70015 research outputs found
Sort by
Exploring feedforward neural network explainability using the layerwise relevance propagation framework
No full textNeural Networks (NNs) can learn very accurate solutions to complex problems, but it is rarely clear how. The Layerwise Relevance Propagation (LRP) framework would explain how a given NN would produce a prediction for given data by assigning a relevance score to each data feature in each data example. This would be achieved by propagating each NN layer's output onto each data feature in its input. Other researchers have shown what hyperparameters and architectural choices lead to these explanations beinanalytically correct, however, it is not always possible to apply these in practice.
The first chapter discusses the problems and solutions that were explored in this research. The second chapter presents background literature about AI, NNs, model shade, explainability, and LRP. The third chapter compared explanations extracted by LRP to those extracted from white-box models. These were most comparable when the NN architecture was large and when the data that it was fitted on contained many data examples. This established a link between explainability and the predictive accuracy of a NN. The fourth chapter found that explanations generated by LRP can be made correct through hyperparameter optimisation, and the newly-proposed Local LRP (LLRP) framework exceeded the explainability of trained LRP over greyscale and colour images by learning the hyperparameters at each NN layer. Chapter five discovered and analysed why the actual and expected negative relevance representations differ, and the sensitivity of positive relevance was maximised individually instead of trying to mutually maximise positive and negative relevance. A final reflection in chapter six shows that this thesis has contributed to NN explainability by improving the relevance produced by LRP. Future research opportunities were highlighted throughout this work
From power source to waste burden: legal implications of using SMRs for AI infrastructure
No full textAs artificial intelligence (AI) compute scales, energy security has emerged as a key constraint. The UK’s AI Growth Zones aim to address this by integrating Small Modular Reactors (SMRs) as a primary power source. However, this strategy faces a structural paradox: while SMRs are being fast-tracked through streamlined licensing, planning, and environmental law, the UK lacks a long-term solution for nuclear waste disposal. This paper examines the regulatory barriers to SMR deployment, the unresolved risks of AI-powered nuclear infrastructure, and the governance challenges posed by nuclear waste and AI energy demand. It argues that unless nuclear policy, AI expansion, and waste governance are treated as interdependent, the UK risks undermining its AI supercomputing strategy. By analysing UK energy law, SMR licensing, and AI compute governance, the paper contributes to current debates on the trade-offs between AI energy security, nuclear governance, and long-term waste sustainability
How interest intensity and cognitive flexibility impact conversational topic maintenance in autistic and neurotypical children
No full text
Wage labor history in Central and Southern Africa
No full textLabor in southern Africa changed rapidly over the nineteenth and twentieth centuries as the region became firmly locked into the global capitalist economy. The development of wage labor in southern Africa adapted to the rapid changes brought about by the minerals revolution and the spread of commercial agriculture. This was also a period of rapid urbanisation, creating both opportunities and challenges for workers. Labor migration and the mobility of both people and ideas, tied the region together often creating a shared experience of working life in both the formal and informal economies. Southern Africa’s political, social, and economic development was fundamentally altered by processes of colonisation and the new forms of economic activity that arose from that intrusion. The large-scale exploitation of mineral resources in the late-nineteenth century and the entrenchment of European colonialism reshaped African societies and changed the nature of how, and why, people worked. Broadly speaking, the push of taxes combined with the pull of wages in new mines, farms, and factories slowly coaxed Africans into the capitalist economy. Growing urban centres, and the new occupations they generated, provided alternatives to mine and farm work, as well as providing opportunities for livelihoods based in the ‘informal’ economy, catering for the needs and desires of the growing number of urban residents
Inter-rater reliability of stress signatures in exfoliated primary dentition - Improving scientific rigor and reproducibility in histological data collection
Get PDFAccentuated Lines (ALs) in tooth enamel can reflect metabolic disruptions from physiological or psychological stresses during development. They can therefore serve as a retrospective biomarker of generalized stress exposure in archaeological and clinical research. However, little consensus exists on when ALs are identified and inter-rater reliability is poorly quantified across studies. Here, we sought to address this gap by examining the reliability of accentuated (AL) markings across raters, in terms of both the presence versus absence of ALs and their intensity (HAL= Highly Accentuated, MAL= Mildly Accentuated, RL= Retzius Line). Ratings were made and compared across observers (with different levels of experience) and pairs of raters (who agreed on AL coding through consensus meetings) (N = 15 teeth, eight observers). Results indicated that more experience in AL assessment does not necessarily produce higher reliability between raters. Most disagreements in intensity ratings occurred in categories other than HAL. Furthermore, when AL assessment was performed by pairs of raters, reliability was significantly higher than individual assessments (Gwet's AC1 = 0.28 to 0.56 for line presence assessment; Gwet's AC1 = 0.48 to 0.64 for line intensity assessment). Based on these results, we recommend a workflow called IRRISS (Improving Reliability and Reporting In Scoring of Stress-markers) to increase rigor and reproducibility in histological analysis of dental collections. The introduction of IRRISS is well-timed, given the surge in studies of teeth occurring across anthropological, epidemiological, medical, forensic, and climate research fields
The Effect of SPECTROM Training on Support Staff Knowledge of Psychotropic Medicine and Attitude Towards Behaviours That Challenge in Adults With Intellectual Disabilities to Help Implement the STOMP Initiative
Get PDFBackground
Overmedication of adults with intellectual (learning) disabilities, particularly the off-licence use of psychotropic medicines for behaviours that challenge in the absence of a psychiatric disorder, is a major public health concern and an example of health inequalities. In the United Kingdom, we coproduced online training resources backed up by face-to-face training for support staff (direct care workers) called SPECTROM involving all stakeholders, including adults with intellectual disabilities and their families, to help reduce the overmedication and implement NHS England's STOMP initiative.
