A description of emergency nursing in a Danish context through focus group interviews.

Artificial intelligence in microscopic hair imaging for scalp disorders: From image acquisition to clinical decisions.

While pre-trained models, such as large language models, can achieve high performance with minimal fine-tuning, the source datasets used for pre-training often contain irrelevant or blackundant data, which can degrade performance on target tasks. Domain Adaptation Information Gain (DAIG)-based source data selection improves performance by pre-training on source data selected based on rough prior knowledge obtained from target data in advance. However, DAIG’s key component, the transition matrix, lacks flexibility and is limited to handling only single-label classification tasks. To address this limitation, we propose the Generalized DAIG (GDAIG)-guided selection process, a novel framework that extends DAIG to support multi-label classification. GDAIG introduces a soft transition matrix to capture inter-label dependencies and employs binary cross-entropy loss to enable adaptation to multi-label data. By leveraging “rough prior knowledge” from initial training on target data, GDAIG actively selects informative and task-relevant source data for pre-training. Experiments on medical image and general object classification datasets demonstrate that GDAIG consistently outperforms baseline approaches, with particularly significant improvements in scenarios involving label mismatch between source and target domains (partial or no label overlap), where conventional transfer learning methods suffer from noise caused by irrelevant source labels. These results highlight GDAIG’s ability to enhance the effectiveness of pre-trained models through strategic source data selection, thereby optimizing performance for specific target tasks. Our framework goes beyond existing approaches that rely solely on pre-trained models, emphasizing the direct utilization of task-relevant source data. Furthermore, GDAIG provides a practical and effective solution for domains with scarce labeled data, such as medical image analysis. • A GDAIG-guided data selection strategy for multi-label classification is proposed. • GDAIG improves target model performance through task-relevant multi-label data selection. • A probabilistic transition matrix captures inter-label dependencies. • “Rough prior” from target data effectively guides source data pre-training. • GDAIG outperforms conventional baselines across diverse multi-label scenarios.

Autoregressive model with discrete feature representation for well log interpretation

In-memory processing can reduce data movement, alleviate the bottleneck between memory and processor, and improve performance and energy efficiency. Associative Processors (AP) in particular have demonstrated substantial performance gains. However, AP hardware designs with full arithmetic support are still limited, with existing implementations often inefficient or tailored to specific tasks, which limits their evaluation across a broader range of applications. This paper presents the design and FPGA implementation of an AP that performs parallel bitwise logic, addition, subtraction, and multiplication directly in memory, making efficient use of LUTRAM and other FPGA logic elements. Using a RISC-V platform as a reference, we compare the AP with a traditional CPU and a vector processor, achieving up to 25x speedup over the CPU on a convolution kernel and an average 3x speedup over the vector processor while using fewer FPGA resources. Compared to a state-of-the-art AP, our implementation consumes 4x less power and requires only one-third of the hardware area. • Data movement limits performance; AP enables in-memory parallelism. • RTL Associative Processor for FPGA using a structural approach. • 25 × faster than RISC-V core, 3 × faster than vector processor. • Smaller than vector processor, one-third of the area of a state-of-the-art AP.

Toward a foundational brain model of intelligence.

Reservoir computers, using recurrent neural networks with fixed random connections, are known to perform a wide range of information-processing tasks. Yet the transformations taking place within the reservoir, the interaction between input matrix, reservoir, and readout layer, and the influence of key design parameters remain insufficiently understood. Here, we shift the focus from performance maximization to the identification of minimal computational requirements for different model tasks. We investigate how many neurons and how much nonlinearity are needed to solve specific tasks, including cases with non-sigmoidal activation functions. Our results show that the division of labor between input matrix, reservoir, and readout layer depends strongly on the task. In many cases, a weakly coupled and only minimally nonlinear reservoir proves sufficient. In addition, features often considered secondary, such as the structure of the input matrix or the steepness of the activation functions, can become decisive for performance.

