National University of Ireland, Maynooth
MURAL - Maynooth University Research Archive LibraryNot a member yet
18981 research outputs found
Sort by
Love Under Duress: How Burnout Mediates the Relationship Between Partner Stress and the Perception of Romantic Partner Support.
Get PDFEmployees face difficulties in the modern workplace that put a burden on both their professional and personal well-being. This study aimed to clarify the intricate interactions between romantic partner stress, burnout, and the support of a romantic partner in the healing process. The literature emphasizes the role of romantic partners as both resource givers and demand producers. This idea is based on the Conservation of Resources (COR) principle. We sampled full-time employees from various industries in committed long-term partnerships (N=277). Using partial least squares structural equation modeling (PLS-SEM) to conduct our research, we provide support to understand the complex dynamics of romantic partner support in reducing work-related stress and its effects on burnout. Our results highlight how vital it is to comprehend how the supporting and demanding roles of romantic partners interact to influence burnout. We present our findings, discuss managerial implications, and outline recommendations for future research
Sea level since the Last Glacial Maximum from the Atlantic coast of Africa
Get PDFConstraining sea level at the Last Glacial Maximum (LGM) is spatially restricted
to a few locations. Here, we reconstruct relative sea-level (RSL) changes along
the Atlantic coast of Africa for the last ~30 ka BP using 347 quality-controlled
sea-level datapoints. Data from the continental shelves of Guinea Conakry and
Cameroon indicate a progressive lowering of RSL during the LGM from
−99.4 ± 5.2 m to −104.0 ± 3.2 m between ~26.7 ka and ~19.1 ka BP. From ~15 ka to
~7.5 ka BP, RSL shows phases of major accelerations up to ~25 mm a−1 and a
significant RSL deceleration by ~8 ka BP. In the mid to late Holocene, data
indicate the emergence of a sea-level highstand, which varied in magnitude
(0.8 ± 0.8 m to 4.0 ± 2.4 m above present mean sea level) and timing (5.0 ± 1.0
to 1.7 ± 1.0 ka BP). We further identified misfits between glacial isostatic
adjustment models and the highstand, suggesting the interplay of different
ice-sheet meltwater contributions and hydro-isostatic processes along the
wide region of Atlantic Africa are not fully resolved
The use of physical optics and machine learning in modelling the QUBIC beam pattern.
Get PDFThe Q and U Bolometric Interferometer for Cosmology (QUBIC) is a ground-based telescope which will observe the cosmic microwave background (CMB) with the goal of detecting its extremely faint primordial B-mode polarization pattern as observational evidence for the theory of inflation in the early Universe.
QUBIC has a novel bolometric interferometer design that allows it to have precise control over instrument systematics and to remove astronomical foregrounds which would otherwise obscure the CMB polarization signal. Due to its interferometric design, QUBIC has a complex multi-peaked antenna beam pattern known as the synthesized beam which must be well known in order to correctly process the QUBIC data. We have modelled the QUBIC beam pattern at multiple frequencies and for detectors located at different positions in the focal plane using precise electromagnetic and physical optics simulations. For the complex setup of the QUBIC telescope these simulations are computationally expensive.
In this work we use machine learning to interpolate the beam at multiple frequencies and detector locations in order to reduce the amount of computation required to model it. For our purposes, two types of neural networks were used: a Multilayer Perceptron (MLP) model and a Long Short-Term Memory (LSTM) model. For the amount of data generated, only the LSTM model was successful in replicating the synthesized beam to a satisfactory degree. We used the LSTM machine learning (ML) model to generate the synthesized beam for all 248 detectors at 150 GHz and compared its use in our analysis pipeline to a fully simulated PO beam and an approximate analytical model. From this it can be seen that although the ML prediction did not replicate the PO synthesized beam perfectly, it was closer than the beam produced with the analytical formula. Finally, we show the effect of the differences in beam pattern prediction on the recovered B-mode spectrum of the CMB
Toward Future Reanalyses That Meet Evolving Needs in Science, Public Services, Policymaking, and Socioeconomic Activity
Get PDFThe Sixth WCRP International Conference on Reanalysis (ICR6)
What: Reanalysis producers, observation data providers, numerical modelers, and members
of the user community came together to discuss progress, challenges, and future
priorities with the aim of guiding the development and use of reanalysis data in
science, public services, policymaking, and social/economic activity.
