Search Results (766)

This study investigates the pedagogical landscape of state language instruction in Hungarian-medium schools in Slovakia. Situated within the wider context of European minority language policies, this study explores the institutional ecosystems, didactic approaches and teaching strategies, and the relationship between teacher- and student-centered methodologies in state language instruction. A questionnaire survey based on a self-developed Multi-Level Diagnostic Model was administered to a representative sample of teachers, accounting for 23% of the total Slovak teacher population working in this distinctive sociolinguistic setting (N = 112). Although the results indicate that the educational process is shaped by various factors and there is an endeavor to promote communicative practice, the competence–use gap persists due to the reliance on conventional teacher-centered teaching approaches. This trend is driven by a methodological vacuum, the absence of specialized L2 teaching materials and the lack of modern digital resources; it also suggests that teachers are forced to prioritize instructional security rather than being resistant to innovation. The findings suggest that the current educational system is ready for change, but it requires systemic investment in resources to promote the balanced development of intercultural communicative competence. Addressing the linguistic distance between Hungarian L1 and Slovak L2 through specialized materials may promote a model of additive bilingualism that ensures professional credibility and the protection of minority cultural identity. Full article

The move to sustainable energy systems has increased the requirement for comprehensive decision support frameworks that are uncertainty-aware to guide the selection of solar power plant sites. The rapid growth of investments in solar energy has increased the demand for systematic and accurate decision-support tools to choose the best sites for photovoltaic (PV) power facilities. The selection of solar power plant sites is a complicated multi-criteria decision-making (MCDM) problem that involves technical, economic, environmental, social, and technological aspects. The process is typically associated with ambiguity and incomplete knowledge of experts. To overcome these problems, this paper offers an interval-valued neutrosophic fuzzy TOPSIS (IVN-TOPSIS) method, which extends the standard TOPSIS methodology by including truth, indeterminacy, and falsity membership degrees as interval values. The methodology is utilized in a real case study in the Mediterranean region of Türkiye, comprising three provinces with great potential: Antalya, Mersin, and Adana. An assessment of a complete set of environmental, economic, social, and technological criteria is performed using expert judgments stated in interval-valued neutrosophic language assessments. They were incorporated into a Geographic Information System (GIS) to produce a suitability map indicating the most suitable sites for the facility. The suggested approach is different from the traditional crisp or fuzzy MCDM techniques since it clearly models the degrees of truth, indeterminacy, and falsehood, thus providing a more detailed representation of the expert evaluations. According to the data, Mersin is the most ideal site for the construction of a solar power plant, followed by Antalya, and the least suitable site is Adana. The results suggest that sustainable solar energy planning must go beyond technical resource potential and include integrated and uncertainty-aware assessments. The suggested IVN-TOPSIS framework can serve as a powerful decision-support tool to policymakers, planners, and investors that wish to encourage regionally balanced and sustainable renewable energy development. Full article

Group Relative Policy Optimization (GRPO) streamlines the alignment of Large Language Models (LLMs) and Vision–Language Models (VLMs) by eliminating the Critic model. However, its efficiency heavily depends on the group size, G. While a larger G improves reward estimation and stabilizes the Advantage, $A_i$, it drastically increases VRAM usage and reduces throughput. Standard heuristics like a fixed G of 64 create significant bottlenecks in resource-constrained settings. This paper introduces an Efficiency-Aware optimization framework utilizing Multi-fidelity Bayesian Optimization and Hyperband (BOHB) to dynamically identify the optimal group size, $G^{*}$. The method uses a multi-objective function that balances reward accuracy, $A_i$ variance, and hardware utilization, applying z-score normalization. By employing Successive Halving to quickly evaluate candidates at low fidelity, the framework reduces search costs by up to 74% compared with random search. Tested across text-only LLMs (Qwen2.5-7B/1.5B) and multimodal VLMs (Qwen2.5-VL-3B), the framework demonstrates that the discovered $G^{*}$ saves up to 72.5% in VRAM compared with the baseline of 64, while maintaining reward accuracy within 5.8%. Sensitivity analyses on hyperparameters like $\lambda$, $\alpha$, and $\beta$ confirm the framework’s robustness. Rather than treating group size as a mere engineering heuristic, this study establishes a principled methodological advance by formalizing the trade-off between statistical estimation stability and hardware constraints into a unified optimization framework for resource-efficient RLHF. Full article