Automatically constructing Deep Neural Networks (DNNs) has become a key focus in artificial intelligence research, as their performance is highly dependent on the architecture and parameters of the network. Careful selection of the architecture and parameters for specific tasks can significantly improve the network's output. This study proposes an intelligent and efficient framework for Neural Architecture Search (NAS) by integrating Particle Swarm Optimization (PSO), the concept of Fisher Duty Interval (FDI), Transfer Learning (TL), and graph-based architecture representation. Given the high dependency of DNN performance on architecture design, narrowing the search space and guiding it toward promising regions is essential for improving model effectiveness. In the initial phase, the Fisher Information Matrix (FIM) is used to compute the statistical distance FDI between the target task and a set of previously solved tasks. This serves as the basis for TL, allowing the generation of informed initial particles from previously successful architectures. PSO then performs a multi-level search, balancing global and local exploration. In early iterations, candidate architectures are evaluated relative to baseline architectures, while in later stages, comparison shifts dynamically toward the global best particle. Architectures are encoded as Directed Acyclic Graphs (DAGs) with cell-based modules, allowing for heterogeneous and flexible designs where each cell can have a distinct structure. Additionally, weight inheritance and FIM-based performance estimation reduce the need for full training during each iteration, minimizing computational cost. Overall, this framework leverages prior knowledge, reduces search complexity, and efficiently explores the architecture space to discover models that outperform baseline designs and random search results. The effectiveness of this approach has been tested through classification tasks on well-known datasets, with comprehensive results reported accordingly.

Motor control: Generating targeted grooming.

Attention is thought to periodically allocate cognitive resources to task-relevant stimuli, enabling the sequential sampling of multiple stimuli over time. However, whether the specific task influences the temporal dynamics of attentional sampling remains not fully understood. This study investigates the dynamics of attentional sampling and how these dynamics are influenced by task demands, using high-frequency steady-state visual evoked potentials (SSVEPs). Participants were instructed to either select one of two visual stimuli or integrate both stimuli into a single target, with the stimuli tagged at 42 Hz and 44 Hz, respectively. Analysis of the SSVEP envelopes revealed that attention periodically samples stimuli within the theta band (4-6 Hz) in both tasks. Notably, the phase relationship between the SSVEP envelopes differed across tasks: distinct phases were observed in the selection task, whereas similar phases were found in the integration task. This task-dependent phase modulation was mediated by theta-band neural oscillations in the brain. Furthermore, the phase difference between SSVEP envelopes correlated with behavioral performance in the integration task but not in the selection task. These findings demonstrate a flexible, task-dependent mechanism for rhythmic attentional sampling, wherein neural oscillations align stimulus processing according to the specific demands of the task.

The deployment of large language models (LLMs) on extended reality (XR) devices has great potential to advance the field of human-AI interaction. In case of direct, on-device model inference, selecting the appropriate model and device for specific tasks remains challenging. In this paper, we present AIvaluateXR, a comprehensive evaluation framework for benchmarking LLMs running on XR devices. To demonstrate the framework, we deploy 17 selected LLMs across four XR platforms-Magic Leap 2, Meta Quest 3, Vivo X100s Pro, and Apple Vision Pro-and conduct an extensive evaluation. Our experimental setup measures four key metrics: performance consistency, processing speed, memory usage, and battery consumption. For each of the 68 model-device pairs, we assess performance under varying string lengths, batch sizes, and thread counts, analyzing the tradeoffs for real-time XR applications. We finally propose a unified evaluation method based on the 3D Pareto Optimality theory to select the optimal device-model pairs from the quality and speed objectives. Additionally, we compare the efficiency of on-device LLMs with client-server and cloud-based setups, and evaluate their accuracy on two interactive tasks. We believe our findings offer valuable insights to guide future optimization efforts for LLM deployment on XR devices. Our evaluation method can be followed as standard groundwork for further research and development in this emerging field.

To systematically evaluate the diagnostic accuracy of artificial intelligence (AI) models in periapical radiography for detection, classification, and segmentation tasks compared to human experts, while critically appraising methodological quality and risk of bias. The systematic review was conducted in accordance with PRISMA guidelines and registered in PROSPERO (CRD420251132033). Strict exclusion criteria were applied to studies with insufficient dataset sizes (< 200 images) lacking augmentation, ambiguous reference standards, or reporting only accuracy without complementary metrics. A comprehensive literature search was conducted on June 23, 2025, across five electronic databases: PubMed/MEDLINE, Scopus, ScienceDirect, Web of Science, and IEEE Xplore. The search strategy combined keywords related to "periapical radiography" (e.g., periapical X-ray), "artificial intelligence" (e.g., deep learning, neural networks), and specific diagnostic tasks, with no restrictions on publication date or language. Studies utilizing artificial intelligence models for diagnostic tasks on periapical radiographs and validating performance against a human reference standard (expert consensus) were eligible for inclusion. Out of 544 identified records, 47 studies met the full eligibility criteria and were included in the qualitative synthesis. AI models possess a high diagnostic potential in periapical radiography, performing at a level comparable to human experts in tasks such as pathology detection, anatomical segmentation, and implant classification. However, their clinical applicability is currently limited by a high risk of bias, lack of external validation, reliance on cropped datasets, and the use of human consensus as a surrogate reference standard. Future research must prioritize full-arch evaluations, anatomical region-specific performance reporting, and adherence to AI-specific standardized metrics.

