Search Results (771)

Background/Objectives: Mosaic speech was used to further investigate the auditory system’s temporal resolution needed for speech intelligibility. Mosaic speech is a form of degraded speech segmented in frequency × time blocks with no discernible temporal fine structure and with degraded amplitude envelope cues. Methods: We performed a listening experiment with mosaic speech consisting of 20 frequency bands segmented into 20, 40, 80, 160, and 320 ms. Younger listeners (<25 years; n = 20), with self-reported normal hearing and having passed a limited screening test, and elderly listeners (>65 years; n = 19) with hearing thresholds ranging from normal hearing to moderate hearing loss listened to Japanese low- and high-familiarity mosaic words and wrote down what they heard. The original words were included as control stimuli. Results: Although younger listeners had significantly higher intelligibility scores, elderly listeners could integrate and parse coarse blocks of mosaic speech of 20 ms and 40 ms with intelligibility scores of 84% or higher for high-familiarity words. For longer block durations, however, the elderly listeners’ intelligibility dropped rapidly for both high- and low-familiarity words. In both the young and the elderly listener groups, intelligibility reached the floor for block durations of 160 and 320 ms. Word familiarity strongly affected intelligibility scores. For blocks up to 80 ms, intelligibility was significantly higher for high-familiarity words than for low-familiarity words in both age groups, with a 10–30% difference. Conclusions: Elderly listeners (n = 19) with normal hearing to moderate hearing loss could maintain relatively high intelligibility for mosaic words segmented in blocks of 20 or 40 ms, provided the words had a high familiarity level. Full article

Selecting a sensing architecture for knee osteoarthritis (OA) screening requires balancing biomechanical information, system complexity, and reproducibility. We audited a public Korean multimodal gait dataset and performed a leakage-resistant internal evaluation. The release contained 180 participants (90 normal, 90 knee OA) measured using a smart insole, instrumented gait mat, and inertial measurement units (IMUs); all 1080 JavaScript Object Notation (JSON) files were checked for structural, value, provenance, and duplication errors. The primary benchmark was a fixed class-balanced L2 logistic regression model using nine gait mat variables, evaluated with subject-level repeated stratified five-fold cross-validation and 10,000 outcome-stratified bootstrap resamples. The audit identified 14 source-path metadata errors and one opposing-label duplicate smart insole payload, but no parsing, schema, range, or cross-partition subject errors. The gait mat model achieved an area under the receiver operating characteristic curve (AUROC) of 0.924 (95% confidence interval [CI], 0.879–0.962), balanced accuracy 0.883 (0.833–0.928), sensitivity 0.856, specificity 0.911, and Brier score 0.102. Adding smart insole and/or IMU features did not improve AUROC. Provider-model reproduction was descriptive because the public Validation partition informed model selection. In this release, the compact gait mat feature set provided the most favorable observed balance of discrimination, interpretability, and sensing complexity; external prospective evaluation is required before clinical use. Full article

This Data Descriptor presents a georeferenced dataset of electromagnetic-field exposure measurements collected in Colombia between 2 November 2023 and 30 June 2024. The records originate from the Sistema de Monitoreo de Campos of the Agencia Nacional del Espectro (ANE), which uses isotropic probes to monitor broadband radiofrequency electromagnetic fields from 100 kHz to 8 GHz and reports six-minute averages of incident power density in W/m². The comma-separated value file contains 1,286,346 timestamped records and seven source variables, including Well-Known Text (WKT) point geometry. The principal 2024 subset comprises 995,238 records from 24 monitoring stations. We provide a reproducible workflow for timestamp and coordinate parsing, structural and numeric validation, station-level aggregation, and spatial sensitivity analysis using inverse distance weighting with the mean, median, and 95th percentile. The dataset does not provide frequency-resolved measurements, calibration certificates, or measurement-uncertainty budgets; these omissions limit the use of the public file for formal regulatory-compliance assessment. The accompanying repository includes validated station-level summaries, descriptive tables, figures, and reproducible R code. The package supports environmental monitoring, geospatial analysis, methodological comparison, and reproducible reuse of electromagnetic-field exposure data. Full article

