Access to Research at National University of Ireland, Galway
Research at National University of Ireland, GalwayNot a member yet
15712 research outputs found
Sort by
Balanced-simplified spatiotemporal memory attention for image captioning
No full textVisual attention and memory-based attention methods have been effectively utilized in image captioning to focus on the most relevant areas of an image during the language generation process. Nevertheless, they face significant challenges, as they are solely guided by the hidden state of the LSTM, leading to attention focused on less relevant areas at various time steps. Furthermore, many approaches apply a uniform focus across all visual information within an image, lacking a mechanism for adjustment of focus intensity. Additionally, the complexity of memory-based attention methods highlights the necessity of developing a simplified memory-based attention mechanism for more efficient and effective image captioning. To address these challenges, a novel attention method for image captioning, named BalancedSimplified Spatiotemporal Memory Attention (BS-STMA), is proposed. The proposed attention mechanism captures spatiotemporal relationships by combining the advantages of LSTM and visual attention in a simple and effective manner. The combination of LSTM memory and the attention mechanism significantly enhances the model’s capacity to retain, convey, and utilize relevant visual information throughout the captioning process. Additionally, an Intensity Balancing Controller (IBC) is introduced and integrated into BS-STMA to enhance its efficiency. IBC allows for adjustments of attention intensity, enabling the model to capture visual information more accurately over time. Extensive experiments on the MSCOCO dataset demonstrate that the method significantly improves image captioning performance, surpassing recent approaches in various evaluation metrics by effectively capturing spatiotemporal.This work is supported by Anhui Provincial Higher Education Institutions Scientific Research Project, Project approval No.: 2023AH050496. This research was conducted with the financial support of SFI under Grant Agreement No SFI/12/RC/2289_P2.peer-reviewe
ChildDiffusion: Unlocking the potential of Generative AI and controllable augmentations for child facial data using stable diffusion and large language models
No full textEnsuring the availability of child facial datasets is essential for advancing AI applications, yet legal, ethical, and data scarcity concerns pose significant challenges. Current generative models such as StyleGAN excel at producing synthetic facial data but struggle with temporal consistency, control over output attributes, and diversity in rendered features. These limitations underscore the need for a more robust and adaptable framework. In this research, we propose the ChildDiffusion framework, designed to generate photorealistic child facial data using diffusion models. The framework integrates intelligent augmentations via short text prompts, employs various image samplers, and leverages ControlNet for enhanced model conditioning. Additionally, we have used large language models (LLMs) to provide complex textual guidance to enable precise image-to-image transformations, facilitating the curation of diverse, high-quality datasets. The model was validated by generating child faces with varied ethnicities, facial expressions, poses, lighting conditions, eye-blinking effects, accessories, hair colors, and multi-subject compositions. To exemplify its potential, we open-sourced a dataset of 2.5k child facial samples across five ethnic classes, which underwent rigorous qualitative and quantitative evaluations. Further, we fine-tuned a Vision Transformer model to classify child ethnicity as a downstream task, demonstrating the framework’s utility. This research advances generative AI by addressing data scarcity and ethical challenges, showcasing how diffusion models can produce realistic child facial data while ensuring compliance with privacy standards. The versatile ChildDiffusion framework offers broad potential for machine learning applications, serving as a valuable tool for AI innovation. The project website, along with the complete ChildRace dataset and the fine-tuned model, is available at (https://mali-farooq.github.io/childdiffusion/).Taighde Eireann–Research Ireland (Grant Number: IRCLA/2023/1992 and EPSPG/2020/40)peer-reviewe
High-latitude cabbeling observations along the east Greenland polar front
No full textCabbeling is the process where water parcels of the same density but different temperatures/salinities combine to form a new parcel of higher density. This can result in a statically stable profile becoming unstable after mixing has occurred. High-latitude cold, fresh, and shallow ocean waters exhibit greater nonlinearity in density dependence on temperature and are prone to cabbeling along fronts. While modeling shows there are important implications of high-latitude cabbeling, harsh polar conditions and the evanescent nature of cabbeling events make direct observations of the small-scale and shifting vertical structure difficult and rare. The East Greenland Polar Front (EGPF), where mixing of cold-fresh Arctic water and warmer-saltier Atlantic water occurs, is a location which has a high potential for cabbeling. Cabbeling-induced density anomalies of up to 0.05 kg m−3 within the upper 40 m were observed along the EGPF in 2023. Shallow stratification shows staircase structures within a strong halocline overlaying warm water intrusions, displaying a characteristic “jagged” shape in TS space. Enhanced turbulence was identified in regions where cabbeling instabilities occurred. These observations offer new insight into the vertical and temporal structure of cabbeling in high-latitude environments via rapidly repeated profiling. The observed anomalies align in magnitude and character with previously documented studies, extended here to include shallower observations coupled with shear measurements within frontal