Nutrient stoichiometry, particularly the balance of carbon (C), nitrogen (N), and phosphorus (P), plays a fundamental role in regulating freshwater ecosystem dynamics, primary production, and biogeochemical cycling. This study presents one of the first dedicated reviews to combine bibliometric mapping with ecological synthesis of C:N:P ratios in inland waters, drawing on 1004 publications indexed in the Web of Science Core Collection (2000–2025), comprising peer-reviewed articles and review articles refined by document type, language, and research area. Bibliometric mapping using VOSviewer (version 1.6.20) identified exponential growth in publications after 2010, with phosphorus dynamics and eutrophication emerging as the most-cited themes, while recent years have shown increasing attention to C:P ratios as reliable ecological indicators. Four dominant thematic clusters were identified: Nutrient Cycling and Biogeochemistry; Phytoplankton and Food Web Dynamics; Eutrophication and Water Quality; and Climate Change and Ecosystem Responses. Ecological synthesis demonstrated substantial deviations from the canonical Redfield ratio (106C:16N:1P), with pronounced stoichiometric variability across trophic states, latitudes, and ecosystem types. Case comparisons revealed high C:P ratios in Arctic and alpine lakes linked to dissolved organic carbon inputs, low N:P ratios in tropical waters that promote cyanobacterial dominance, and stable, low phosphorus concentrations in deep African lakes. These findings emphasize the significance of flexible stoichiometry in predicting ecosystem tipping points, managing harmful algal blooms (HABs), and guiding nutrient restoration strategies. By integrating bibliometric and ecological evidence, this study identifies C:P ratios as a promising candidate indicator that merits further field validation for freshwater management, while underscoring persistent research gaps in microbial stoichiometry, cross-scalar modeling, and policy uptake in the Global South. Full article

The rapid scaling of large language model (LLM) training has intensified demand for Graphics Processing Unit (GPU) clusters balancing throughput with energy efficiency. While NVIDIA’s H100 and B200 architectures are increasingly deployed in production datacenters, their comparative behavior under distributed training remains insufficiently characterized beyond vendor specifications, leaving datacenter operators without empirical guidance on metrics such as TFLOPs/kW and tokens-per-kilojoule. This work presents a system-level evaluation of single-node 8× H100 and 8× B200 configurations using Distributed Data Parallel (DDP) training across LLMs and vision–language models (VLMs) ranging from 7B to 32B parameters, spanning various real AI workload scenarios. We benchmark end-to-end throughput, utilization, power, energy, TFLOPs/kW, and tokens-per-kilojoule, complemented by architectural analysis explaining observed behavioral differences. Across LLM workloads, B200 achieves higher utilization (1–6%), faster training (up to 15%), and greater compute efficiency (up to 32% higher TFLOPs/GPU), attributable to higher memory bandwidth and large streaming multiprocessor (SM) count. However, B200 exhibits lower TFLOPs/kW and tokens-per-kilojoule, revealing a fundamental trade-off: throughput gains come at a measurable energy cost per useful token. VLM results further expose model-dependent asymmetries, with B200 consuming disproportionately more energy for lighter compute kernels due to elevated baseline power draw. These findings provide an empirical framework distinguishing compute efficiency from energy efficiency across next-generation GPU nodes, offering practical guidance for energy-aware AI datacenter design. Full article