Machine learning (ML) models trained on bulk-crystal descriptors are increasingly used to prescreen catalysts by predicting adsorption energies, yet reported performances often rely on random K-fold cross-validation that permits the same bulk formula to appear in both training and test sets. We construct a reproducible benchmark that fuses 936 CatApp DFT adsorption energies with bulk descriptors from the Materials Project for H*, O*, and OH* on metal and alloy surfaces. We compare random K-fold cross-validation with GroupKFold grouped by parsed formula, the latter mimicking the realistic task of predicting adsorption on entirely new catalyst compositions. Under formula-grouped evaluation, random CV materially overestimates apparent generalization performance, with the largest and most robust effects for H* and OH* (protocol-inflation gaps up to approximately 0.8). The H* and OH* results are based on only 20 and 31 unique formulas, so their GroupKFold Spearman point estimates should be read as directional evidence rather than quantitative estimates. O* shows a smaller and statistically fragile protocol-inflation signal and, even where composition-plus-bulk features improve Random Forest and Ridge, the usable signal is best described as a very coarse pre-filter within a limited domain. Bulk descriptors are adsorbate-dependent: they improve O* prediction for Random Forest and Ridge, but degrade H* and OH*—a qualitative, directional observation given the small formula counts—whose binding is poorly captured by bulk crystal descriptors, consistent with the established view that it is governed by surface-localized electronic structure. These results outline a realistic performance boundary for bulk-to-surface ML in this benchmark: O* can be very coarsely prioritized from bulk descriptors within a limited domain, whereas H* and OH* are unlikely to be quantitatively predicted from bulk descriptors alone and would benefit from surface-aware models. We therefore recommend that bulk-to-surface adsorption-energy benchmarks report formula-grouped cross-validation alongside random cross-validation as a more robust and transparent practice. Full article

This paper takes up the important questions whether (i) machines—by means of software—can translate texts, thereby imitating translations made by humans, and (ii) publications in the fields of scientific domains and engineering can be translated by machines to an almost perfect level. This paper examines some issues in this context, among them the question if software can be compared with the minds processed by human brains, if machines handle information or something else, if knowledge is found in software (including some deliberations on what knowledge is), and if the latest software developments are what they declare. This paper concludes that machines may assist translators but that they will not, by principle, be able to reach an almost perfect level. Accordingly the specific aim (I shall not label it objective) is to point to theoretical and analytic (parsing) obstacles to the idea that machines can produce acceptable translations in all contexts. I demonstrate that certain combinations of words will present systems with insurmountable challenges; challenges which also present human translators with questions that have almost unattainable answers. Therefore I object to the huge amount of money spent on software development for systems with that objective. I also point to the option of developing more relevant alternatives, including simpler solutions. If you ask for new insights—like you would do when perusing a research paper in order to update your professional frame of reference—there are, basically, no new insights in this theoretical scientific article. I just make it transparent what all linguists know, i.e., that in the field of machines handling language, less focus should be on what electronic systems can, or cannot, do and more focus on what details in language, and languages, will make certain kinds of handling particularly arduous; to the verge of being insurmountable. Full article

Generative named entity recognition (NER) systems must identify clinical concepts and map them to exact source spans. We investigated how span-reference design affects this mapping and whether failures arise from recognition or source localisation. Four representation-and-grounding pipelines—inline XML, mention-list JSON, a tab-separated mention list, and direct-offset JSON—were compared for disease, procedure, and symptom recognition in a public Spanish clinical case-report collection. Two instruction-tuned 3-billion-parameter models were fine-tuned for each task, and outputs were evaluated for syntactic validity, parsing, grounding, and official strict-span performance. Inline XML, mention-list JSON, and the tab-separated format achieved F₁ scores of 0.637–0.736, whereas direct-offset JSON achieved 0.001–0.005. With the same unadapted checkpoints, zero-shot F₁ was at most 0.131 and three-shot F₁ at most 0.386; three-shot prompting improved the mention-list outputs, while inline XML and direct offsets remained near zero. Direct-offset outputs usually contained relevant mention text but incorrect character positions. On a separate 75-document confirmation partition, grounding the emitted strings recovered F₁ of 0.603–0.701, while replacing absolute offsets with mention-occurrence numbers achieved 0.652–0.740. Without further training, the external CARMEN-I evaluation on 458 hospital-record sections yielded F₁ of 0.610–0.659 for string-grounded or occurrence-index outputs, compared with 0.113–0.177 for inline XML and at most 0.002 for direct offsets. These results show that output validity alone does not establish usable span extraction. Separating recognition from source localisation identifies where an otherwise valid generation fails, while occurrence-based references provide a more reliable alternative to absolute character offsets under exact matching in the tested settings. Full article

