53417 research outputs found
Sort by
Transforming jet flavour tagging at ATLAS
No full textJet flavour tagging enables the identification of jets originating from heavy-flavour quarks in proton–proton collisions at the Large Hadron Collider, playing a critical role in its physics programmes. This paper presents GN2, a transformer-based flavour tagging algorithm deployed by the ATLAS Collaboration that represents a different methodology compared to previous approaches. Designed to classify jets based on the flavour of their constituent particles, GN2 processes low-level tracking information in an end-to-end architecture and incorporates physics-informed auxiliary training objectives to enhance both interpretability and performance. Its performance is validated in both simulation and collision data. The measured c-jet (light-jet) rejection in data is improved by a factor of 3.5 (1.8) for a 70% b-jet tagging efficiency, compared to the previous algorithm. GN2 provides substantial benefits for physics analyses involving heavy-flavour jets, such as measurements of Higgs boson pair production and the couplings of bottom and charm quarks to the Higgs boson, and demonstrates the impact of advanced machine learning methods in experimental particle physics
Generative AI for Human-Aware Scene Reconstruction and Multi-Modal Positioning
Get PDFThis study proposes a diffusion-based generative model capable of synthesizing realistic 3D indoor scenes and human motion trajectories from textual descriptions. By leveraging Variational Diffusion Transformers (VDiT), the model effectively captures spatial relationships and semantic context while generating coherent object layouts and dynamic human movements. The generated outputs provide significant potential for creating synthetic datasets to support Indoor positioning systems (IPS) development—for instance, converting trajectories into synthetic Inertial measurement unit (IMU) data to train IMU-based systems or extracting RGB and depth images from generated scenes to enrich datasets for vision based IPS. Additionally, the research advances IMU-based positioning by developing a robust pedestrian dead reckoning (PDR) method that mitigates cumulative errors through continuous motion analysis and Vision Transformer (ViT)-based heading estimation across varying time scales. Results indicate that the model achieves 83% accuracy at optimal time scales while maintaining robustness across different environments. Notably, shorter time scales improve fault tolerance in turning sections, and increasing the positioning update frequency to 10Hz significantly enhances real-time IPS performance. For vision-based IPS, the study introduces a transformer-based multi modal fusion framework that combines RGB and depth data to improve absolute pose regression accuracy. The proposed MCAPR model integrates cross-attention mechanisms to enhance feature extraction and scene representation, achieving 7.03° orientation error and 0.16m translation error, outperforming state-of-the-art single-scene and multi-scene models on the 7-Scenes dataset. By addressing challenges like lighting variations and occlusions, this framework demonstrates up to 20% improvement in localization accuracy compared to the baseline method MSPN. The key contributions of this research include the development of III innovative generative modeling techniques, enhanced IMU and vision based positioning algorithms, and the exploration of multi-modal data fusion for IPS. These advancements lay the foundation for creating more accurate, robust, and scalable indoor positioning systems
Essays on methods to cast nonlinear unobserved components
No full textThis dissertation consists of three self-contained essays that develop and apply novel econometric methods to uncover nonlinear unobserved components in macroeconomic time series. Across these chapters, the analysis explores how incorporating flexible nonlinear structures—such as time-varying volatility, quantile heterogeneity, and nonparametric func- tion approximation—can enhance the understanding of latent macroeconomic dynamics and the transmission of structural shocks. While the chapters cover distinct applications, they are unified by a shared methodological goal: to extend existing frameworks for modeling unobserved components in the presence of asymmetries, nonlinearities, and high-dimensional information sets
Barriers to domestic violence disclosure in healthcare settings: a scoping review of victim and provider perspectives.
No full textBACKGROUND: Domestic violence (DV) is a global public health issue with far-reaching physical, psychological, and social consequences. Although prior reviews have identified barriers to DV disclosure in healthcare settings, these have predominantly focused on female victims in Western contexts. This scoping review builds on the work of Heron and Eisma (2021) by including male victims, along with females, and studies from Asian countries, along with western countries, offering a more inclusive and culturally diverse understanding of disclosure barriers. METHODS: A thorough search of four databases: PubMed, Scopus, Embase, and the Cochrane Database, was conducted to identify relevant studies published between January 2018 and April 2023. Studies were included if they examined barriers to DV disclosure in healthcare settings from the perspectives of either victims or healthcare professionals (HCPs). Title and abstract screening, full-text review, and data extraction were performed independently by two reviewers. A thematic analysis was conducted to synthesise victim- and HCP-related barriers. RESULTS: Fifteen studies met the inclusion criteria. Victim-reported barriers included fear of retaliation, social stigma, low self-esteem, mental health challenges, and lack of privacy during healthcare encounters. Male victims highlighted societal disbelief and stigma around male victimhood. In Asian countries, cultural norms around family honour and obedience were particularly influential in discouraging disclosure. Practical barriers, such as the presence of abusers and limited access to services, were common in both high-income and low- and middle-income settings. HCP-reported barriers included inadequate training, absence of standardised protocols, time constraints, and a lack of culturally sensitive tools. CONCLUSION: This review identifies complex, context-specific barriers to DV disclosure, especially for male victims and individuals in non-Western healthcare systems. Addressing these barriers requires gender-sensitive training, culturally appropriate interventions, and systemic improvements to healthcare delivery. These findings call for inclusive, evidence-based strategies to support disclosure and improve care for all DV survivors in healthcare settings
The structural organisation of pentraxin-3 and its interactions with heavy chains of inter-α-inhibitor regulate crosslinking of the hyaluronan matrix
No full textProtocol for a biomarker discovery study to identify correlates of risk for future tuberculosis disease progression in South African children (INTREPID).
