224092 research outputs found
Sort by
Improving robustness to domain shift in machine learning
No full textMachine learning models often underperform when the test data characteristics differ from the training data, a phenomenon known as domain shift. Improving robustness to domain shift has been a longstanding goal in machine learning, and is crucial to the widespread deployment of AI. This thesis addresses four underexplored but important aspects of this field: imbalanced domain adaptation, dataset filtering, model selection, and variance reduction of domain alignment losses. To this end, novel algorithms, perspectives, methodologies, and theoretical results are introduced, resulting in improved out-of-domain performance on these tasks. Particular emphasis is placed on developing methods that are both theoretically grounded and practically useful, and understanding their assumptions and limitations.A central motivating application for this work is the automated detection and classification of marine mammal vocalisations, where domain shift is especially prevalent. This thesis serves to underscore the importance of adopting robust training and evaluation practices in this context. To support progress in this area, a novel domain shift benchmark based on humpback whale detection is also introduced.Overall, this thesis contributes to advancing the reliability and trustworthiness of machine learning models, at a time when AI systems are increasingly being deployed to dynamic, uncertain, and open-ended settings
Impacts of salmonid fish production on river water quality: a critical appraisal of the evidence
No full textRivers are recognised as providing valuable goods and services to society, and as hotspots of global biodiversity. Fish farming is an economically important activity associated with river systems; however, the effects of this industry on river health are not entirely understood. Studies evaluating interactions between freshwater fish farms and rivers are limited; therefore, we conducted a systematic review and an appraisal of evidence quality to assess the impacts of salmonid fish farming on river water quality. The review revealed that two main types of water quality indicators are generally applied: physicochemical and biological. The response variables used to determine the effects of fish farm effluent on water quality vary considerably, yet most studies indicate that aquaculture negatively affects the environment. Overall, we highlight the knowledge gaps, including a lack of studies in freshwater aquaculture hotspots. The effects of fish farming on bacterial communities are also not clear due to limited research on these bioindicators. We suggest improvements in study design, including thorough consideration of confounding factors, and the provision of background information on the fish farms
Safety, feasibility and efficacy of exercise as an airway clearance technique in cystic fibrosis - a randomised pilot feasibility trial
No full textObjectives: To test the feasibility and safety of exercise as an airway clearance technique (ExACT) for people with cystic fibrosis (pwCF) versus usual care (UC). Methods: Dual-site, two-arm randomised pilot trial. Fifty pwCF (≥10 years, forced expiratory volume in 1 s (FEV 1) ≥40% predicted), stable on Elexacaftor/ Tezacaftor/Ivacaftor, were recruited, of whom 48 were randomly assigned (1:1 with minimisation) to daily ExACT (stopping all other airway clearance techniques) or UC. Feasibility was measured by recruitment, retention and adherence against preset progression criteria. Key measures of safety and signals of efficacy included spirometry (FEV 1), lung clearance index (LCI 2.5), pulmonary exacerbations, physical activity, treatment burden and quality of life across 28 days. Qualitative interview data and preliminary health economic data were also collected. Findings: ExACT was safe over 28 days, measured by change in LCI 2.5 (ExACT −0.1 (0.6) vs UC 0.2 (0.8), mean (SD)) and FEV 1 (ExACT +2.1 (6.6) vs UC −0.8 (5.5), % predicted mean (SD)). Relative (ExACT/UC) differences of 0.97 (0.92, 1.02) for LCI 2.5 and absolute differences (ExACT-UC) of 3.2 (−0.6, 6.9) % predicted for FEV 1 suggest potential intervention efficacy. Few adverse events were reported; none serious. Recruitment and retention data suggest progression to a definitive trial, with 48/117 (41% of approached) randomised, 45/48 (92%) completing the study and a 60% overall adherence rate. Discussion: Testing of our primary hypothesis within a feasibility trial showed ExACT to be a safe, acceptable and feasible intervention for pwCF. These data support advancement to a definitive, longer-term, multisite trial evaluating the safety, efficacy and cost-effectiveness of ExACT, following minor refinement.</p
Adrien Allorant’s contribution to the discussion of ‘Some statistical aspects of the Covid-19 response’ by Wood et al
No full textAPOLLO: an open platform for LLM-based multi-agent interaction research
