This study examines how artificial intelligence (AI) and machine learning (ML) can be more effectively integrated into Ireland's Leaving Certificate Computer Science (LCCS) curriculum, where these topics are currently underrepresented. Using a pragmatist, convergent mixed-methods design, it draws on surveys from 41 teachers, interviews with eight teachers, and analysis of international literature. The Technology Acceptance Model (TAM) was used as a theoretical lens and extended into the TAM-AI-Edu framework to capture perspectives on curriculum gaps and future trends. Findings show both enthusiasm for AI/ML integration and concerns around resources, training, and assessment. The study offers teacher-informed recommendations for curriculum reform and introduces TAM-AI-Edu, a model that better reflects the complexities of AI/ML adoption in post-primary education.

We investigate how successful bias mitigation reshapes the embedding space of encoder-only and decoder-only foundation models, offering an internal audit of model behaviour through representational analysis. Using BERT and Llama2 as representative architectures, we assess the shifts in associations between gender and occupation terms by comparing baseline and bias-mitigated variants of the models. Our findings show that bias mitigation reduces gender-occupation disparities in the embedding space, leading to more neutral and balanced internal representations. These representational shifts are consistent across both model types, suggesting that fairness improvements can manifest as interpretable and geometric transformations. These results position embedding analysis as a valuable tool for understanding and validating the effectiveness of debiasing methods in foundation models. To further promote the assessment of decoder-only models, we introduce WinoDec, a dataset consisting of 4,000 sequences with gender and occupation terms, and release it to the general public. (https://github.com/winodec/wino-dec)

Prototypical Part Networks (ProtoPNets) provide interpretable image classification through prototype-based explanations. However, existing methods suffer from prototype collapse, where multiple prototypes converge to similar patterns, limiting both interpretability and performance. We present ProtoPDiv, a novel extension that explicitly optimizes for prototype diversity through four complementary mechanisms: (1) margin-based diversity loss preventing prototype clustering, (2) orthogonality constraints ensuring prototypes capture independent features, (3) attention-guided learning to focus prototypes on semantically rich regions, and (4) entropy regularization promoting uniform prototype utilization. Our three-phase training strategy progressively introduces these mechanisms while maintaining classification performance. Experiments on CUB-200-2011 demonstrate that ProtoPDiv achieves 93.90

Media bias in news reporting significantly influences public perception and societal dynamics. Personalised news recommender systems, widely adopted by digital platforms, often exacerbate patterns of polarisation by amplifying users’ pre-existing preferences, creating filter bubbles, and reinforcing ideological divides. To address this issue, this work explores sentence rewriting as a novel strategy to mitigate polarisation in the dissemination of biased news content. We propose automated techniques leveraging word replacement methods and large language models (LLMs) to rewrite biased news content into neutral versions while preserving meaning and readability. Extensive experiments evaluate the effectiveness of sentence rewriting across various components of the news recommendation feedback loop, including user reading histories, candidate news datasets, and training datasets. Experimental results demonstrate the ability of sentence rewriting to reduce media bias polarisation in the dissemination of biased news content while maintaining or enhancing recommendation accuracy. Furthermore, this work provides a comprehensive assessment of media bias polarisation mitigation through a multidimensional evaluation framework and examines the ethical and societal challenges of addressing media bias polarisation in news recommendations. It also highlights the need for collaborative solutions involving platforms, news organisations, and users to foster more transparent news recommender systems.

While multi-modal deep learning approaches trained using magnetic resonance imaging (MRI) and fluorodeoxyglucose positron emission tomography (FDG PET) data have shown promise in the accurate identification of Alzheimer’s disease, their clinical applicability is hindered by the assumption that both modalities are always available during model inference. In practice, clinicians adjust diagnostic tests based on available information and specific clinical contexts. We propose a novel MRI- and FDG PET-based multi-modal deep learning approach that mimics clinical decision-making by incorporating uncertainty estimates of an MRI-based model (generated using Monte Carlo dropout and evidential deep learning) to determine the necessity of an FDG PET scan, and only inputting the FDG PET to a multi-modal model when required. This approach significantly reduces the reliance on FDG PET scans, which are costly and expose patients to radiation. Our approach reduces the need for FDG PET by up to 92% without compromising model performance, thus optimizing resource use and patient safety. Furthermore, using a global model explanation technique, we provide insights into how anatomical changes in brain regions—such as the entorhinal cortex, amygdala, and ventricles—can positively or negatively influence the need for FDG PET scans in alignment with clinical understanding of Alzheimer’s disease.