Method
In a feasibility cluster randomised controlled trial, we trained service managers and support staff using two core modules of SPECTROM, namely, (a) Medicine/STOMP and (b) Alternatives to Medicine (ATM) using an online platform. These two core modules introduced 12 other modules and other online resources. We assessed trainees' knowledge of psychotropic medicines using the Psychotropic Knowledge Questionnaire-Revised (PKQ-R) and their attitude to behaviours that challenge using the Management of Aggression and Violence Attitude Scale-Revised-Intellectual Disabilities (MAVAS-R-ID) using a pre–post training design.
Results
The research team delivered SPECTROM training to 18 service managers and 122 support staff. Of the 140 trainees, 126 completed PKQ-R at baseline before and within 4 weeks after the training. There was a post-training improvement in PKQ-R scores in 42 of the 43 questions (97.7%), 22 of which were statistically significant differences (p < 0.001). The MAVAS-R-ID was completed at baseline and within 4 weeks of training by 125 trainees. The MAVAS-R-ID total score showed statistically significant post-training improvements (p < 0.01). Individual domain score analysis showed a statistically significant improvement in one of the five domains related to attitude regarding the use of medicine for behaviours that challenge.
Conclusions
The SPECTROM training seems to improve staff knowledge of psychotropic medicine, at least in the short-term, and attitude towards behaviours that challenge, particularly concerning the use of psychotropic medicine
Payors, Caregivers, and Teens: Aligning Priorities for Effective Digital Mental Health Tools
Get PDFDigital mental health (DMH) tools are a promising solution to the growing need for mental health care among teens because of their scalability and accessibility. Yet the key stakeholders-payors (who reimburse), caregivers (who facilitate adoption and access), and teens aged 13-17 (the primary users of these tools)-often have conflicting priorities. These misaligned priorities can limit investment in tools that align with user needs and promote long-term engagement, ultimately hindering the effective deployment of DMH tools for teens. This viewpoint paper outlines payor, caregiver, and teen priorities for DMH tools and proposes novel strategies to align these priorities. We argue that this alignment is critical for driving meaningful improvements in teen well-being
The Animals
No full textThis Practice as Research PhD investigates the use of the animal in magical realist fiction, with a particular focus on representations of lesbian identity and queer relationships. It situates magical realism within the contemporary landscape, and considers how animals function both as symbols/metaphors, and as characters with agency. Alongside this, the thesis explores intimacy across human-human and human-animal relations, including the spectre of bestiality and the possibility of identity transition. These concerns inform the creative component, a novel entitled The Animals.
The thesis addresses four key questions:
1. How have animals been deployed in contemporary magical realist literature, and what functions have they served?
2. How might animals operate simultaneously as metaphors of identity and suppression, and as characters in their own right?
3. How can animals illuminate the complexities of intimacy, desire, and power in lesbian and queer dynamics, within magical realism works?
4. How might human-animal crossings - intimacies, bestiality, and transitions of identity - interrogate the boundaries between species, selves, and communities?
The Animals is a magical realist novel concerned with identity, sexuality, and suppression. Structured in four sections, it moves backwards in time through the life of its protagonist, Alba. Each section is marked by a distinctive animal presence - horses, tigers, snakes - converging in the final part. Framed as a letter to her former partner Elise, the narrative combines confession and retrospective commentary in a futile attempt to 'set the record straight.'
Across the novel, animals express both marginalisation and resistance, exposing suppressed aspects of Alba's identity and the conflicts most sharply felt in her relationship with Elise.
The thesis demonstrates how magical realism can interrogate the animal as both metaphor and material presence, extending the mode by situating such explorations within a lesbian context rarely addressed in the field
Interventions to facilitate access to long-term care for community-dwelling older adults: A scoping review
Get PDFAccessing long-term care can be challenging due to complex care systems. This study aimed to identify evidence on interventions designed to improve access to long-term care for community-dwelling older adults. A systematic search of five databases identified 16 studies published between 2013 and 2023. Interventions were categorized into four types: early identification (n = 6), integrated care (n = 5), partnership (n = 2), and other (n = 3). Common strategies included providing information, making referrals, and providing coordinated care. Most interventions were delivered by healthcare professionals, targeted high-risk older adults, and focused on nursing home admissions, with less attention to other long-term care services. Interventions could target older adults with advanced age or specific health conditions and identify unmet needs at early stages. Future research could document intervention length, explore its impact on service access, and evaluate the feasibility and effectiveness of involving non-healthcare professionals in delivering interventions
Access Repository Dashboard
Do you manage Kent Academic Repository? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!