observations from low-cost trail-cam phenological observations show distinctions between these two plant communities, and Juniperus plants are found to have longer growing seasons at higher altitudes (published in the journal Progress in Physical Geography).
• Across the entire Himalayan arc we found consistent upward trends in the location of the vegetation line, with varying rates from 1.42 m/year in Bhutan (the most eastern region) to 6.95 m/year in Manthang (central Himalayas; accepted to the journal Ecography).
• Across all regions, greening trends were more prevalent than browning trends, and larger area of greening with higher annual changing rates were particularly shown in central regions (Reckong, Ngari and Manthang; accepted to the journal Ecography).
This PhD is interdisciplinary, working between physical geography, remote sensing and social science. The problems that this PhD addresses are a complex combination of the current ecosystem situation, their likely changes under future climate warming, and the recognition that there is lack of long-term in-situ data existing for the Himalayan alpine zone, resulting from the lack of scientific attention and complicated by the remote geographical mountainous location.</p
Application of manganese oxide minerals to treat acidic waters contaminated with vanadium
No full textVanadium (V) is both toxic environmental contaminant and a strategically valuable metal, increasingly being released from metallurgical and energy industries. Its high mobility and toxicity under acidic conditions challenge conventional remediation technologies, while its growing demand creates opportunities for selective recovery. This thesis investigated manganese(III/IV) oxides (MnOx) as sustainable, reuseable sorbents for V removal from acidic aqueous systems, combining mechanistic analysis with validation in real industrial wastewater. A systematic comparison of natural, commercial, synthetic, and biogenic MnOx revealed strong contrasts in V uptake that were governed primarily by surface chemistry rather than surface area. Under acidic conditions (pH 3), natural marine MnOx exhibited the highest mass-bassed adsorption capacity (54.0 mg/g), reflecting its amorphous, defect-rich structure and very low point of zero charge. Synthetic MnOx (SynMnO) showed slightly lower uptake (26.0 mg/g) but superior structural stability and reusability, retaining performance over multiple adsorption-desorption cycles. In contrast, commercial crystalline MnOx (ComMnO) displayed limited reactivity due to a lower density of accessible surface sites. Biogenic MnOx produced by the cyanobacterium Synechococcus leopoliensis displayed a distinct adsorption behaviour. Despite its low Brunauer-Emmett-Teller surface area, it achieved the highest area-normalised V uptake (18.1 mg/m2), outperforming abiogenic oxides (0.36 mg V/m2 for SynMnO and 0.77 mg V/m2 for ComMnO). This superior reactivity is attributed to its nanoscale, disordered structure, low point of zero charge (pHpzc = 2.0), and biologically derived surface functionalities, underscoring the importance of defect chemistry and surface composition in acidic conditions. However, the precise role of associated organic matter in stabilising reactive sites or influencing V binding remains equivocal and could not be fully resolved within this study. Application of SynMnO to a real acidic industrial effluent from V2O5 production (pH 2.1, V = 41.1 mg/L, Cr = 454 mg/L, high Cl-/SO42- background) demonstrated rapid and selective V removal, with >95 % V uptake achieved within 1 minute at optimised dosages, while Cr removal was slower and diffusion-limited. Mn loss was negligible (</p
Photophysical Properties of Emerging 2D Materials
No full textPhotophysics is a branch of physical science that underpins a wide range of light-based applications. This field has emerged as a powerful tool for exploring light–matter interactions, enabling researchers to understand the behaviour of materials through optical and optoelectronic properties of light under different environmental conditions and timescales. Recently, the integration of two-dimensional (2D) materials into nanostructure-based devices has played a vital role in enhancing physical and chemical properties, particularly for optoelectronic applications. Since their introduction, 2D ultra-thin materials, with their efficient performance and low cost, have attracted considerable attention as a promising semiconducting material owing to their enhanced stability under ambient conditions as well as tuneable optical and electrical/optoelectronic characteristics associated with layer thickness.
The present study explores mechanically exfoliated 2D flakes as active materials for high-performance photodetectors fabricated via electron-beam lithography (EBL). The investigation includes 2D indium selenide (α-In₂Se₃), a layered semiconductor with unique ferroelectric behaviour and a tunable bandgap, where its performance is evaluated both in transistor and photodetector configurations. The results demonstrate thickness-dependent transport and photoresponse characteristics, revealing the potential of α-In₂Se₃ as a multifunctional 2D material for advanced optoelectronic devices. In parallel, efforts focus on hybrid 2D perovskites, which are examined as light-absorbing materials due to their strong excitonic effects and favourable optical properties. Despite their promise, perovskite crystals present significant challenges for device integration, as they are prone to structural degradation under ambient exposure and highly susceptible to the damaging effects of high-energy electron beams during EBL. These limitations often restrict their application to larger-scale devices fabricated by shadow masks or low-resolution patterning methods, which hinders progress toward miniaturized and scalable devices. To address this, lithographic parameters and resist processing are systematically optimized to minimize beam-induced damage, enabling the reliable definition of nanoscale electrodes while preserving the fragile perovskite crystal structure. These advances make it possible to fabricate top-down lithography-based perovskite photodetectors with well-defined device geometries, whose figures of merit, including responsivity, detectivity, and response speed, are systematically evaluated and benchmarked against state-of-the-art devices. In particular, for the first time, I will show how new strategies, such as fast lift-off and mitigating the dose of the electron-beam, contribute to enhancing their optoelectronic performance. Beyond fabrication and performance optimization, device stability is further enhanced through a novel encapsulation method using a beeswax/PMMA layered structure. This encapsulant provides long-term protection against moisture and environmental degradation, allowing the devices to operate stably under water for extended durations. The encapsulated photodetectors are also demonstrated in turbidity sensing experiments, confirming their robustness and versatility for real-world applications.</p