The semantic interpretation of remote sensing imagery through segmentation has become indispensable for a wide range of applications, including resource exploration, environmental assessment, and land-use analysis. Yet, accurate parsing of such images remains challenging because complex object boundaries and large scale differences often weaken the ability of conventional Convolutional Neural Network (CNN)-based methods to preserve local details. In response, this study constructs a segmentation framework that couples wavelet convolution with the Mamba architecture. To strengthen feature learning in the intermediate stages, an Auxiliary Segmentation Module (ASM) is employed to provide additional supervisory guidance, which supports optimization and encourages the representation of subtle semantic details. Wavelet-transform convolution is also introduced into the downsampling path, enabling spatial cues and frequency-related information to be exploited in a more coordinated manner for finer boundary and texture modeling. Experiments on public remote sensing datasets and mining area imagery further confirm the effectiveness of the method. Compared with several existing segmentation approaches, the proposed model delivers better overall performance in mIoU, F1-score, and recognition accuracy, particularly in scenes where multiple land-cover categories are heavily interlaced. Moreover, these gains are obtained with relatively low model complexity, suggesting good potential for practical deployment in land monitoring and ecological management. Full article

To address the challenges of complex multi-objective trade-offs, tightly coupled physical constraints in reservoir dam safety dispatching, and fulfill the significant cognitive gaps in human–machine interaction, a framework with four deep cognitive layers and a physical computation layer is proposed which integrates large language models (LLMs) with multi-agent collaboration. The framework stratifies cognitive intelligence into interface translation, strategic cognition, tactical reasoning, and operational understanding layers; performs computation in the physical computation layer; and achieves deep coupling among agents in different layers through the Blackboard information sharing mechanism. The physical computation layer consists of the gate-opening discharge, water-level storage capacity, runoff and inflow, downstream risk calculation agents and a Pareto multi-objective optimizer to realize non-dominated sorting of multi-dimensional objectives encompassing dam safety, ecological loss, downstream risk, and operational complexity. Illustrative case analysis indicates that this framework can effectively parse user requirements expressed in natural language, generate dispatching schemes conforming to physical constraints, achieve error control and quantify the downstream risk. This research provides a scalable framework for the implementation of intelligent reservoir dispatching and can enhance the intelligence of digital twins of river basins. Full article

The widespread penetration of generative artificial intelligence is reshaping adolescents’ social ecosystems and emotional experiences, while challenging the interpretive boundaries of traditional connectedness theories. Following the logical path of “connotation reconstruction–extension transformation–concept construction”, this study integrates the ethics of care and neo-ecological theory to systematically construct a theoretical framework of artificial intelligence connectedness. First, tracing theories across philosophy, sociology, and psychology, this research reconstructs the core connotation of connectedness rooted in the ethics of care, proposes the continuum hypothesis of caring relationships, and clarifies AI’s unique position on this continuum. Second, from the neo-ecological perspective, this paper sorts out the extended structure of connectedness and demonstrates the ecological shifts brought by the rise of virtual microsystems and the entry of AI actors. On this basis, the study formally defines artificial intelligence connectedness and establishes its three-dimensional structure: demand identification, two-way behavioral engagement, and responsive confirmation. Through systematic comparison with adjacent concepts, this paper identifies its uniqueness and positions it as a specific subtype of connectedness for the digital era. It defines it as a perceived bond that is both psychologically real and ethically asymmetric, carrying asymmetric risks under particular usage conditions and design logics. Finally, this paper builds a dual interpretive framework integrating traditional connectedness and artificial intelligence connectedness, verifies its incremental validity and unique predictive power, and puts forward falsifiable research propositions. This study expands the boundary of connectedness theory and provides an integrated analytical framework for parsing the complex mental health mechanisms of adolescents in the digital age. However, it should be noted that the artificial intelligence connectedness proposed in this study is currently a theoretical construct; its scientific validity and applicability remain to be verified through the development of standardized measurement tools and systematic empirical research. Full article

Model-based systems engineering (MBSE) takes Systems Modeling Language (SysML) as the industrial standard modeling language, yet cloud Large Language Model (LLM)-based SysML generation faces limited domain data, model hallucinations, high hardware cost and confidential data leakage risks. This paper builds a 914-sample SysML PlantUML corpus and proposes a fully offline lightweight framework based on Qwen2.5-Coder-7B-Instruct, integrating 4-bit NF4 Quantized Low-Rank Adaptation (QLoRA) fine-tuning, vector-free Jaccard same-diagram reference retrieval and a three-round syntax correction loop. PlantUML executes syntax parsing while Graphviz only renders layouts. Tested on 131 samples covering five structural and behavioral SysML v1 diagram types, the plain-prompt baseline achieves word-set semantic F1 of 52.94%, and the retrieval-enhanced variant lifts the zero-retry syntax pass rate from 92.37% to 99.28%, with F1 slightly dropping to 50.64%. Running fully local without cloud data transmission, this pipeline offers a privacy-safe lightweight solution for SysML PlantUML modeling and does not support SysML-exclusive requirement or parametric diagrams. Full article