When: 28 October–1 November 2024
Where: Tokyo, Japa
Leveraging EU non-discrimination law to make the cultural and creative sectors more inclusive of professionals with disabilities: socio-legal perspectives
Get PDFWorking in the cultural and creative sectors is often seen as an atypical exercise, which differs from mainstream practices in the labour market and operates outside the standard regulatory framework. In this context, the situation of cultural and creative professionals with disabilities and the applicable EU legislation can be overlooked, both in research and in practice. Following a socio-legal approach, this article associates desk-based legal research and empirical research, and considers how the participation of cultural and creative professionals with disabilities could be fostered within the current EU regulatory framework. It identifies gaps and potential in EU cultural policy, disability law and labour law, and, supported by the findings from a qualitative study with EU cultural stakeholders, it discusses the challenges that cultural and creative professionals experience, including those with disabilities. Contributing valuable insights into the participation of persons with disabilities in cultural life, the article argues that the Employment Equality Directive, a pillar of EU labour law and non-discrimination law, can play a key role in making those sectors more inclusive
Co-Creating Change: Seedbed Interventions as Catalysts for Equitable Urban Planning—The Case of Umeå
Get PDFThe ongoing urbanisation and densification at the intersection with increasing environmental and health crises demand a holistic, equitable, and inclusive approach to urban planning, which has also been highlighted in the EU Green Deal’s inclusive approach to sustainable urban planning aligned with the UN SDGs’ “Leave No One Behind.” This article introduces the seedbed intervention as a novel, community-driven, co-creative approach to Nature-based Solutions (NbS) that addresses gaps in equitable and inclusive urban planning frameworks. On the case of Umeå (Sweden), the article introduces the seedbed intervention approach and demonstrates how the approach facilitates the development of locally appropriate and sustainable NbS. The results show that the seedbed intervention approach improved the alignment between local needs and NbS design, connected diverse user groups, and catalysed curiosity, interest, and participation among citizens with the help of applying art-based methods. By demonstrating the practical application of a seedbed intervention, this research contributes to the development of scalable frameworks for more equitable and inclusive urban planning
Parallel Genetic Algorithms for Multi-Criteria and Diverse Path Routing on Large Transportation Networks
Get PDFFinding an optimal path between a source and target node in a graph or
network is a well-established problem in Computer Science. This problem has
been studied extensively, with a range of algorithms developed to accommodate diverse
objectives, domains, and real-world application. Classical approaches, such as Dijkstra’s
algorithm, are well-known for computing a single optimal path based on one objective, variable
or criteria. This is typically distance or time and algorithms like this are not designed
to handle multiple objectives or generate a usable set of diverse alternative routes. In the
case of transportation networks, as urban environments become more complex and
mobility expectations increase, the ability to consider multiple objectives and
offer diverse routing options on road networks has shifted from a theoretical
challenge to a practical necessity. A real-world example of this might be a driver
who wishes to travel between two points in a road network and is seeking a route which
minimises travel distance, maximises the number of EV charging stations available on the
route, and minimises the overall cost of the journey in terms of toll charges. Traditional
algorithms (such as Dijkstra’s) cannot solve scenarios where a path (or route)
has multiple objectives to optimise. While such methods are effective at returning
a single shortest path, many real-world applications require a set of optimal (or nearoptimal)
alternative paths. Algorithms such as the k-shortest paths algorithm improve
upon, and extend, earlier approaches by producing multiple paths of equal or near-equal
length between two points. Nevertheless, most algorithmic approaches remain limited
to single-objective optimisation and do not guarantee diversity among the
final set of generated paths.
To address these limitations, population-based methods such as Evolutionary Algorithms
(EAs) have emerged as promising alternatives, capable of producing
diverse and/or multi-objective solutions suited for the challenges outlined
above. In this thesis, we make several contributions to this rapidly growing research
field by introducing novel computational frameworks capable of generating diverse, multiobjective
routes or paths in large-scale real-world transportation networks. This thesis
introduces a novel framework for solving the K-Most Diverse Near-Shortest
Paths (KMDNSP) problem, which aims to find a set of k near-optimal paths that
maximise path diversity while having near-optimal overall route lengths. To achieve this
we have used a special class of EAs known as Parallel Genetic Algorithms (PGAs). The
proposed PGA, named the MultiPath Island-Based Genetic Algorithm (MIBGA),
serves to facilitate vehicle routing mechanisms for the KMDNSP problem. This effectively
balances the need for path diversity and optimality through innovative migration and
adaptive selection strategies. Empirical results presented confirm the efficacy of MIBGA
and demonstrate how its design supports the generation of diverse, near-shortest paths.