zones as identified by sea surface temperature. These findings suggest that cabbeling-induced mixing moderates locally enhanced heat flux, with implications for ice-ocean interaction and the broader high-latitude climate system.The author acknowledges the support from the Irish Center for Research in Applied Geosciences iCRAG, part of the Research Ireland Centre for conducting this research funded by a research Grant from Research Ireland under Grant 13/RC/2092_2. Additional funding was provided from the European Union's Horizon 2020 Research and Innovation Programme under Grant agreement 821001 awarded to BW. AM acknowledges Grant in Aid funding from the Marine Institute for research expedition CE23011 on the RV Celtic Explorer with special thanks to the captain and crew. In addition, AM acknowledges the financial support of Research Ireland and the Geological Survey of Ireland under the Frontiers for the Future Programme 21/FFP-P/10261.peer-reviewe
Direct Flux via Virtual Faces (DFVF-overset): Interpolation-free, conservative, overset CFD using a generalised finite volume method
No full textWe present DFVF-overset (Direct Flux via Virtual Faces), a conservative overset scheme based on a general form of the finite volume method, originally derived for a meshless method, which intrinsically supports overlapping cells. Fluxes pass between overlapping cells through virtual faces which have rigorously defined area. Exact conservation is retained, and the method does not require interpolation between constituent grids. The new technique has been implemented as a preprocessor for the open-source CFD library OpenFOAM, and validated for a number of 1D and 2D cases. In a 1D diffusion case, the method converges to an analytical solution in the second order. For the lid-driven cavity, DFVF-overset results are close to single-grid solutions and display similar convergence towards a benchmark solution. The new method produces smooth velocity fields, and on a relatively coarse grid, it resolves a tertiary vortex which is absent in interpolation-based overset solutions. In static and dynamic multiphase cases solved with a volume-of-fluid method, conventional overset schemes display loss of liquid mass, whereas DFVF-overset demonstrates strict conservation of mass and close agreement with single-grid solutions. The new technique shows promise for applications where conventional overset is unsuitable due to interpolation errors or lack of conservation.The research conducted in this publication was funded by the Irish Research Council under award number
GOIPG/2021/516
Evaluation of the potential for genomic selection in an Irish breeding population of sitka spruce (Picea sitchensis (Bong.) Carr))
No full textGenomic selection is a form of breeding predicting for phenotypic traits within a population, using allelic data drawn from individuals within this population as the explanatory variable. Genomic selection has been widely evaluated within livestock and crop species, and is now an established part of breeding endeavours within such species. In the era of increasingly cheaper whole-genomic sequencing strategies, the potential for genomic selection is increasingly being investigated in more niche species with features that otherwise discourage evaluation such as long generational intervals or particularly large genomes. Sitka spruce (Picea sitchensis) is one such species, yet equally is also one of great importance to Irish forestry. Efforts to breed Sitka spruce in Ireland have been have resulted in a national breeding program, but have not yet attempted to incorporate any of the recent advances regarding genetic and genomic breeding. In this thesis, I investigated the the potential for genomic selection among the Irish breeding population of Sitka spruce. I evaluated the suitability of phenotypic data generated from the breeding program for the used as traits of interest within a genomic selection model. Comparing the historical data of the program to contemporary data, I investigated the assumptions of correlation of the genetic control of a traits across different ages which underpins breeding in many long-lived species. Combining the available phenotypic data of the breeding program, derived from families of half-siblings, with genomic data generated from their maternal parent through Genotype-by-Sequencing, I explored the potential for genomic selection in Irish Sitka spruce with respect to the nature of these data. Finally, using a sub-population of full-siblings derived from the breeding program's genotypes, I examined the effectiveness of genomic selection against more conventional breeding methods such pedigree-based breeding. In totality, this thesis demonstrates a clear potential for viable genomic selection in the Irish Sitka spruce breeding population as is, and highlights opportunities for changes to current practices which would further improve genomic selection within this population
An effective graph-based diffusion method for top-n recommendation
No full textThe primary interest of this thesis is to design a graph-based recommendation approach to improve the recommendation result. The traditional recommendation approaches did not address data sparsity and insufficient information utilization issues. Consequently, we are motivated to build a user-item combination graph and apply the graph traversals on that graph to address these problems. Firstly, we use probabilistic graph traversals to solve the data sparsity problem by exploring the indirect relationships between users-to-users and items-to-items. Besides, we investigate graph kernels to effectively measure the similarity between a pair of graph nodes and combine the diffusion kernel with the graph-based recommendation approach. Then, to solve the insufficient information utilization problem, we aim to explore the item’s semantic information from knowledge graphs using deterministic graph traversals. We build the semantic inter-item graph and combine it with the graph-based recommendation approach to improve the recommendation result further. Finally, we experiment with our proposed methods on publicly well-known datasets to investigate the recommendation performance.Insight Centre for Data Analytics, School of Computer Science, University of Galwa
