Twisted convolutional networks (TCNs): Enhancing feature interactions for non-spatial data classification.

Advancing Whole-Person Health through Informatics: A Narrative Review of Knowledge Resources for Complementary and Integrative Health.

Even though the ocean covers the majority of the planet’s surface, it remains the least explored ecosystem. As light and radio waves do not propagate through water, underwater acoustics is the main choice for various ocean applications ranging from marine biology to pollution monitoring. Increasing levels of anthropogenic noise from ships contribute significantly to underwater sound pollution, posing risks to marine ecosystems. This makes monitoring crucial to understand and quantify the impact of the ship radiated noise. Passive Acoustic Monitoring (PAM) systems are widely deployed for this purpose, generating years of underwater recordings across diverse soundscapes. Manual analysis of such large-scale data is impractical, motivating the need for automated approaches based on machine learning. Recent advances in automatic Underwater Acoustic Target Recognition (UATR) have largely relied on supervised learning, which is constrained by the scarcity of labeled data. Transfer Learning (TL) offers a promising alternative to mitigate this limitation. In this work, we conduct the first empirical comparative study of transfer learning for UATR, evaluating multiple pretrained audio models originating from diverse audio domains. The pretrained model weights are frozen, and the resulting embeddings are analyzed through classification, clustering, and similarity-based evaluations. The analysis shows that the geometrical structure of the embedding space is largely dominated by recording-specific characteristics. However, a simple linear probe can effectively suppress this recording-specific information and isolate ship-type features from these embeddings. As a result, linear probing enables effective automatic UATR using pretrained audio models at low computational cost, significantly reducing the need for a large amounts of high-quality labeled ship recordings. • This study presents a comparative empirical analysis of pretrained audio models for UATR. • Transfer learning reduces the need for a large labeled ship-noise datasets. • Linear probing enables effective ship-type classification from frozen audio embeddings. • Ship-type information is linearly decodable from a small subspace of the embeddings. • Pretrained audio models provide a low-cost solution for automatic UATR under minimal supervision.

Purpose This study proposes a novel hybrid artificial neural network (H-ANN) framework, inspired by reinforcement learning (RL), to proactively detect Internet connection speed problems using enriched datasets from multiple sources of an Internet service provider. Design/methodology/approach The problem is challenging due to the high dimensionality, unbalanced class distribution and continuous influx of new data. To address these issues, the proposed hybrid framework integrates supervised learning methods – radial basis function network (RBFN) and multi-layer perceptron (MLP) – with the unsupervised self-organizing map (SOM). RL is employed to accelerate learning, reduce feature and instance space complexity and improve the detection of underrepresented classes. The framework is first validated on benchmark open-source datasets and subsequently applied to real-world company databases combining network, business and customer information. Findings The results demonstrate that the proposed H-ANN significantly improves both classification accuracy and computational efficiency compared to conventional machine learning approaches. Importantly, the framework enables the early identification of slow Internet connections before customers submit complaints, allowing the service provider to take proactive measures. Originality/value The proposed H-ANN framework not only enables the early identification of slow Internet connections before customers submit complaints – allowing service providers to take proactive measures – but also offers a generalizable solution for large-scale, imbalanced and dynamic data classification problems across diverse domains.