Get PDFINTRODUCTION: Young children and children living with HIV are at high risk of progressing to tuberculosis (TB) disease following Mycobacterium tuberculosis (Mtb) exposure and infection, and also of developing severe forms of disease and TB-related mortality. Identifying children who have very early (sub-clinical) TB disease, prior to progression to clinically apparent TB, would mean that TB preventive treatment (TPT) could be more efficiently targeted to this group. Identifying biomarker changes on drug therapy in children with Mtb infection or very early disease could pave the way for the development of tests that can identify which children have viable bacilli and are therefore at increased risk of disease progression. METHODS AND ANALYSIS: The INTREPID study will use already collected samples taken from well-phenotyped paediatric cohorts in three clinical studies conducted in South Africa in children <5 years, including a drug-resistant TPT trial (TB-CHAMP), an observational household contact study (interferon-gamma release assay studies) and a prospective diagnostic study (Umoya), all conducted in a setting with a high burden of TB and HIV. We will employ transcriptomic, proteomic, metabolomic and serology approaches to analyse changes in host blood profiles at every stage along the TB continuum, from Mtb exposure to disease and from children treated for Mtb infection and early TB disease, as well as targeted Mtb antibody analysis. Data on viral co-infections and relevant clinical and epidemiological parameters will be integrated and evaluated to identify the optimal biosignatures that can predict future progression to clinically overt disease in children below 5 years of age, including those living with HIV. ETHICS AND DISSEMINATION: The study protocol received ethical approval from the Stellenbosch University Health Research Ethics Committee (N23/03/025). The study findings will be disseminated through peer-reviewed publications, scientific conferences and formal presentations to healthcare professionals and to local communities, in collaboration with the Desmond Tutu TB Centre Community Advisory Board
Laboratory-based additive modifications in glass ionomer cements: A scoping review using a systematic data mining and trend analysis framework (2015-2024).
No full textOBJECTIVES: This scoping review aimed at identifying laboratory-based additive modifications in commercial glass ionomer cements (GICs), analyzing trends between additives and properties improved, and highlighting research gaps relevant to clinical performance using an expert-guided informatics framework (EGIF), a human-in-the-loop approach grounded in materials informatics (MI). DATA/SOURCES: Reporting followed the PRISMA Extension for Scoping Reviews. PubMed and ScienceDirect were screened from 01 January 2015 to 31 December 2024. Additional relevant articles were identified through reference mining. STUDY SELECTION/RESULTS: A total of 1,222 articles were screened, and 79 unique records selected and further analyzed. Additives to GICs were categorized into nine classes according to MeSH classification. Based on the final selected articles, trends were analyzed regarding the GIC base materials used in experiments, the types of additives incorporated, and the properties improved. Research on GIC modification has been steadily increasing, with Fuji™ Ⅸ GP and Fuji™ Ⅱ LC being the most frequently used base materials. Recent studies have primarily focused on improving physicochemical properties while maintaining or even enhancing bioactivity. CONCLUSIONS: EGIF based on these data suggests that the incorporation of nano-sized metals or bioactive glass or glass fiber may compensate for the limitations of commercially available GICs and enhance their bioactivity. Furthermore, AI-assisted approaches are expected to facilitate the prediction of the optimal additive and their ratios. CLINICAL SIGNIFICANCE: This review provides an evidence-based map of additive strategies for GIC modification. As it synthesizes only in vitro findings, the results may guide rational development of next-generation GICs but do not directly reflect clinical outcomes
Corrigendum to “Computational and experimental investigation of an aerosol extraction device for use in dentistry” (Journal of Aerosol Science, (2025), 183, C, (106478), (S0021850224001459), 10.1016/j.jaerosci.2024.106478)
No full textThe authors regret leaving out the MRC grant code MC_PC_19031 from the Acknowledgements. The authors would like to apologise for any inconvenience caused
Safeguarding Pastiche in the Age of Algorithmic Constitutionalism: Reconstructing User Rights through CMO Accountability and ECL Governance
No full textThis thesis examines the structural failures of the European Union’s current framework for governing online copyright, focusing on the inability of Article 17 of the CDSM Directive to safeguard transformative user expression within platform dominated environments. Although Article 17 seeks to rebalance the interests of rightsholders, platforms, and users, platform driven private copyright governance, shaped by automated filtering systems and confidential licensing agreements, continues to determine what content is accessible and monetizable online. The consequences are particularly acute for pastiche, a lawful transformative use recognised under EU copyright law but rendered practically unworkable by doctrinal ambiguity and algorithmic detection tools that default to treating such content as infringing. The thesis argues that the core deficiency lies not in Article 17’s legal design, but in the absence of institutional mechanisms capable of enforcing its safeguards against the operational power of large platforms. It proposes that Collective Management Organisations, strengthened through Extended Collective Licensing and enhanced obligations of transparency, representativeness, and non-discrimination, can serve as constitutional intermediaries between rightsholders, platforms, and users. Properly reformed CMOs offer procedural oversight and accountability capable of converting Article 17’s theoretical protections into meaningful guarantees for pastiche and freedom of expression. The thesis therefore reconceptualises CMOs as rights protective institutions that are essential to restoring public oversight in an increasingly privatised digital copyright environmen