No full textTraditional decision-making processes often struggle to capture diverse stakeholder perspectives and anticipate potential outcomes. Complex decisions and persuasions might rely on insights and perspectives which might not be available. In this paper, we leverage recent advances in large language models and retrieval-augmented generation to introduce APOLLO—an Architecture and oPen-source system that Orchestrates Large Language mOdels. APOLLO coordinates multiple LLMs by engaging them in collaborative discourse to reach a consensus on user-defined prompts. This system enables HCI and AI researchers and practitioners, and allows them to explore and experiment with LLM-based multi-agents systems in a user-configurable and customisable manner. By providing this flexible platform, APOLLO enables new avenues for studying and designing human-AI interactions, investigating the impact of multi-agent interaction on human behaviour, and ultimately facilitates a deeper understanding of how AI-driven collaboration can enhance human-AI interaction and decision making.</p
Cross-Cultural Consortium on Irritability (C3I): An International Network for Research on Cultural Similarities and Differences in Irritability
No full textObjectiveIrritability is among the top reasons for youth mental health referrals worldwide. Cultural factors may affect how irritability manifests and develops; how it is experienced by youth and responded to by their caregivers; and how it is treated. However, the influences of cultural context on irritability have received little systematic investigation.MethodThe Cross-Cultural Consortium on Irritability (C3I; https://m.yale.edu/c3i) is an international research network created to increase the limited evidence base on cross-cultural similarities and differences in irritability. By bringing together researchers worldwide, C3I provides an innovative and collaborative approach to address unmet needs and explore novel research questions regarding cultural variation in irritability. Additionally, combining resources and data across the globe helps produce robust, reproducible, and generalizable results using large mega-data. One important initiative involves pooling existing datasets to support manuscript collaborations. The first three such projects focus on cross-cultural comparisons of the following irritability-related topics: boundaries of normative behavior; association with suicidality and self-harm; and informant effects. Another ongoing effort involves conceptualization of irritability across cultures. Other efforts include promoting projects of primary data collection using qualitative and quantitative methods, harmonization across measures, and facilitating/supporting community-based participatory research and engagement.DiscussionC3I is an innovative, collaborative research structure to build a robust, reproducible, and generalizable evidence base on irritability and its characteristics, including socio-cultural influences. This evidence base will facilitate recognition and assessment of irritability and, ultimately, inform development of effective, culturally informed prevention and intervention to benefit the largest possible number of youth and their families
Chiral time crystal - modelling the matter-to-life transition
No full textNonreciprocal molecular interactions are now widely discussed as a mechanism of the matter-to-life transition that is accompanied by emergence of structural chirality. Here we demonstrate an achiral photonic nano-mechanical metamaterial that transitions to a chiral oscillating time crystal state, under illumination with circularly polarized light
Revisiting cross-domain problem for LiDAR-based 3D object detection
No full textDeep learning models such as convolutional neural networks and transformers have been widely applied to solve 3D object detection problems in the domain of autonomous driving. While existing models have achieved outstanding performance on most open benchmarks, the generalization ability of these deep networks is still in doubt. To adapt models to other domains including different cities, countries, and weather, retraining with the target domain data is currently necessary, which hinders the wide application of autonomous driving. In this paper, we deeply analyze the cross-domain performance of the state-of-the-art models. We observe that most models will overfit the training domains and it is challenging to adapt them to other domains directly. Existing domain adaptation methods for 3D object detection problems are actually shifting the models’ knowledge domain instead of improving their generalization ability. We then propose additional evaluation metrics – the side-view and front-view AP – to better analyze the core issues of the methods’ heavy drops in accuracy levels. By using the proposed metrics and further evaluating the cross-domain performance in each dimension, we conclude that the overfitting problem happens more obviously on the front-view surface and the width dimension which usually faces the sensor and has more 3D points surrounding it. Meanwhile, our experiments indicate that the density of the point cloud data also significantly influences the models’ cross-domain performance.</p
Data sources:
Access Repository Dashboard
Do you manage e-Prints Soton? Access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard!