Search Results (945)

Eye tracking is a valuable behavioral signal for human-centered AI, yet the reliability of feature-based machine-learning models for inferring task type and workload across users and tasks remains uncertain, because experimentally defined workload labels may reflect task type and visual structure as much as cognitive demand. The practical problem is that designers of gaze-adaptive systems need to know which inferences are dependable enough to act on, and reported accuracies alone do not answer this, because the choice of prediction target and validation split can determine the result. This study systematically evaluates feature-based machine-learning models with post hoc explainability across three prediction targets: task type, binary load-versus-rest, and three-level workload. Eye-movement features derived from fixations, saccades, pupils, and blinks were extracted from short temporal windows collected from 54 participants performing attention, visual-spatial, and memory tasks under rest, easy, and difficult conditions, and evaluated using leave-one-subject-out (LOSO) and leave-one-group-out (LOGO) validation. Task type was classified most reliably (85.9% LOSO, 83.4% LOGO), binary load-versus-rest showed moderate, validation-sensitive robustness (81.4% LOSO, 63.9% LOGO), and three-level workload classification was substantially more challenging (56.4% LOSO, 44.3% LOGO). SHAP and statistical analyses consistently identified fixation dispersion, pupil-related measures, and subject-normalized features as the strongest contributors across all three targets. These findings show that prediction target definition, validation strategy, and post hoc explainability jointly determine what can be reliably inferred from gaze-based machine-learning models. Eye tracking alone therefore appears promising for task-type recognition and may support coarse engagement-related inference when the deployment task family is represented during model development, whereas task-independent fine-grained workload estimation remains unsupported by the present evidence. Full article

This study investigates the effects of different graded crushing durations in a blade crusher on the crushing mechanics of wheat and the properties of the resulting flour. Mechanical models were established for blade–particle collisions, radial sliding of particles along the blade surface, and particle–chamber wall collisions. Under reasonable simplifying assumptions, the models analytically characterize the theoretical relationships of impact force and crushing energy with blade rotational speed, rotational radius, and particle incidence angle. The models were used to provide a qualitative mechanistic interpretation of the experimental trends rather than to quantitatively predict flour particle size distribution or damaged starch content. Two graded crushing processes were evaluated, with crushing durations of 10 s per pass (F10) and 15 s per pass (F15). Observation of particle-size evolution during the crushing of wheat particles showed that as the number of crushing passes increased, the proportion of coarse particles continuously decreased, the proportion of fine particles gradually increased, and the proportion of intermediate-sized particles initially increased and then decreased, demonstrating a progressive coarse-to-fine fragmentation pattern. Particle size analysis of the resulting wheat flour showed that the particle size distribution for the F15 process peaked below 5 μm and shifted toward smaller particle sizes relative to that for the F10 process. Nevertheless, the wheat flour obtained from both processes exhibited relatively concentrated particle size distributions, with Span values ranging from 2.46 to 2.68. Damaged starch content increased significantly with the number of crushing passes and was generally higher for the F15 process than for the F10 process. Moisture content decreased from 14.30% to 12.86% under the F10 process and from 14.25% to 12.73% under the F15 process, whereas ash content ultimately increased to 0.48% under both processes. Protein content initially increased and subsequently decreased under both processes. These findings provide experimental evidence for the effects of graded milling on grain refinement, starch damage, and physicochemical composition of wheat flour. Full article

In this paper, we present deep learning models for classifying ancient-coins using transfer learning applied to a custom dataset curated from a specialized collection. The dataset was meticulously developed through rigorous preparation and quality screening; the retained fine-grained classes are imbalanced. We initially performed binary classification before progressing to separate fine-grained assessments of DenseNet121, EfficientNetB0, EfficientNetV2S, and ResNet50. In confirmatory experiments using a physical-coin-aware 70/15/15 split and five training seeds, the coarse CITY-versus-NABATAEAN task remained approximately perfectly separable. Fine-grained performance was lower: ResNet50 achieved 68.28 ± 2.45% CITY accuracy, whereas DenseNet121 achieved 72.89 ± 2.79% NABATAEAN accuracy. These results show that near-perfect coarse classification does not by itself imply reliable fine-grained attribution. This analysis underscores the importance of meticulous dataset preparation and strategic model selection in optimizing performance. A matched DenseNet121 ablation further showed that ImageNet initialization substantially improved the difficult fine-grained tasks compared with random initialization. The proposed framework supports digital documentation and analysis of ancient coin collections, while the current single-collection evaluation still limits claims of cross-collection generalization. Full article