Central venous access devices (CVADs) are essential in paediatric care but pose significant risks. Synthesising existing evidence is needed to guide safe, effective, and equitable practice amid evolving interventions and complex management needs. To develop an evidence and gap map (EGM) to identify, categorise, and visualise paediatric evidence on interventions aimed at improving CVAD outcomes. Following Campbell Collaboration guidance, systematic searches were conducted in PubMed, CINAHL, Scopus, and CENTRAL (date limits: 2014 to 30 June 2024). Eligible studies included patients (0-18 years) evaluating an intervention to improve CVAD outcomes, including randomised and non-randomised trials, implementation studies, and systematic reviews. Two reviewers independently screened and extracted data on CVAD type, intervention, setting, outcomes, and study design. Data were descriptively analysed and visualised in Tableau. Of 952 studies in the broader EGM, 151 were paediatric-specific. Most were conducted in high-income countries (72%) and high-acuity settings, including critical care (41.9%) and oncology (38.5%). CVAD type was unspecified in 80.1% of studies. Systematic reviews (22.5%) and randomised controlled trials (28.5%) were available, though 40.4% of studies were before-and-after studies without controls. Common interventions addressed infection prevention, insertion technologies, and flushing. Clinical outcomes, particularly bloodstream infection (27.8%), dominated reporting, while patient-reported, economic, and device removal outcomes were rarely reported (< 2%). Only studies from the last 10 years and English-language databases were included. No formal quality appraisal was conducted. Significant evidence gaps exist. Future research should prioritise rigorous, paediatric-specific studies across diverse settings and outcome domains. Open Science Framework (OSF) q6gcr: https://osf.io/q6gcr/overview.

Systematic reviews (SRs) are critical for evidence-based research but are time-consuming and labor-intensive. The rapid expansion of academic publications further challenges the performance and applicability of existing screening and classification methods. While large language models (LLMs) present new opportunities for automation, limited research has examined whether they can achieve classification performance comparable to human reviewers in large-scale, multi-class settings. With the goal of improving classification performance, we proposed an LLM-based framework that leverages full-text key-insight extraction to enhance literature classification. We constructed a manually curated dataset of 900 articles from 17 published SRs to quantitatively evaluate the classification capabilities of LLMs. The results provided empirical evidence of LLMs' potential in supporting large-scale SRs and introduced a practical pathway for improving efficiency and reliability in evidence synthesis. Empirical results showed that key-insight-based classification (KBC) significantly outperforms abstract-based classification (ABC). We implemented a confidence-weighted voting (CWV) mechanism using multiple LLMs to improve robustness. The CWV method achieved the highest macro F1-score of 0.796, substantially exceeding KBC (0.732), ABC (0.676), and unsupervised K-means clustering (0.446). By employing zero-shot LLMs, our approach demonstrated the potential for enhanced adaptability across diverse domains and classification tasks without requiring fine-tuning, demonstrating that a carefully designed pipeline can enable LLMs to achieve classification performance comparable to human reviewers.

Multivariate mixture data analysis presents numerous challenges and constitutes a vital area of interest in the fields of statistics and data science. Research into multivariate mixture structures holds relevance across diverse application domains and plays a pivotal role in the advancement of artificial intelligence and machine learning. In this article, we focus on nonparametric estimation techniques for multivariate mixture data. Specifically, we assume a known number of subpopulations and propose a binomial likelihood method, along with an efficient numerical algorithm, to estimate the mixing proportions and cumulative distribution functions of these subpopulations without relying on parametric assumptions. Through extensive numerical experiments, we demonstrate three key advantages of our approach: (a) Our method eliminates the need for tuning parameters. (b) It does not require the assumption of continuous component density functions. (c) Our method consistently delivers stable performance. Under mild regularity conditions, we provide theoretical proofs for the asymptotical properties of our estimators. To illustrate the practical performance of our method, we include a real-data example. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

Edge-cloud systems, which bring computing, storage, and networking resources closer to end-users, offer significant advantages in reducing latency and enabling real-time data processing. These systems are increasingly deployed across diverse domains, such as smart manufacturing, autonomous vehicles, and large-scale IoT networks, to support big data-driven services that require continuous analytics and rapid response. Ensuring software reliability in these environments is critical, which has led to growing attention on just-in-time (JIT) defect prediction as an effective technique for prioritizing testing efforts by identifying code changes likely to introduce defects. However, existing techniques struggle to perform accurately on new or low-data projects due to insufficient training data. In this paper, we propose PROPER-SDP, a prompt-based approach that leverages large language models. By incorporating project evolution data directly into prompts, our approach enables LLMs to effectively capture the contextual information essential for accurate JIT defect prediction. By doing so, we effectively address the cold-start problem, allowing accurate JIT defect prediction even in the absence of project-specific training data. Evaluation results demonstrate that our method significantly improves prediction performance, surpassing baseline methods by an average of 19.7% in F1-score. Our approach enables reliable JIT defect prediction even in rapidly evolving, resource-constrained edge-cloud systems.

Active learning (AL) aims to maximize model performance while minimizing annotation costs. With the rapid adoption of deep learning, AL approaches have evolved to meet contemporary demands. We systematically examine the literature published from 2018 to May 2025, focusing on four key trends: batch-mode selection, transfer learning integration, multi-strategy querying, and extension to diverse application domains. In addition, we summarize classical AL approaches. While observations show that combining AL with deep learning significantly enhances data efficiency, a critical limitation remains: the lack of standardized evaluation protocols across studies hinders precise comparisons. Nevertheless, we find that AL is well-aligned with modern trends, and we offer insights into underexplored opportunities to guide future research within the machine learning community.

The rapid expansion of digital data poses a unique challenge for retrieving relevant and insightful information efficiently. In particular, the increasing volume of scientific publications has made literature reviews time-consuming. The emergence of large language models (LLMs) offers new opportunities to streamline this process. This paper explores the use of generative artificial intelligence (GenAI) for query reformulation and evaluates the performance of nine massive text embedding models, varying in size and fine-tuning strategies, in the context of document retrieval. We apply multiple prompt engineering techniques to evaluate the ability of LLMs to generate effective queries, comparing them with human-crafted queries. These are used to retrieve documents utilizing nine embedding models. The evaluation is across five datasets using metrics such as recall, average precision, and rank-based measures. Results show that embedding models fine-tuned for semantic similarity consistently outperform general-purpose models, with UAE Large proving most robust across diverse domains. Furthermore, queries generated using zero-shot and few-shot prompting techniques often surpass the performance of human-formulated queries. These findings highlight the value of integrating LLMs and massive text embeddings to reduce manual effort in literature reviews. GenAI provides a reliable starting point for query formulation, with human input reserved for refinement when needed.

Retrotransposons play outsized roles in the evolution of gene regulation, genome function, and disease pathogenesis, and more recently they have sparked interest as instruments for new gene therapy approaches. R2 retrotransposons insert site-specifically into the multicopy genes encoding 28S ribosomal RNA at a target sequence conserved broadly across eukaryotic evolution. R2 retrotransposons have been detected in many animals, but previous surveys have been limited in scope and methodology. Here, we substantially expand the known distribution of R2 retrotransposons from previously unrepresented or underrepresented taxonomic groups, ranging from ctenophores to amphibians and reptiles. We discover diverse R2 domain architectures and motifs and identify many new avian R2 candidates for genome engineering development. Overall, phylogenetic analyses reveal two highly successful R2 lineages. We describe lineage-distinctive features of the N-terminal DNA recognition domains and reverse-transcriptase domain signatures. Within a lineage, R2 protein sequences do not necessarily preserve the unifying configuration of N-terminal domains assumed in the current clade classification scheme. We show that recombinant R2 proteins with distinctive domain architectures and distribution across major animal classes support target-primed reverse transcription with conserved site specificity. Our analysis of the surprisingly varied domain architectures that support target-site specificity informs new R2 classification criteria and provides a greatly expanded foundation for additional structure/function insights about DNA binding selectivity. This expanded perspective on R2 evolution informs approaches for engineering therapeutic gene insertion technologies and offers an opportunity to investigate the conservation and diversification of retrotransposons.

Non-suicidal self-injury (NSSI) persists as a major public health challenge worldwide. Identifying and strategically targeting risk factors for NSSI constitutes a practical approach to its prevention. We aim to synthesize existing knowledge concerning the range and magnitude of risk factors for NSSI among children and adolescents, and to critically assess the robustness of the available evidence. In this umbrella review, six bibliographic databases were systematically searched for articles published from database inception to Dec 2024. For the assessment of evidence credibility, pre-specified criteria for classifying evidence were utilized, categorized as convincing ("class I"), highly suggestive ("class II"), suggestive ("class III"), weak ("class IV"), or no evidence ("class V"). The Amstar-2 framework was employed to evaluate the quality of the evidence which graded as "high," "moderate," "low," or "critically low" quality. The study included meta-analyses of observational studies in the past 30 years on risk factors for NSSI in children and adolescents. We identified 16 meta-analyses comprising 410 primary studies on 43 risk factors from 38 countries, involving 2,659,156 children and adolescents. Twenty-three (e.g. LGBTIQ) risk factors were categorized as individual, followed by family level (n = 8, e.g. childhood maltreatment), school/peer level (n = 8, e.g. bully victims) and multifactorial level (n = 4, e.g. no religion). Eighteen (41.86%) risk factors provided highly suggestive (Class II) evidence of association with NSSI. Suggestive evidence (class III) indicated that NSSI was associated with adverse childhood experiences (2.31, 1.77-3.01) and being left-behind children (1.37, 1.11-1.69). A multitude of risk factors spanning diverse domains were identified, highlighting the multifactorial nature of NSSI in adolescents and children. Comprehensive prevention strategies and measures should be conducted for children and adolescents to decrease the risk of NSSI and associated harms in multilevel approaches.

Abstract – Multi-Agent Systems (MAS) have emerged as a powerful and flexible paradigm within distributed artificial intelligence, enabling the efficient solution of complex, large-scale, and dynamic problems by decomposing them into smaller, manageable subtasks handled by multiple autonomous agents. An agent can be defined as an independent computational entity capable of perceiving its environment through available inputs, processing this information using predefined rules or learning algorithms, and taking appropriate actions to achieve its assigned objectives. In a MAS, these agents do not operate in isolation; instead, they interact, collaborate, and sometimes compete with one another, sharing knowledge and coordinating their actions to achieve both individual and collective goals. This collaborative behavior significantly enhances system performance, allowing MAS to exhibit key characteristics such as scalability, adaptability, robustness, and fault tolerance, which are essential in real-world applications. The distributed nature of MAS reduces the dependency on a central controller, thereby eliminating single points of failure and enabling the system to continue functioning even if some agents fail or behave unexpectedly. MAS have been widely applied across diverse domains including computer networks, cloud computing, robotics, smart grids, transportation systems, social networks, and urban infrastructure, where they facilitate intelligent decision-making, efficient resource management, and real-time problem-solving. For instance, in cloud computing, agents can dynamically allocate resources and balance workloads, while in robotics, they enable coordinated movement and task execution among multiple robots. Furthermore, the integration of advanced learning techniques such as reinforcement learning and evolutionary algorithms allows agents to adapt to changing environments, improve their decision-making over time, and handle uncertainty more effectively. However, despite their numerous advantages, MAS also face several significant challenges that must be addressed to fully realize their potential. These include coordination and consensus among agents, efficient communication in large-scale systems, task allocation based on agent capabilities, fault detection and isolation, maintaining system security, and managing dynamic and unpredictable environments. Additionally, issues such as scalability, synchronization, and maintaining connectivity among agents further complicate system design and implementation. This comprehensive exploration of Multi-Agent Systems provides an in-depth understanding of their fundamental principles, architectural characteristics, and operational mechanisms, along with a detailed examination of their applications and associated challenges. By analyzing both theoretical foundations and practical implementations, this work offers valuable insights into the design and development of intelligent, distributed systems, making it a useful resource for researchers, engineers, and practitioners aiming to build advanced agent-based solutions in modern computing environments. Key Words: Multi-Agent Systems (MAS), Autonomous Agents, Distributed Artificial Intelligence, Agent Communication, Coordination and Collaboration, Task Allocation, Scalability, Adaptability, Reinforcement Learning, Cloud Computing, Robotics, Smart Systems, Fault Tolerance, Security, Intelligent Decision Making.

Internet of Things (IoT) enables seamless connectivity and intelligent automation across diverse domains, from healthcare and agriculture to smart cities and industrial systems. However, conventional IoT architectures often rely on centralized servers for data processing and coordination, resulting in poor scalability and decreased system reliability. The integration of blockchain with IoT offers a promising approach to address these limitations of centralized IoT architectures. In practice, however, existing blockchain consensus mechanisms are often unsuitable for resource-constrained IoT devices and dynamic network conditions. To overcome this limitation, we propose Proof of Periodic Inference (PoPI), a machine learning model-based consensus mechanism tailored for blockchain-based IoT systems. PoPI uses a supervised machine learning model to periodically select a group of block producers and maintains security through random block generation within the group. It incorporates both static and dynamically changing device features, such as battery level and resource usage, to select capable nodes with high reliability, and implements fair participation mechanisms to balance network involvement over time. Theoretical analysis and experimental evaluation demonstrate that PoPI achieves high scalability, low latency and improved applicability compared to the state-of-the-art consensus protocols in dynamic IoT environments.

Multi-camera target tracking remains a fundamental challenge in intelligent video surveillance, particularly when maintaining consistent identity associations across spatially disjoint camera views with overlapping or non-overlapping fields of view. This paper introduces VisionNet, a unified deep learning framework that integrates state-of-the-art detection, single-camera tracking, and cross-camera re-identification into a cohesive and computationally efficient pipeline. VisionNet employs YOLOv8 for high-accuracy object detection, Deep Simple Online and Realtime Tracking (DeepSORT) for intracamera trajectory persistence, and Omni-Scale Network (OSNet) for discriminative cross-camera identity association. The framework leverages GPU-accelerated inference to sustain real-time throughput while preserving tracking fidelity under challenging surveillance conditions such as occlusion, lighting variation, and crowded scenes. A Flask-based monitoring dashboard provides operators with live visualization of tracked targets, inter-camera handoff events, and aggregate behavioral analytics. Extensive evaluation on benchmark datasets including MOT17, DukeMTMC-reID, and Market-1501 demonstrates that VisionNet achieves an IDF1 score of 76.4%, a MOTA of 73.9%, and a Rank-1 re-identification accuracy of 96.2%, outperforming competing baselines on identity-switch minimization. The modular architecture facilitates seamless integration with existing closed-circuit television (CCTV) infrastructure and supports diverse application domains including intelligent security systems, crowd flow analytics, retail behavior monitoring, and multi-athlete sports event tracking.

ABSTRACT Early childhood education and care (ECEC) is expected to provide all children with the opportunity to enhance their skills across diverse learning and developmental domains. Although professionals play a crucial role in assessing and promoting key competences, numerous ECEC centers and educational teams remain inadequately prepared to do so. This study seeks to integrate a comprehensive characterization of professional competences acquired during initial and continuing training and current practices in the implementation of assessment methodologies in ECEC while exploring influencing factors at both the professional and ECEC center levels. The participants (n = 430) comprised teachers working with children aged 3–6 from public, social, and private ECEC centers across Portugal. The findings indicate that the ability of ECEC professionals to design effective learning and developmental assessment systems can be enhanced, underscoring the need for improvements in both initial training and continuous professional development. In addition, only a small proportion of professionals reported having opportunities to develop skills in assessment methodologies, innovation, and curriculum development, which are crucial for effective, child-centered educational responses. In light of our findings, we propose a set of recommendations for policy enhancement regarding professional development and training in the Portuguese ECEC context.

Foundation models (FMs) have achieved remarkable success across diverse machine learning domains through large-scale pretraining on large, diverse datasets. However, pretraining on such datasets introduces significant challenges due to substantial mismatches in data distributions, a problem particularly pronounced with time series (TS) data. In this article, we tackle this issue by proposing a domain-aware adaptive normalization strategy within the transformer architecture. Specifically, we replace the traditional LayerNorm with a prototype-guided dynamic normalization mechanism (ProtoNorm), where learned prototypes encapsulate distinct data distributions, and sample-to-prototype affinity determines the appropriate normalization layer. This mechanism effectively captures the heterogeneity of TS characteristics, aligning pretrained representations with downstream tasks. Through comprehensive empirical evaluation, we demonstrate that our method significantly outperforms conventional pretraining techniques across diverse downstream tasks, while effectively mitigating the adverse effects of distribution shifts during pretraining. Incorporating ProtoNorm is as simple as replacing a single line of code. Extensive experiments on diverse real-world TS benchmarks validate the robustness and generalizability of our approach, advancing the development of more versatile TS FMs.