Consistency verification of drilling reports is critical for engineering quality control because a single data item may be distributed across reports with different formats, units, and page structures. Existing retrieval-augmented generation methods remain sensitive to retrieval and parsing errors in such documents, whereas ultra-large models impose substantial local computing and memory costs. This study proposes a lightweight tool-augmented framework based on dual-teacher distillation and sparse mixture-of-experts (MoE) modeling. Qwen3-235B-A22B serves as the primary teacher and Qwen3-30B-A3B as the assistant teacher. Their tool-use and task-planning capabilities are transferred to a sparse MoE student upgraded from a Qwen3-1.7B dense backbone through trajectory pruning, sample decomposition, and token-level Kullback–Leibler (KL) distillation. The student adopts an eight-expert Top-2 routing architecture. Experiments on 1000 drilling reports containing 30,127 verification instances show an F1 score of 58.0 ± 0.5%, with file-level, location-level, and exact-match accuracies of 66.5%, 55.2%, and 45.0%, respectively. The model contains 9.1B total parameters and 2.8B activated parameters, and reaches a latency of 12.1 ms per forward pass and a memory footprint of 18.4 GB under bfloat16 (BF16) precision. The reported F1 score characterizes the end-to-end verification task rather than an autonomous safety decision capability. The framework is intended to support evidence localization, anomaly prioritization, and expert review in local deployment settings. Full article

Automatic semantic parsing of architectural floor plans provides a two-dimensional semantic layer for building information modeling (BIM)-related workflows, intelligent plan checking, renovation, and large-scale drawing management. Unlike natural images, floor plans contain sparse textures, dense linework, small architectural symbols, and strong axis-aligned geometric regularities. These properties make conventional segmentation networks vulnerable to small-symbol dilution during downsampling, noisy skip-feature fusion, and fragmented predictions along long wall boundaries. Inspired by principles of hierarchical visual processing, selective attention, and spatial encoding, this study presents PCP-Net, an end-to-end Planar Component Parsing Network for room, icon, and boundary-aware floor plan parsing. PCP-Net uses a Grouped Residual Encoder to extract multi-scale local patterns, a Cross-Scale Feature Recalibration (CSFR) pipeline to recalibrate skip features through Adaptive Channel Gating, Spatial Response Amplification, and Axis-Decompositional Positional Embedding, and a Structural Gradient Propagation branch to provide training-time boundary regularization. Experiments on CubiCasa5K and three external datasets (R3D, CVC-FP, and ROBIN) evaluate PCP-Net under in-domain and zero-shot cross-domain protocols. On CubiCasa5K, PCP-Net attains 71.3% mIoU, 90.1% overall accuracy, and 78.9% mean accuracy; in the current ablation setting, ADPE improves mIoU by 0.8 percentage points over the ADPE-ablated configuration. These results indicate that cross-scale feature recalibration with axis-aware positional cues can improve floor plan semantic parsing within the evaluated datasets and pixel-level metrics, while downstream BIM generation still requires additional vectorization, topology graph construction, and attribute extraction. Full article

Background: Ochratoxin A (OTA) is a widespread mycotoxin with well-established nephrotoxic, immunotoxic, and carcinogenic properties. Despite decades of research, the structure, thematic evolution, and emerging directions of the scientific literature on OTA toxicity remain incompletely mapped. Methods: A bibliometric analysis was conducted using 883 publications indexed in the Web of Science Core Collection (1965–2025). Bibliometrix, VOSviewer, and BibExcel were integrated within a comparative cross-software workflow to assess publication trends, collaboration networks, thematic evolution, and thematic research directions in in vitro and in vivo OTA toxicity studies, while enabling cross-validation of the main bibliometric outputs. Results: After normalisation, authors, journals, author keywords, and Keywords Plus showed strong cross-software agreement, whereas country and institution outputs retained software dependent differences related mainly to country aggregation and affiliation parsing. Publication output increased substantially over the analysed period, particularly during the most recent decade, reflecting the growing visibility of OTA research in food safety and toxicology. The field is highly collaborative and multidisciplinary. Oxidative stress-related terms were prominent in the keyword and thematic analyses and co-occurred with terminology related to apoptosis, DNA damage, and mitochondrial dysfunction. Thematic evolution analyses showed a transition from early studies focused on nephrotoxicity and animal models toward more recent investigations addressing molecular pathways, cellular responses, and microbiota host interactions. Comparatively less prominent or emerging bibliometric themes included the gut immune axis, co-exposure to multiple mycotoxins, and the broader representation of animal-health and productivity-related research. Conclusions: The bibliometric and science mapping analyses indicate that the literature on OTA toxicity has evolved from predominantly organ, and animal model-related research, toward increasing attention to molecular, cellular, intestinal, and microbiota-related topics. These patterns describe changes in the conceptual structure of the literature rather than direct evidence of biological causality. Comparatively, less prominent themes, including human-relevant models, combined exposure scenarios, and microbiome-related research, may warrant further investigation. Full article