Background/Objectives: Accurate estimation of nutritional content from food images has important applications in dietary assessment and public health surveillance. While large language models (LLMs) have shown promise for this task, the effects of prompt design and model selection on estimation accuracy remain poorly characterized. Methods: We evaluated three Claude models (Haiku 4.5, Sonnet 4.6, Opus 4.6) for visual estimation of five mandatory nutritional components (energy, protein, fat, carbohydrate, and salt equivalent) across three datasets: NutriImage (691 Japanese meal photographs with dietitian-validated ground truth, after OCR-mask quality filtering), SNAPMe (1463 US meal photographs from a publicly available benchmark), and the Japan Branded Food Database (JBFD; 989–1000 packaged food product images). We systematically compared a default prompt and a visual estimation prompt explicitly instructing the model not to read any text or numbers visible in the image. Results: The visual estimation prompt substantially improved accuracy when paired with a sufficiently capable model (energy R²: 0.23 for Haiku to 0.60 for Sonnet, JBFD). Sonnet and Opus substantially outperformed Haiku across all datasets, while differences between Sonnet and Opus were small (MedAPE difference 1–3 percentage points). Packaged food images (JBFD) yielded higher R² than meal photographs. Salt equivalent showed consistently poor accuracy (MedAPE 34–64%). On SNAPMe, Sonnet achieved lower energy MAE (116.9 vs. 123.0 kcal, −4.9%) and lower MAE for protein (5.9 vs. 7.9 g, −25.7%) and fat (6.6 vs. 8.7 g, −24.5%) compared with a recent ChatGPT-5 study. Conclusions: Claude Sonnet offers the best cost-performance balance for LLM-based nutritional estimation. Prompt design substantially affects accuracy, but only when paired with a sufficiently capable model; model visual recognition capability appears to be a key determinant of performance. These findings highlight the inherent difficulty of this task and provide practical guidance for dietary assessment system development. Full article

Small and medium enterprises routinely face a paradox in financial monitoring: their accounting documents exist, but the cost of converting heterogeneous PDFs into timely financial signals is prohibitive without dedicated analytical staff or specialized software. This paper presents a two-layer artifact, designed under Design Science Research, that bridges this gap using only public-web large language models (LLMs) and a parsimonious multi-criteria decision routine. Layer 1 implements a structured LLM-driven workflow that extracts account–value pairs from annual tax balance sheets without code, APIs, or fine-tuning. Layer 2 reconstructs auditable accounting aggregates and ranks yearly financial condition through TOPSIS with min–max normalization—a deliberate replacement for classical vector normalization, which fails when profitability indicators are negative, as routinely occurs in distress years. To avoid size effects and algebraic redundancy, the decision matrix uses only three criteria spanning liquidity, profitability, and solvency. The artifact is demonstrated in a four-year case study of an anonymized construction SME (2021–2024), with accountant-verified document-level match rates of 0.810, 0.998, 0.950, and 0.909. Equal weighting is the only weighting configuration used; a supplementary entropy-based dispersion diagnostic yields the same ordinal ranking—2024 > 2023 > 2021 > 2022—and 10,000 Monte Carlo replications, with uncertainty injected at the reconstructed-aggregate level, confirm that the extreme ranks are invariant across all runs. The contribution is methodological and practical: a transparent, low-infrastructure pipeline that brings first-pass financial screening within reach of SMEs operating under severe data and budget constraints. Full article

Manipulated satellite imagery threatens analytic workflows, policy decisions, and trust in geospatial intelligence. Operational systems increasingly benefit from capabilities for both manipulation detection and manipulation-family attribution to support verification, triage, and downstream analysis. We present a unified benchmark for characterizing three representative manipulation families in geospatial imagery—generative manipulations, pixel-level perturbations, and adversarial patches—using a controlled, class-balanced design and 20 modern vision architectures spanning conventional, Earth-observation-pretrained, and vision-language models. Across architectures, the dominant failure boundary is between authentic imagery and subtle pixel-level perturbations, whereas generative manipulations and adversarial patches are generally more separable under matched in-domain conditions. Additional analyses reveal important generalization limitations under unseen manipulation variants and external-domain transfer, demonstrating that strong benchmark performance does not necessarily translate to reliable operational screening. The framework also enables systematic comparison of unified multi-attack and specialized detection strategies, providing insight into their relative strengths and limitations. Rather than proposing a new defense, this work provides a reproducible methodology for characterizing manipulation artifacts, model failure modes, and deployment-relevant screening behavior in geospatial imagery, with applications to analyst triage, verification workflows, and trustworthy use of satellite data. Full article