The performance of Active Learning (AL) methods varies widely, influenced by the query strategy, model, and dataset, with the reasons for variation in performance still unclear and insufficiently studied. However, commonly used metrics like accuracy, precision, and recall provide only limited analytical perspectives. No research has effectively uncovered or explained the reasons behind these performance variations, leaving a gap in understanding of the factors that influence the success or failure of AL methods. To address this issue, we propose a novel method and tool leveraging Voronoi Diagrams to visualize AL processes by illustrating interactions between classification decision boundary changes and queried samples across AL iterations. We perform experiments on synthetic and real-world datasets to validate the effectiveness of our method and analyze various AL query strategies. By visualizing the AL process, we illustrate how different query strategies progressively select samples and influence performance in each iteration. This reveals the potential benefits of adapting query strategies at different learning stages to improve active learning efficiency.

We report on a trans-Atlantic PhD student mobility program that connects two graduate research training initiatives in the US and Ireland, centered on developing future researchers in artificial intelligence (AI) and machine learning (ML). We discuss both the structure of the student exchange experiences and share key learnings from this international collaboration. The most important lesson learned is that providing a structured mobility program and matched visiting pairs is a highly effective way to improve learning outcomes compared to more typical ad-hoc individual visits.

Subclinical mastitis in cows is a major challenge to the global development of the dairy sector, as it can reduce milk yield and lead to significant financial losses for dairy farmers. Given the difficulty in detecting subclinical mastitis and the growing availability of sensor data during milking, the application of machine learning techniques may offer a solution to more effectively and efficiently detect the problematic condition. Farmers and managers are more likely to use subclinical mastitis prediction tools if they can easily understand how the models work and why they make certain predictions. The advent of Explainable Boosting Machines (EBM) represents an optimal solution, offering results that are both transparent in the decision-making processes and interpretable to domain experts, while maintaining strong predictive power. The study compares models built with readily available data from milking machines to those using more expensive milk characteristic data to determine if there’s an “explanation deficit” when relying on the cheaper, more accessible data. The study finds that while models using the readily available features can perform well in prediction, the explanations derived from them are often less interpretable and harder to align with established veterinary knowledge compared to explanations from models using milk composition data. This suggests a trade-off between data accessibility and the clarity of the model’s reasoning for detecting the condition.

In the rapidly evolving Generative AI (GenAI) landscape, source code and natural language are being mixed and used in new ways. This presents opportunities for rethinking teaching practice in Introductory Programming (CS1) courses that includes, but goes beyond, assessment. In this paper we examine the reasons why and how instructors who are early adopters of GenAI are using it in their teaching, and why others are not. We also explore the changes and adaptations that are currently being made to practice. This is achieved by synthesizing insights from several recent studies that have collected primary data from introductory programming instructors who are teaching with, considering teaching with, or actively not teaching with GenAI. Due to the fast pace of GenAI development and adoption, the fixed-pace and cyclical nature of education, and the relatively slow pace of research (including ethical approvals) and publication cycles, research with primary data from instructors is only being published relatively recently. In computing education, there is not yet enough published research with primary data from CS1 instructors to warrant a systematic literature review, although in the next year this will likely be possible. Based on an analysis of the nascent research that has been published, we propose emerging and flexible guidelines on how CS1 instructors could adapt their practice based on what others have done so far. These guidelines highlight important factors to consider when integrating GenAI in CS1 courses, which for many is only beginning.