Vision Foundation Models(VFMs) have achieved remarkable success in various computer vision tasks. However, their application to semantic segmentation is hindered by two significant challenges: (1) the disparity in data scale, as segmentation datasets are typically much smaller than those used for VFM pre-training, and (2) domain distribution shifts, where real-world segmentation scenarios are diverse and often underrepresented during pre-training. To overcome these limitations, we present Rein++, an efficient VFM-based segmentation framework that demonstrates superior generalization from limited data and enables effective adaptation to diverse unlabeled scenarios. Specifically, Rein++ comprises a domain generalization solution Rein G and a domain adaptation solution Rein-A. Rein-G introduces a set of trainable, instance-aware tokens that effectively refine the VFM's features for the segmentation task. This parameter efficient approach fine-tunes less than 1% of the backbone's parameters, enabling robust generalization. Building on the Rein-G, Rein-A performs unsupervised domain adaptation at both the instance and logit levels to mitigate domain shifts. In addition, it incorporates a semantic transfer module that leverages the class-agnostic capabilities of the segment anything model to enhance boundary details in the target domain. The integrated Rein++ pipeline first learns a generalizable model on a source domain (e.g., daytime scenes) and subsequently adapts it to diverse target domains (e.g., nighttime scenes) without any target labels. Comprehensive experiments demonstrate that Rein++ significantly outperforms state-of-the-art methods with efficient training, underscoring its roles an efficient, generalizable, and adaptive segmentation solution for VFMs, even for large models with billions of parameters. The code is available at https://github.com/wloves/Rein.

Recent studies have investigated various forms of the time-fractional and distributed-order nonlinear Klein–Gordon equations (KGEs). In this work, we focus on the nonlinear KGE involving distributed time derivatives and propose a novel numerical framework for its efficient and accurate solution. For cases where the nonlinear component in the KGE appears as sin ( ψ ( v ) = sin ⁡ ( v ) ), this particular form of the equation is commonly known as the Sine-Gordon equation, whose numerical solution holds significant importance. A Gaussian quadrature-based approach is employed to approximate the distributed derivative term, ensuring high precision in the numerical treatment. The proposed method combines the Generalized Moving Least Squares (GMLS) technique for spatial discretization with a first-order accurate finite difference approximation in the time domain. The experimental findings provide substantial evidence for the efficacy of our novel methodology in accurately capturing the complex dynamics of the equation, highlighting its potential for solving similar problems with distributed derivatives in diverse computational domains.

Background: Although al-Zahrāwī's contributions to medical history are widely acknowledged, modern literature often fragments discussions regarding the specific pedagogical mechanics of his work. Objective : To provide a focused thematic analysis of the 30th volume of Al-Taṣrīf, elucidating how its detailed instrument design and stepwise procedural documentation offered a foundational blueprint for surgical instruction. Methods: A narrative historical review was employed for an in-depth qualitative exploration. Primary data were sourced from the highly acclaimed 1973 Spink and Lewis English translation. To mitigate interpretive bias, findings were systematically cross-referenced with contemporary peer-reviewed historical analyses to validate technical translations and confirm the broader historical context. Findings: Our analysis demonstrates that al-Zahrāwī established a systematic methodological framework that integrated patient positioning, specialized site-specific instrumentation, and stepwise intraoperative guidance. This structured approach served as an early structural precursor to the elements found in modern operative reporting across diverse domains. Conclusion: This review systematically elucidates how the 30th volume of Al-Taṣrīf functioned as a pivotal pedagogical masterwork. By successfully integrating theoretical anatomy with practical applications, it provided a foundational blueprint for surgical education; however, its comparison to modern practice must be carefully contextualized within the scientific limitations of the 10th century.