A variable-to-variable (V2V) length code parses a source sequence into phrases of variable length and maps each phrase to a binary codeword of, generally, a different random length. After encoding n phrases, the realized compression ratio $R_n={\Lambda }_n/{\Sigma }_n$—total codeword length over total source-symbol count—is the finite-sample counterpart of the code’s asymptotic rate $\rho$, to which it converges only as $n\to \infty$. This paper first derives exact formulas for all integer moments of $R_n$ for a given discrete memoryless source (DMS). Specifically, we obtain a closed-form formula for every moment $\mathbb{E}\left\{R_n^k\right\}$ as a one-dimensional integral involving only single-phrase moment generating functions of the pair $(L,\mathcal{l})$—the phrase length, in source symbols, and codeword length, in bits. From these moments we derive an Edgeworth approximation to the cumulative distribution function (CDF) of $R_n$ that is substantially more accurate than the central limit theorem (CLT) approximation. Using the Laplace method of integration, we also derive explicit closed-form formulas for the bias constant $C={\lim }_{n\to \infty }n(\mathbb{E}\left\{R_n\right\}-\rho )$ and for the variance constant ${\lim }_{n\to \infty }n·\mathrm{Var}\left\{R_n\right\}$. The analysis extends to Markov sources via state-indexed matrices with a redundancy formula obtained in closed form. On the coding-theoretic side, we cast V2V length codes as finite-state encoders and apply a generalized Kraft inequality for a compression-rate lower bound, and give a structural decomposition of the bias coefficient that separates cleanly across variable-to-fixed (V2F) length codes, fixed-to-variable (F2V) length codes, and V2V length codes. Applied to the Khodak code of Bugeaud, Drmota, and Szpankowski, this decomposition shows that its improved performance is reflected in its smaller bias constant. Full article

Pointer gauges remain widely deployed in industrial environments because they are inexpensive, resistant to electromagnetic interference, and readable from a distance. However, automatic reading remains difficult in practice because reliable prediction requires jointly interpreting pointer geometry, scale layout, and unit-type consistency under challenging conditions such as glare, scratches, blur, and oblique viewpoints. Although multimodal large language models (MLLMs) offer a promising unified interface for visual understanding and structured output, their direct application to gauge reading is limited by weak geometric grounding, unit confusion, and unstable numeric generation. Rather than claiming a new model architecture or a new reading algorithm, this work frames the contribution as a practical adaptation and evaluation framework for applying existing MLLM and PEFT components to structured industrial gauge reading. Our framework combines three components: (i) a dedicated dataset and VQA-style annotation protocol covering multiple noise types and intensity levels; (ii) a unified screening pipeline for selecting a suitable MLLM backbone under zero-shot settings; and (iii) parameter-efficient adaptation of the selected model with Projector-LoRA, together with training and decoding mechanisms designed to improve reading robustness and output consistency. On our test set, the fine-tuned Granite-Vision 3.2 model achieves 99.9% type accuracy, 43.77% reading accuracy, and 43.60% joint accuracy. It obtains an MAE of 3.98 over valid numerical predictions, a parsing coverage of 98.65%, and an all-sample penalized normalized MAE of 0.052. These results substantially outperform the evaluated zero-shot MLLM baselines in structured prediction accuracy, although the lightweight CNN baseline remains slightly better in all-sample normalized numerical error.These results should be interpreted as evidence of promise and measurable improvement over untuned MLLMs, not as evidence that the system is already sufficient for safety-critical or fully autonomous industrial deployment. More broadly, the proposed framework offers a practical, traceable path for adapting large multimodal models to visual measurement tasks that require structured numerical outputs. Full article