Zero-shot remote sensing is attractive for scene classification because new regions, sensors, and label taxonomies often appear before sufficient annotated data are available for supervised adaptation. We present OATS-RS, an inference-centric framework that keeps a remote sensing vision–language model (VLM) backbone frozen and improves zero-shot decisions through ontology-aware prompt construction, hierarchical and contrastive scoring, adaptive multi-view aggregation, unlabeled transductive refinement, ambiguity-aware local re-ranking, and selective prediction. The method targets the common remote sensing regime in which neighboring classes such as annual crop, permanent crop, forest, pasture, herbaceous vegetation, river, and sea or lake overlap strongly in red–green–blue (RGB) appearance, meaning that they require more than a single class-name prompt. On the supplied final EuroSAT RGB evaluation with a GeoRSCLIP Contrastive Language–Image Pre-training (CLIP)-family Vision Transformer Base with 32 × 32-pixel patches (ViT-B-32) backbone, the complete pipeline obtains top-1 accuracy of 0.522, balanced accuracy of 0.522, macro-averaged F1 score (macro-F1) of 0.535, and top-3 accuracy of 0.887. The strongest classes are industrial area, residential area, river, highway, and pasture, whereas the weakest classes remain herbaceous vegetation and several fine-grained vegetation categories. Selective prediction increases accepted-example accuracy to 0.538 at 0.934 coverage, but the expected calibration error (ECE) remains high at 0.384. These results support a qualified conclusion: ontology-guided zero-shot inference can already recover useful semantic shortlists for structured remote-sensing scenes, but fine-grained natural-class disambiguation, calibrated confidence, multi-dataset transfer, component-level ablations, and measured runtime remain essential before dependable deployment claims can be made. Full article

SWANP-AI (Smart Water Network Partitioning with Artificial Intelligence) is a web application with AI natively embedded in its core engines for automated Water Network Partitioning (WNP) of water distribution networks. It is presented as the web-based evolution of SWANP 4.0, whose computational routines have already been tested in operational and research applications. The paper clarifies the full development chain of the platform, from graph-based grouping of candidate District Metered Areas/pressure management Areas (DMA/PMA) to multi-objective boundary pipe optimization and operational decision support. The methodology combines spectral and multilevel k-way partitioning for district generation, NSGA-II for cost–resilience boundary selection, hydraulic simulation through EPANET/WNTR, and AI-supported modules for solution interpretation, sensor placement, natural language editing, and Bayesian leak localization. The application to a real water distribution network shows that SWANP-AI can transform natural language engineering requests into formal optimization tasks, identify hydraulically meaningful candidate interventions, and select balanced solutions through Utopia point analysis, thus reducing manual trial-and-error in DMA/PMA design. The main contribution is a structural generative AI workflow that supports engineers not only in analyzing a network as it is, but also in designing how the network should be partitioned and operated. Full article