Building on the island structure of MIBGA and integrating it with a global parallelisation model, we present the Parallel Optimal-route Search (POS) as a multi-objective
routing framework. POS is designed to address the real-world routing problems in the context
of general public mobility needs by providing end users with a collection of optimal
paths optimised across multiple criteria. To achieve this, a novel fitness metric called
Positional Count (PC) is introduced to mitigate the bias caused by longer routes potentially
offering better values for certain objectives. PC assigns weights to features based
on their position and density along the route, shifting the focus from sheer quantity to the
contextual relevance and spatial distribution of features along the route.
To demonstrate the robustness of our methods across different network topologies and
scales, we perform evaluation on complex real-world road networks rather than the simpler
synthetic networks commonly used in much of the key literature in this research. The
smallest network tested is from Arizona, with 834 nodes and 1, 547 edges, while the largest
is Texas, featuring 10, 886 unique nodes and 24, 464 edges. Our framework for solving
the KMDNSP problem, the development of MIBGA and POS demonstrate
significant contributions to the area of path optimisation on real-world transportation
networks. A number of opportunities for future work are presented at the
end of the thesis and these address theoretical, practical, and application-oriented aspects
of the proposed methods and metrics in this work. All software code and network data
used in our research is available as open-source resources to enhance the opportunities for
reproducibility and reuse
Psychological Readiness is the Main Barrier to Return to Play After Revision Anterior Cruciate Ligament Reconstruction
Get PDFBackground:
Despite advances in modern surgical techniques, return-to-play (RTP) rates after revision anterior cruciate ligament reconstruction (R-ACLR) often fall short of patients’ expectations. There is growing awareness that a patient’s psychological recovery is as important as the functional recovery of their knee.
Purpose/Hypothesis:
The primary purpose of this study was to determine the RTP rate and identify the barriers to RTP after R-ACLR. Secondarily, we compared the progression of psychological readiness (using the Anterior Cruciate Ligament–Return to Sport after Injury [ACL-RSI] scale) throughout rehabilitation between those who achieved RTP and those who did not. Finally, we assessed if RTP could be predicted for patients aged <23 years and patients aged ≥23 years based on their ACL-RSI scores at different time points during rehabilitation.
Study Design:
Case-control study; Level of evidence, 3.
Methods:
This investigation included 301 patients who underwent R-ACLR at our institution. Preoperatively, patients completed a questionnaire detailing their demographic characteristics and target level of RTP. The ACL-RSI scale was also administered preoperatively and at 3 months, 6 months, and 9 months. At 2 years postoperatively, patients indicated whether or not they had returned to play. Those who did not return provided their reasons for not doing so.
Results:
The mean age was 25.4 ± 6.3 years, and 84.5% of patients were male. Although 95.1% of patients undergoing R-ACLR intended to return to play before surgery, only 63.4% achieved this goal at 2-year follow-up. The main barriers to RTP were a fear of reinjury (44%), a lack of confidence in performance (13%), and ongoing knee pain (11%). The mean preoperative ACL-RSI score was significantly higher in patients who returned to play than in those who did not (51.2 ± 23.4 vs 42.6 ± 23.3, respectively; P = .027), indicating greater psychological readiness to RTP. The mean ACL-RSI score was also significantly higher in those who achieved RTP at 3 months, 6 months, and 9 months. A preoperative ACL-RSI score of 42.9 points predicted RTP in patients aged <23 years, with a sensitivity of 76% and a specificity of 77% (area under the curve = 0.73). The ACL-RSI score was able to predict RTP at all time points, demonstrating the most accuracy preoperatively and at 6 months postoperatively. At 6 months, a cut-off score of 60.4 points predicted RTP in patients aged <23 years (sensitivity = 62%; specificity = 81%), and a cut-off score of 56.7 points predicted RTP in patients aged ≥23 years (sensitivity = 54%; specificity = 77%).