This study explores the development and implementation of machine learning (ML) models on Edge-AI boards, aiming to identify the most effective solution for anomaly detection systems on space missions. We investigate ML anomaly detection techniques including Autoencoders, Long Short-Term Memory (LSTM) cells, Isolation Forests, and Transformers. These models were trained on a univariate dataset derived from real space missions and deployed on diverse hardware platforms engineered for space environments to comprehensively assess performance. Specifically, we explore space flight ready boards (Ubotica CogniSAT-XE1 and XE2, which incorporate Intel's Myriad 2 and X chips, respectively); commercial, non-space flight ready, edge-AI boards (NVIDIA's Jetson Nano and Google Coral); and Field Programmable Gate Array (FPGA) implementations (from Microchip, AMD, and NanoXplore). We compare the performance of anomaly detection models run on space flight ready and commercial boards (using CPU performance as a benchmark) to provide a thorough comparison of available platforms for onboard anomaly detection. This paper provides a detailed examination of both the optimal ML models and hardware platforms for deploying univariate anomaly detection systems in space flight contexts and draws conclusions about which ones are most suitable.

This study collected insights from post primary Computer Science teachers regarding the integration of Artificial Intelligence (AI) and Machine Learning (ML) into the Leaving Certificate Computer Science (LCCS) curriculum in Ireland. By surveying teachers, the research evaluates current practices, identifies educational needs and challenges, and seeks to inform future developments to enhance AI education for students. Preliminary findings suggest a range of perspectives, highlighting the importance of professional development and the growing interest in AI within the curriculum.

Personalised news recommender systems are effective in disseminating news content based on users’ reading histories but can also amplify and proliferate biased media. This work examines the potential of automated sentence rewriting methods, utilising word replacement methods and large language models (LLMs), to mitigate this side effect of recommender systems. We present a two-step workflow: the application of automated sentence rewriting methods to rewrite biased sentences, and the integration of these rewritten sentences into the recommendation process. We evaluate the effectiveness of sentence rewriting approaches in a simulation framework, to assess how well they mitigate the spread of biased news. Our study demonstrates that applying sentence rewriting to users’ reading histories can result in a significant reduction in the propagation of biased media. Our contributions are threefold: we pioneer the use of LLMs for mitigating the spread of biased news by recommender systems; we demonstrate that algorithms trained on debiased content maintain or improve recommendation accuracy; and we provide a comprehensive exploration of the effectiveness of applying sentence rewriting methods to various components within a recommender system, as well as an investigation of the underlying reasons for their efficacy. This work advances our understanding of media bias mitigation in news content and recommendation algorithms, providing valuable insights into how news recommender systems can prevent the dissemination of biased information.

The negative effects of media bias, such as influencing readers’ perceptions and affecting their social decisions, have been widely identified by social scientists. However, the combined impact of media bias and personalised news recommendation systems has remained largely unstudied, especially in real-world news recommendation datasets. Our study bridges this gap by analysing how leading algorithms influence the spread of biased news among news recommendation system users with diverse preferences. In this article, we show that current state-of-the-art news recommendation algorithms amplify the amount of biased media that readers consume and that, while the quality of their recommendations is largely similar, different news recommendation algorithms have differing sensitivities to media bias. We present experimental results that compare the performance of different news recommendation algorithms for users with different subject interests and different levels of prior history of reading biased media. Our analysis reveals that some state-of-the-art news recommendation algorithms that perform well at the recommendation task also lead to large amounts of biased news being recommended to readers. These findings suggest significant potential for negative impacts from increasing volumes of biased media being promoted by news recommendation algorithms. This highlights the importance for organisations to offer more trustworthy personalised news recommendations to mitigate the propagation of bias in news consumption.

In this study, we aim to reduce generation latency for Named Entity Recognition (NER) with Large Language Models (LLMs). The main cause of high latency in LLMs is the sequential decoding process, which autoregressively generates all labels and mentions for NER, significantly increase the sequence length. To this end, we introduce Parallel Decoding in LLM for NE} (PaDeLLM-NER), a approach that integrates seamlessly into existing generative model frameworks without necessitating additional modules or architectural modifications. PaDeLLM-NER allows for the simultaneous decoding of all mentions, thereby reducing generation latency. Experiments reveal that PaDeLLM-NER significantly increases inference speed that is 1.76 to 10.22 times faster than the autoregressive approach for both English and Chinese. Simultaneously it maintains the quality of predictions as evidenced by the performance that is on par with the state-of-the-art across various datasets.