Ultra-thin metal bipolar plates are critical components of proton exchange membrane fuel cells (PEMFCs), and their forming characteristics decisively influence service performance. This study proposes a constitutive model incorporating size effects to elucidate how sheet thickness and grain size govern stress–strain responses and formability of ultra-thin plates. The verified model is employed in finite element analysis for formability of ultra-thin plates. Based on the results of simulation, key stamping die parameters were optimized using Random Forest and XGBoost surrogate models. Results indicate that increasing grain sizes reduces grain boundary density, leading to stress localization within coarse grains and promoting local thinning. This effect increases stored elastic energy and simultaneously raises the maximum stress, thinning rate, and springback angle. Conversely, the increasing sheet thickness strengthens triaxial constraint and raises forming stress, while suppressing thinning and springback through enhanced strain redistribution and plastic dissipation. Thus, grain coarsening degrades formability overall, whereas increasing thickness introduces a trade-off between higher forming stress and improved dimensional stability. Both surrogate models demonstrated high predictive accuracy on unseen data (maximum error is 2.01%), identifying a non-standard parameter combination (α = 16.0°, R = 0.30 mm, h = 0.48 mm, W = 1.46 mm, S = 0.73 mm) that yields a thinning rate of 4.43% and a springback angle of 0.151°, a level of precision that is difficult to achieve using conventional orthogonal experimental design. This result was verified by additional finite element simulations. The proposed constitutive model and optimization approach provide a theoretical framework and practical guideline for micro-scale bipolar plate die design. Full article

Landslide-induced surges and subsequent dam breaching constitute severe cascading hazards in mountainous gorges. Conventional steady-flow sediment theories fail to describe these extreme, unsteady processes, and existing research focuses on wave propagation rather than surge-driven erosion mechanisms. Using the Baige landslide dam as a prototype, this study combines 1:100 physical model tests with CFD-DEM simulations to investigate how landslide fall height, water depth, and sediment gradation govern surge propagation, dam scour, and sediment transport. The results show the surge amplitude reaches 38.21 cm under high-fall, deep-water conditions and decays nonlinearly. Fine-grained beds exhibit suspended-load transport (max concentration 15.2%), whereas coarse-grained beds develop scour pits via bedload transport, with deposition volume increasing ~230%. Sediment transport follows a three-stage spatial pattern: intense erosion near the dam (max depth 2.9 cm), grain-size-sorted deposition in the middle reach (max height 4.6 cm), and fine-sediment accumulation downstream. The numerical results agree well with experiments. The constructed “water depth–gradation–energy” risk assessment matrix supports refined prediction and mitigation of landslide dam-break cascading hazards. Full article

Lithology prediction is a fundamental component of geological exploration and hydrocarbon reservoir characterization, playing a critical role in improving subsurface structural interpretation and enhancing resource prediction accuracy. However, well log data are typically characterized by high dimensionality, strong nonlinearity, severe class imbalance, and asymmetric geological feature distributions, which significantly restrict the predictive accuracy and generalization capability of conventional machine learning methods. To address these challenges, this study proposes a symmetry-aware lithology classification framework based on a Hybrid Grey Wolf Optimizer–Genetic Algorithm optimized Extreme Gradient Boosting (GGA-XGB) model. The proposed framework establishes a symmetric collaborative optimization mechanism by integrating the global exploration capability of the Grey Wolf Optimizer (GWO) with the local exploitation ability of the Genetic Algorithm (GA), thereby achieving a balanced optimization strategy between exploration and exploitation. Specifically, GWO first performs coarse-grained global hyperparameter optimization of XGBoost to improve search efficiency and optimization stability, while GA subsequently refines the parameter space to further enhance local optimization accuracy. Experimental results on a multi-class well logging dataset demonstrate that the proposed method achieves outstanding classification performance, with precision, recall, and F1-score all reaching 0.9862. Compared with several conventional machine learning methods, the proposed GGA-XGB framework exhibits superior predictive accuracy. The symmetry-aware optimization strategy provides an effective solution for intelligent lithology prediction under complex geological conditions and offers both theoretical insights into symmetry-aware optimization mechanisms and practical value for intelligent geoscience and hydrocarbon exploration. Full article

Multimodal emotion recognition in conversation (MERC) aims to identify the emotions in each utterance by modeling textual, acoustic, and visual evidence. Compared with unimodal emotion recognition in conversation (ERC), MERC can leverage complementary textual, acoustic, and visual information to support more accurate and consistent emotion inference. However, coordinating intramodal contextual dependencies, cross-modal alignment, and temporal affective dynamics within a unified framework in MERC is challenging. Existing solutions have advanced MERC through contextual modeling, multimodal fusion, and graph-based reasoning, but they still often rely on static relational assumptions or stage-wise coordination of modalities. This limits their ability to jointly model fine-grained relations, selective cross-modal interactions, and dynamic changes in emotion. To address these issues, we propose DGL-PCA (dynamic graph learning with pairwise cross-modal alignment), a dynamic graph-based framework for MERC. The model combines time-aware relation construction, dynamic time-aware heterogeneous graph modeling, and pairwise cross-modal alignment to improve prediction accuracy. This coordinates temporal affective dynamics, structured dialogue context, and multimodal interaction more explicitly than conventional coarse fusion or static graph formulations. Extensive experiments on IEMOCAP and CMU-MOSEI show that DGL-PCA consistently improves weighted F1 by 1.08–19.93% across all reproduced baselines. It achieves 70.02% and 83.91% weighted F1 on the IEMOCAP 6-way and 4-way settings, respectively, and 44.93% and 84.01% weighted F1 on the CMU-MOSEI 7-way and 2-way settings, respectively. Utterance-masking results demonstrated the robustness of the proposed method under dynamic emotional changes. In a 79-utterance IEMOCAP dialogue with 39 adjacent emotion transitions and up to nine transitions within a 15-utterance window, average weighted F1 slightly decreases by 1.87% in the case of any missing utterance, indicating long-context prediction stability under frequent emotion shifts. The proposed model paves the way for developing human-level emotion understanding capability of robots. Full article