Conclusion:
Psychological readiness, especially fear of reinjury, was a significant barrier to RTP after R-ACLR. Patients with lower psychological readiness who were less likely to return to play could be detected using the ACL-RSI scale
Generative AI and Blockchain-Integrated Multi-Agent Framework for Resilient and Sustainable Fruit Cold-Chain Logistics
Get PDFThe cold-chain supply of perishable fruits continues to face challenges such as fuel wastage, fragmented stakeholder coordination, and limited real-time adaptability. Traditional solutions, based on static routing and centralized control, fall short in addressing the dynamic, distributed, and secure demands of modern food supply chains. This study presents a novel end-to-end architecture that integrates multi-agent reinforcement learning (MARL), blockchain technology, and generative artificial intelligence. The system features large language model (LLM)-mediated negotiation for inter-enterprise coordination, Pareto-based reward optimization balancing spoilage, energy consumption, delivery time, and climate and emission impact. Smart contracts and Non-Fungible Token (NFT)-based traceability are deployed over a private Ethereum blockchain to ensure compliance, trust, and decentralized governance. Modular agents—trained using centralized training with decentralized execution (CTDE)—handle routing, temperature regulation, spoilage prediction, inventory, and delivery scheduling. Generative AI simulates demand variability and disruption scenarios to strengthen resilient infrastructure. Experiments demonstrate up to 50% reduction in spoilage, 35% energy savings, and 25% lower emissions. The system also cuts travel time by 30% and improves delivery reliability and fruit quality. This work offers a scalable, intelligent, and sustainable supply chain framework, especially suitable for resource-constrained or intermittently connected environments, laying the foundation for future-ready food logistics systems
Approximate Newton Methods for Distributed Learning over Communication-Constrained Wireless Networks
Get PDFTraditional Machine Learning (ML) methodologies typically involve aggregating datasets
on a central server for analysis and model training. While effective in certain contexts,
this centralized approach presents significant limitations, particularly concerning data
sensitivity, privacy, and security. These constraints hinder the full realization of ML’s
potential, thereby slowing progress across a range of applications.
Distributed Machine Learning (DML) offers a promising alternative by decentralizing
the learning process. In this paradigm, data remains at its source, while learning
models are transmitted to the data, thereby eliminating the need for data extraction.
Federated Learning (FL), a prominent subset of DML, is specifically designed to address
challenges related to data privacy. FL commonly employs distributed stochastic gradient
descent (DSGD) for optimization, yet it faces several practical challenges, including slow
convergence and high communication overhead.
This thesis investigates enhanced alternatives to conventional FL through the application
of Hessian-based optimization methods. By leveraging second-order information, the
learning process can be accelerated, requiring fewer iterations and reduced communication
to accomplish learning tasks efficiently.
The work addresses key challenges in FL and Fully Distributed Learning (FDL) over
wireless networks, with a particular emphasis on minimizing communication costs while
exploiting the advantages of second-order optimization.
In the FL setting, we propose Distributed Approximate Newton with Determinantal
Averaging (DANDA), a Newton-type method that significantly reduces the number of
communication rounds required for convergence. To accommodate the limited computational
capabilities of client devices, we incorporate approximation techniques for Hessian
computation. DANDA operates over over-the-air Multiple Access Channels (MAC), and
its performance is analyzed under realistic wireless conditions with channel fading and
noise.
Building on this, we develop Lazy-DANDA and Lazy-DANTA, approximate Newtonbased
algorithms tailored for fading MAC environments. These methods integrate subsampled
Hessians, weighted Hessian averaging, and an adaptive Hessian update strategy
that transmits updates only when necessary, thereby further reducing communication
overhead. To counteract distortions from channel fading, we incorporate channel inversion
and power control mechanisms, preserving signal quality while regulating power
consumption.
In the FDL scenario, we extend the server-dependent GIANT algorithm into Network-
GIANT, enabling decentralized node-to-node learning without a central server. This is
achieved by combining gradient tracking, Newton-type iterations, and consensus-based
averaging of Newton updates. Network-GIANT achieves semi-global exponential convergence
under strong convexity and smoothness assumptions, addressing the slow convergence
issue inherent to fully distributed setups. We further refine this approach into
Network Exact Convergence-GIANT, which employs finite-time distributed consensus to
match the exact convergence properties of GIANT in the FL setting.
Collectively, these contributions advance the state of the art in second-order optimization
for FL and FDL, delivering faster convergence, reduced communication overhead, and
improved robustness under realistic wireless network conditions