Mastitis is a significant disease on dairy farms and can have serious negative animal performance and economic consequences if not controlled. While clinical mastitis is often easily identified due to visibly abnormal milk, subclinical mastitis presents a more insidious challenge. Somatic cell count (SCC) is commonly used to monitor and detect subclinical mastitis, however, SCC is not available at a high sampling frequency rate at the cow level on most farms due to the manual effort involved in collecting it. With the rise of precision dairy farming technologies such as milk meters, however, there is increasing interest in using data-driven approaches (especially approaches using machine learning) for detecting subclinical mastitis based on indicators more easily collected by modern sensors. In this article we introduce milk flow profiles, a new, easy-to-collect data type that can replace more difficult-to-collect data sources (e.g., those that require laboratory tests or manual measurements) in precision dairy farming. The results of our experiments demonstrate that milk flow profiles, combined with other easily accessible milking machine data, can be employed to train machine learning models that accurately detect subclinical mastitis (as evidenced by high SCC measurements), with an AUC of 0.793. Moreover, these models perform better than models trained using features from milk characteristic data that are expensive to collect and are only collected at low frequency on commercial farms. Our experiments used data from 16 weeks of milking events from 285 cows on Irish farms, and their results demonstrate the value of milk flow profiles as an easily accessible and valuable data source for precision dairy farming applications.

Medical imaging is a critical component of clinical decision-making, patient diagnosis, treatment planning, intervention, and therapy. However, due to the shortage of qualified radiologists, there is an increasing burden on healthcare practitioners, which underscores the need to develop reliable automated methods for interpreting medical images to reduce the time spent on commonplace cases and to support radiologists on more complex cases. Despite the development of novel computational techniques, automatically interpreting medical images remains challenging due to the subtlety and nuance of the patterns to be interpreted as well as the presence of noise and varying acquisition conditions. One promising solution to improve the reliability and accuracy of automated medical image analysis is interactive machine learning (IML), which integrates human expertise into the model training process. However, IML methods often lack compelling explanations to help users understand how a model is processing an image. To overcome this limitation, this study introduces a novel approach that leverages active learning (AL) to iteratively query for high-uncertainty samples while utilizing explanations from a prototypical part network to improve model classification. The proposed approach utilizes prototypical parts, which are snapshots of image sections, to determine an unlabelled image’s class based on the presence of the prototypical parts. Interaction occurs during the selection of prototypes and the AL phase, where a set of decision rules is designed to consider the contributions of which combinations of prototypical parts are the most representative of the unlabeled image output by the AL. The proposed explainable interactive machine learning (XIL) framework empowers medical experts to interact with the model’s training process, enabling more efficient and personalized learning through explanation and interaction.

Much of named entity recognition (NER) research focuses on developing dataset-specific models based on data from the domain of interest, and a limited set of related entity types. This is frustrating as each new dataset requires a new model to be trained and stored. In this work, we present a "versatile" model—the P rompting-based Unified NER system (PUnifiedNER)—that works with data from different domains and can recognise up to 37 entity types simultaneously, and theoretically it could be as many as possible. By using prompt learning, PUnifiedNER is a novel approach that is able to jointly train across multiple corpora, implementing intelligent on-demand entity recognition. Experimental results show that PUnifiedNER leads to significant prediction benefits compared to dataset-specific models with impressively reduced model deployment costs. Furthermore, the performance of PUnifiedNER can achieve competitive or even better performance than state-of-the-art domain-specific methods for some datasets. We also perform comprehensive pilot and ablation studies to support in-depth analysis of each component in PUnifiedNER.

There is often a mixture of very frequent labels and very infrequent labels in multi-label datatsets. This variation in label frequency, a type class imbalance, creates a significant challenge for building efficient multi-label classification algorithms. In this paper, we tackle this problem by proposing a minority class oversampling scheme, UCLSO, which integrates Unsupervised Clustering and Label-Specific data Oversampling. Clustering is performed to find out the key distinct and locally connected regions of a multi-label dataset (irrespective of the label information). Next, for each label, we explore the distributions of minority points in the cluster sets. Only the minority points within a cluster are used to generate the synthetic minority points that are used for oversampling. Even though the cluster set is the same across all labels, the distributions of the synthetic minority points will vary across the labels. The training dataset is augmented with the set of label-specific synthetic minority points, and classifiers are trained to predict the relevance of each label independently. Experiments using 12 multi-label datasets and several multi-label algorithms show that the proposed method performed very well compared to the other competing algorithms.