The representation of the refugee experience in Jane Mitchell’s Run For Your Life and ‘There and Here’
No full textJane Mitchell’s novel, Run For Your Life (2022), and short story, There and Here (2015), document the lives of two young female protagonists as they navigate and endure life in Direct Provision in Ireland. Mitchell’s texts encourage young readers to have empathy for the plight of international protection applicants in Ireland by detailing the systemic problems with the deeply flawed Direct Provision system. My discussion will examine Mitchell’s texts for child and young adult (YA) readers in the context of the Direct Provision system and the rise of far-right, anti-migrant and anti-refugee sentiment present within contemporary Irish society. I draw upon Ekaterina Strekalova-Hughes’ RefugeeCrit framework and Julia Hope’s development of this framework to examine Mitchell’s representation of the refugee experience in Direct Provision and contemporary Irish society, focusing particularly on the details omitted from the texts regarding the protagonists’ nationalities, ethnicities, cultural backgrounds, and reasons for their displacement. I argue that to fully evoke empathy in the reader; to avoid a hierarchisation of nationalities, ethnicities, and cultures; and to prevent a homogenisation of refugee experiences, it is vital that within narratives of displacement for children, characters’ nationalities and their cultural backgrounds are specified and the reasons for their displacement provided.peer-reviewe
An nuachainteoir ina (h)aisteoir cruthaitheach i gceantair Ghaeltachta Chorca Dhuibhne agus Uíbh Ráthaigh
No full textFeiniméan fadaitheanta é aos dána agus cruthaitheach a bheith ag tarraingt ar Chorca Dhuibhne, scríbhneoirí na Gaeilge, go speisialta. Ba ghnách le lucht, Innti imeacht siar go Corca Dhuibhne, áit a soláthraíodh ceárta anama agus teanga dóibh (Ó Dúshláine 2011, Ní Ghairbhí, 2011). Bhí nós ag scríbhneoirí tréimhse sealbhaithe teanga a chaitheamh sa Ghaeltacht ó thús na 20ú haoise ar aghaidh, a luaitear mar chineál ‘turasóireacht chultúrtha’ (Ní Ghairbhí, 2011: 70). Cothaíodh na mórfhilí Ní Dhomhnaill, Ó Ríordáin, Davitt agus Ó Muirthile go speisialta ag an bpobal ‘neamhphollta teanga’ tríd na ‘turasanna fionnachtana’ a thugadar siar go Corca Dhuibhne (Ní Ghairbhí, 2011: 70). Deir Nic Eoin (2005: 270) ‘gur freastalaíodh ar riachtanais iomlána an duine chomh maith lena riachtanais teanga’.peer-reviewe
Conveying identity through place: Understanding Bronze Age people via nucleated settlements
No full textThe Irish Bronze Age covers nearly two millennia, from 2200–750 BC. During this period, changing cultural customs, such as developments in domestic architecture, suggest an increase in social stratification. Two settlement types, the roundhouse farmstead, and the hillfort, have provided primary evidence for developing Bronze Age theories of society, which include heterarchical and hierarchal interpretations.
Recent evidence for a third settlement type, the nucleated settlement, suggests a more complex societal narrative than previously thought. Consisting of 20 to upwards of 200 agglomerated roundhouses, the nucleated settlements have an average area of 12 hectares. They may loosely be thought of as ‘villages’ which is a radically different concept to the more familiar concepts of dispersed farmsteads and hillforts.
Evidence for at least nine nucleated settlements has been found across Ireland in a range of landscapes, from coastal areas to hilltops, but they have not been considered as a coherent group. As such, this project focuses on compiling comprehensive data for the sites via fieldwork, desk-top research, and GIS (Geographic Information System) landscape analysis. A comparative analysis of the data will then help to better understand how the nucleated settlements relate to each other and to their landscapes. Interpretive results will help to establish the role of nucleated settlements within Irish Bronze Age society, with a focus on sense of place and impact on identity. Furthermore, the results will allow connections to be made with similar settlement types found in the wider European Bronze Age
PPFL-DCS: Privacy-preserving federated learning using neural transformer and leveraging dynamic client selection to accommodate data diversity
No full textThe vulnerabilities and security issues of industrial Cyber-Physical Systems (CPSs), such as Intrusion Detection Systems (IDSs), have significantly increased due to the rapid integration of conventional industrial setups with advanced networking and computing technologies like 5G, software-defined networking, and artificial intelligence. Coping strategies for such challenges frequently involve transferring data to a central location, which raises concerns about latency, efficiency, and privacy. To address these issues, Federated Learning (FL) was developed as a solution to mitigate both the privacy concerns of organizations and the complexities of networked systems. However, FL-based techniques still have shortcomings, FedAvg equally weights weak models, risking suboptimal results; FL also faces Membership Inference privacy attacks. To address these challenges, we propose PPFL-DCS, an FL framework that incorporates a weighted mechanism for dynamic client selection, accounting for the performance of each local model and data size of each client in integration with a Neural Transformer System (NTS) that enhances the system‘s robustness against the MIA attacks. The NTS limits the impact and gains of attackers, thereby reducing the effectiveness of MIAs. Extensive experiments demonstrate that PPFL-DCS achieves a high detection accuracy of 97.424% for cyber threats in industrial CPSs, and highlight its efficiency over state-of-the-art techniques.peer-reviewe