Wearable technologies offer practical opportunities for assistive communication and educational support in introductory sign language learning. This paper presents an IoT-enabled smart glove for deterministic static sign language recognition over a bounded vocabulary of 15 isolated static gestures, comprising digits (0–9) and five vowel handshapes (A, E, I, O, U). The system is intended for foundational static gesture and posture practice and is not designed or validated for dynamic gestures, coarticulated signing, continuous sign language recognition, or sentence-level translation. The prototype integrates five 2.2-inch (55.9 mm) resistive flex sensors and an MPU6050 3-axis accelerometer, performs acquisition, exponential moving average filtering, user-specific calibration, normalization, and deterministic classification on a NodeMCU ESP32 board, and transmits selected processed variables to Arduino Cloud through MQTT for remote monitoring. A 10 s calibration routine maps user-specific open-hand and closed-fist responses into normalized flex-sensor ranges, allowing the same deterministic rule structure to operate across participants without model retraining. Experimental evaluation with 10 healthy adult participants aged 20–41 years (mean age: 27 years), all familiar with sign language and all providing written informed consent, produced a balanced dataset of 1500 labeled steady-state sensor vectors. The class-averaged recognition rate was 92.8%, and leave-one-subject-out validation produced a subject-wise accuracy of $92.80\pm 2.03\%$, with individual participant accuracies ranging from 90.00% to 96.00%. The local embedded processing pipeline required less than 2 ms per cycle, the complete path including MQTT visualization produced approximately 150 ms end-to-end latency, and the device operated for up to 14 h using a 3.7 V, 1000 mAh Li-Po battery. The results indicate that calibrated deterministic sensor fusion can provide a low-cost, low-latency, edge-executed solution for bounded static sign-language gesture learning tasks while maintaining stable short-term subject-wise performance under controlled experimental conditions. Full article

Generative Artificial Intelligence (GenAI) in cardiac surgery refers to the integration of advanced computational models, such as Large Language Models (LLMs), to automate and enhance clinical decision-making, preoperative risk assessment, and surgical education. In the context of surgical training, it functions as a personalized pedagogical tool that supports various learning activities, ranging from information acquisition and clinical inquiry to procedural practice, while requiring rigorous human oversight to ensure patient safety and clinical accuracy. (1) Background: Generative Artificial Intelligence (GenAI) is increasingly integrated into health professions education, offering new opportunities for learning; however, its specific application and pedagogical mapping in high-stakes fields such as cardiac surgery remain underexplored. This systematic review investigates how GenAI is utilized in cardiac surgery and surgical education, aligning these uses with Laurillard’s six learning types. (2) Methods: Following the PRISMA 2020 guidelines, we searched the Web of Science Core Collection for studies on GenAI in cardiac surgery, resulting in 42 studies that met the inclusion criteria. Study quality was appraised using the Medical Education Research Study Quality Instrument (MERSQI). (3) Results: GenAI applications most frequently supported clinical inquiry (93.8%) and practice (68.8%), demonstrating expanding efficiency across commercial and open-source models (including ChatGPT-4o, Gemini AI, and emerging reasoning architectures such as DeepSeek) for knowledge acquisition and medical production. While it significantly improves individualized learning and preoperative assessment workflows, its practical role in Discussion and Collaboration remains heavily underutilized, highlighting a distinct shift toward individualized solo professional workflows. (4) Conclusions: GenAI provides a transformative and scalable approach to cardiac surgical training by offering personalized and accessible knowledge retrieval. However, clinical educators and governance bodies must deliberately balance these immediate productivity benefits with long-term concerns regarding structural “hallucinations,” data verifiability, and the preservation of collaborative competencies within modern multidisciplinary Heart Teams. Full article

Background/Objectives: A current debate addresses where syntactic Merge primarily resides: the left-hemisphere posterior inferior frontal gyrus (IFG) or the temporo-parietal cortex. For proponents of the former, the temporo-parietal cortex supports more effortful processing; for the latter, the IFG supports integration and conflict resolution. We examine aperiodic activity in processing wh-filler-gap dependencies in French for evidence from network dynamics addressing engagement in syntax across L1 and L2. Methods: We extracted aperiodic activity 1/f components (considering offsets as a reflection of neuronal spiking and exponents as a reflection of excitatory–inhibitory balance) out of power spectrum density at 0.5–40 Hz across occipital and bilateral frontal and temporo-parietal regions of interest (ROIs) in reading. Results: Greater exponents arose in temporo-parietal than frontal ROIs in L1 and L2, with strong spiking and regulation suggested by greater offsets and exponents in the occipital ROI in L2—unlike L1—and with potential modulation by L1–L2 representation overlaps. These patterns suggest distributed cell assemblies for L1 and L2 processing. Increased regulation in temporo-parietal ROIs across L1 and L2 cell assemblies might suggest a structural function across temporo-parietal cortices in syntactic processing. Conclusions: Aperiodic activity reflecting connectivity in L1 and L2 processing supports distinct L1 and L2 cell assemblies, with L2 patterns suggesting potential overlap between L1 and L2 circuit modules. Greater exponents in bilateral temporo-parietal ROIs across L1 and L2 indicate increased regulation, supporting the engagement of lateralized temporo-parietal cortices in computations. These effects are discussed by considering advances in syntactic theory and the biology of language readiness. Full article