Artificial intelligence (AI) is increasingly used in healthcare but often lacks clinician and patient trust. Explainable AI (XAI) aims to clarify predictions and to make AI decisions more transparent, interpretable, and clinically actionable. Yet, current methods fall short. In this Perspective, we argue that, for XAI to be clinically useful in medical imaging and to build trust with clinicians, it must satisfy three guiding principles: technical robustness, adaptation to end users, and alignment of explanations with the specific clinical task. We introduce a conceptual framework, incorporating these principles, to guide future XAI design and deployment based on expectations and shared responsibilities for developers, vendors, and healthcare institutions. By ensuring robustness, personalizing outputs, and aligning explanations with use cases, XAI can move beyond one-size-fits-all approaches to task- and user-centered design, to support effective and trustworthy AI adoption in healthcare.

Various mathematical models have been formulated to describe the changes in synaptic strengths resulting from spike-timing-dependent plasticity (STDP). A subset of these models include a third factor, dopamine, which interacts with spike timing to contribute to plasticity at specific synapses, notably those from cortex to striatum at the input layer of the basal ganglia. Theoretical work to analyze these plasticity models has largely focused on abstract issues, such as the conditions under which they may promote synchronization and the weight distributions induced by inputs with simple correlation structures, rather than on scenarios associated with specific tasks, and has generally not considered dopamine-dependent forms of STDP. In this paper we introduce forms of dopamine-modulated STDP adapted from previously proposed plasticity rules. We then analyze, mathematically and with simulations, their performance in two biologically relevant scenarios. We test the ability of each of the three models to complete simple value estimation and action selection tasks, studying the learned weight distributions and corresponding task performance in each setting. Interestingly, we find that each plasticity rule is well suited to a subset of the scenarios studied but falls short in others. Different tasks may therefore require different forms of synaptic plasticity, yielding the prediction that the precise form of the STDP mechanism present may vary across regions of the striatum, and other brain areas impacted by dopamine, that are involved in distinct computational functions.

ABSTRACT Dealing with the chances and challenges of genAI has become ubiquitous for teachers and learners but one group of students specifically finds itself in a peculiar situation: pre-service language teachers. They need to learn how to use genAI both as students and pre-service language teachers. Therefore, this paper focuses on a group of pre-service English language teachers, their perceptions on the pedagogical benefits of using genAI for in-class activities and how second language teacher educators can support their students’ learning about the pedagogical benefits of genAI tools. By applying a case study approach, data were collected from 15 pre-service English language teachers through reflective learning journals and a post-task online survey, focusing on their experiences in two university classes in which a didactic approach with genAI-supported tasks was implemented. Findings indicate that the students recognised the diverse applications of genAI for language learning and identified specific tasks, such as revising lesson plans, as particularly beneficial. The study highlights the role of second language teacher educators in fostering AI literacy and suggests that updated curricula and frameworks are essential for preparing pre-service language teachers to navigate the challenges and opportunities presented by genAI in language teaching.