Background: Falls and fall-related injuries are common in older adults and are frequently accompanied by fear of falling (FoF), which may lead to activity avoidance and functional decline. Because many interventions target falls or FoF in isolation, we conducted a systematic review and meta-analysis of randomized controlled trials (RCTs) to identify, describe, and evaluate interventions reporting both falls and FoF outcomes in older adults. Methods: This systematic review and meta-analysis were registered in PROSPERO (CRD420251113137) and conducted in accordance with PRISMA guidelines. PubMed, Embase, and Web of Science were searched from inception to 4 November 2025. Eligible studies were English-language RCTs that included adults aged ≥60 years, evaluated nonpharmacological interventions, and reported both FoF and falls. Methodological quality was assessed using the PEDro scale. Random-effects meta-analyses were performed for FoF (Hedges g), and Bayesian random-effects binomial meta-analyses were conducted for falls. Results: Ten RCTs published between 1998 and 2018 (sample sizes per trial: n = 27–540) were included. Interventions included cognitive–behavioral therapy-based programs, Tai Chi, physiotherapist-led strength and balance training, computerized visual feedback, and video-guided home exercise. PEDro scores ranged from 6 to 9 (mean, 7.7). Pooled analyses showed no significant effect on FoF at the end of intervention (g = −0.20, 95% CI −1.45 to 1.05; p = 0.68; high heterogeneity) or at follow-up (g = −0.14, 95% CI −0.60 to 0.33; p = 0.50). For falls, postintervention evidence favored the null (BF₁₀ = 0.16; pooled estimate −0.01, 95% credible interval [CrI] −0.30 to 0.14). Follow-up results were inconclusive (BF₁₀ = 2.07; pooled CrI −0.56 to 0.00), with substantial uncertainty. Conclusions: Across RCTs that measured both outcomes, interventions did not consistently improve both FoF and falls outcomes. These findings may suggest a partial dissociation between psychological and physical fall-related outcomes, highlighting the need for integrated, adequately powered trials that utilize standardized measures and longer follow-up periods. Full article

Vision-and-language navigation (VLN) traditionally relies on explicit reasoning chains, which, despite being interpretable, impose severe constraints on inference efficiency and scalability in long-range environments. Existing multimodal large language models (MLLMs) frequently encounter latency bottlenecks due to the generation of verbose textual narratives during decision-making. To address these limitations, we propose spatial reasoning vision-and-language navigation (SR-VLN), a novel framework that shifts the paradigm from explicit chain-of-thought (CoT) to an implicit spatial representation space. SR-VLN introduces a pyramidal hierarchical history framework integrated with perceptual compression to condense historical trajectories into multi-scale representations, effectively minimizing token overhead while preserving critical spatial semantics. Rather than generating verbose textual reasoning steps, SR-VLN employs compact, learnable spatial tokens (S-Tokens) to perform agile inference directly within the latent feature space. To establish robust causal mappings between these implicit states and navigational actions, we employ a hybrid training strategy that combines sparse reward supervision with reinforcement learning via GRPO. Extensive evaluations on the R2R, REVERIE, and SOON datasets demonstrate that SR-VLN achieves state-of-the-art overall navigation performance, while maintaining a comparable balance between accuracy and efficiency. Compared to explicit reasoning baselines, our method reduces token consumption by 68% and achieves a 4.1× speedup in inference while reaching a 76.02% success rate and a 73.80% SPL on the R2R unseen split, thereby facilitating near-real-time action prediction in long-range navigation environments. Full article