Transcranial alternating current stimulation (tACS) is a noninvasive brain stimulation method that modulates neural activity by applying low-intensity alternating current to the scalp. Although tACS has shown promise in enhancing cognitive and motor functions and alleviating neuropsychiatric symptoms, variations in stimulation parameters led to inconsistent outcomes. While stimulation frequency and electrode montage have been extensively explored, systematic analyses focusing on stimulation duration remain limited. Resting-state EEG, recorded under relaxed conditions without specific tasks, minimizes variability due to individual performance and external factors, thus providing a stable measure of tACS-induced neuromodulation. Therefore, we aimed to investigate the differences in neuromodulatory effects between three tACS durations for effective neuromodulation using resting-state electroencephalography (EEG). Ten participants completed three randomized tACS sessions on different days, each with a duration of 10, 20, or 30min. Resting-state EEG was recorded before and after stimulation under eyes-open (EO) and eyes-closed (EC) states. Power spectral density (PSD) and network indices were analyzed for neuromodulatory effects. The omnibus analysis revealed no significant main effect of stimulation duration on neuromodulatory outcomes. However, within-condition analyses revealed significant increases in PSD in the post-tACS EO recording after 10min (r = 0.79) and 20min (r = 0.66), whereas in the post-tACS EC recording significant increases were observed only at 10min (r = 0.63). Network efficiency also increased significantly in the EO state after 10 (r = 0.79) and 20min (r = 0.85) for clustering coefficient and after 10 (r = 0.73) and 20min (r = 0.85) for path length, respectively. Moreover, the observed patterns differed between brain states, with more consistent effects observed in the EO state. These findings suggest that neuromodulatory responses may vary depending on both tACS duration and brain state, highlighting the importance of considering both factors in the design of tACS protocols and interpretation of neuromodulatory effects.

Women with Vocal Fold Nodules: Analysis of Voice Quality and Vocal Self-Perception.

Cell surface LPxTG motif protein (LMP) plays a crucial role in adhesion and biofilm formation during lactic acid bacteria (LAB) colonization. LMP anchors to the cell wall via its C-terminal LPxTG motif and contains functional domains that perform specific tasks. By integrating bioinformatics analysis with existing literature, this review systematically evaluates, for the first time at the domain level, the diversity of LMP in LAB and constructs a functional domain framework for understanding their roles in adhesion and biofilm formation. Annotation analysis reveals domain diversity across LAB strains: mucin-binding and extracellular-matrix-binding domains are widespread, while others exhibit strain-specific characteristics. We also speculate that certain LMP domains may form pilus-like adhesion structures via sortase C-mediated polymerization. Overall, this domain-level elucidation of LMP probiotic mechanisms provides a theoretical basis for probiotic strain screening and functional optimization, and offers new insights into structural and functional localization of LMP on the LAB surface.

Background: Early motor milestones play a critical role in shaping developmental trajectories across motor, cognitive, social, and functional domains. Increasing evidence indicates that motor competence facilitates environmental exploration, learning opportunities, and social engagement during infancy and early childhood. Methods: The present scoping review aimed to map and synthesize the existing evidence on early motor interventions in children aged 0–6 years across diverse pediatric populations. A comprehensive literature search was conducted across PubMed, Scopus, and Web of Science. Studies were selected based on predefined inclusion criteria, and data were extracted and synthesized using a descriptive and thematic approach. Results: A total of 30 studies were included, encompassing a wide range of populations, including preterm infants, children at risk of cerebral palsy, and typically developing children. Across studies, early motor interventions were associated with improvements in motor outcomes and, in many cases, broader developmental domains such as cognition and social interaction. Intervention effectiveness appeared to be influenced by factors such as timing, intensity, task specificity, and caregiver involvement. Conclusions: The review provides a cross-population synthesis of early motor interventions and proposes a conceptual framework that integrates shared mechanisms underlying effective intervention across diverse pediatric groups. This approach offers a more unified understanding of how early motor interventions influence developmental trajectories beyond diagnosis-specific perspectives.

In recent years, robotics has emerged as a promising option in many industries to automate tasks that humans manually perform. However, for maintenance tasks, which often occur in unpredictable or confined areas, nonmobile or fully autonomous robots are not a practical solution. Mobile robot teleoperation offers a promising solution for executing tasks while preserving safety and reliability. This paper presents a safety-oriented framework for mobile robot teleoperation in nuclear maintenance. The framework provides a systematic approach to select end-effectors for a given, specific maintenance task. Various control strategies for the selected design are evaluated within a high-fidelity nuclear power plant digital twin (DT) environment. The DT setup enables technicians to remotely control robots in a virtual world using Meta Quest 3. Last, physical system deployment is conducted on a shape-tracing benchmark task to confirm safe and effective performance in real-world settings, with iterative refinement loops using the DT. This framework lays the groundwork for safe guidelines for mobile robot teleoperation methods in the nuclear industry. In this project, a primary focus is placed on the interaction between the manipulators of mobile robots and the safety-critical components in nuclear facilities, with an emphasis on DT integration to enable safety factors. This approach not only highlights the advantages of mobile robot teleoperation but also provides a practical template for the end-to-end development of robotic systems that can be applied to nuclear maintenance tasks.