Timely assessment of crop growth is essential for greenhouse management because irrigation, nutrient supply, pruning, and harvest scheduling all depend on reliable information on plant development. In practice, greenhouse imagery is affected by illumination variation, occlusion, and cluttered backgrounds, whereas deployment-oriented vision models must remain compact enough for resource-constrained computing environments. To address this challenge, FSA-GhostNet was developed as a lightweight visual backbone for greenhouse crop growth monitoring, and a Cucumber Growth Dataset (CGD) was constructed for dense temporal observation of cucumber development. The model was evaluated on ImageNet-100 and CGD under a unified protocol, and a joint learning setting was further used for simultaneous growth-stage classification and continuous Days After Planting (DAP) regression. FSA-GhostNet achieved accuracies of 78.98% on ImageNet-100 and 98.22% on CGD with 2.07 million trainable parameters. Under a unified RTX 3060 runtime setting, the model required 8.08 MB of storage, 0.868 G FLOPs, 13.96 ms/image latency, and 161.58 MB of peak GPU memory. In the joint prediction setting, it achieved 96.03% classification accuracy, a mean absolute error of 0.93 days, a root mean square error of 1.36 days, and an $R^2$ score of 0.9962. These results show that FSA-GhostNet maintains a strong balance between predictive performance and compactness for greenhouse crop monitoring, while CGD provides a useful temporal resource for fine-grained agricultural growth analysis. More broadly, the findings suggest that lightweight agricultural vision models can extend beyond coarse stage recognition toward more continuous and agronomically meaningful monitoring of crop development. Full article

Accurate detection of transmission line components is crucial for the stability and security of power grid operations. However, accurate power line inspection is always affected by complex factors, such as multi-scale objects, complex background interference, and object occlusion, etc. To tackle these complexities, this paper leverages the long-range dependency modeling advantages of the Transformer architecture, and an improved real-time end-to-end Detection Transformer (RT-DETR) with multi-scale feature reconstruction, referred to as MFRRT-DETR, for Unmanned Aerial Vehicle (UAV) inspection systems is presented. Specifically, an enhanced attention-based backbone network integrated via an aggregated pixel-focus attention (APFA) module is built which uses a dual-path design with fine-grained and coarse-grained branches to combine pixel-level focus with global perception to enhance the interaction between local and global features, alleviating the limitations of the local receptive field in Convolutional Neural Networks (CNNs). To further overcome the issues of target overlap, occlusion, and foreground–background confusion, a context-guided spatial feature reconstruction feature pyramid network (CGR-FPN) module is proposed which strengthens foreground representation and effectively fuses multi-scale features, improving performance in crowded scenes. Additionally, a Focaler–Shape IoU loss function is introduced to mitigate class imbalance issues and localization errors by focusing on hard samples and optimizing bounding box regression, particularly for long and wide irregular rectangular targets. Experiments show that the proposed MFRRT-DETR significantly outperforms advanced detection models, which effectively validates the detection efficiency and accuracy of the proposed model in complex inspection scenarios, making it a promising solution for UAV-based power line inspection. Full article

Application-specific instruction set processors (ASIPs) often employ specialized hardware to improve performance, but this introduces complexity in resource management and programming. Existing compiler solutions, including LLVM’s default schedulers, lack fine-grained structural conflict analysis for complex arithmetic logic unit (ALU) instructions, leading to suboptimal performance or runtime errors. This limitation becomes critical when targeting very long instruction word (VLIW) architectures with instruction fusion units that exhibit pipeline-stage-level resource contention. In this paper, we propose a compile-time instruction scheduling method that models sub-cycle resource usage and analyzes both data and structural dependencies at fine granularity. Unlike coarse-grained resource tables used in existing compilers, our approach tracks functional unit occupancy at the pipeline stage level, enabling precise detection of structural hazards in complex execution units. We implement this scheduler as a backend pass in the LLVM compiler framework and validate it on the Sayram VLIW processor for wireless communication. Experimental results show that our approach achieves 100% scheduling correctness while improving execution efficiency by 23% on average compared with in-order scheduling, with benefits up to 38% for highly parallel kernels such as PRACH, and reducing average running time by 66% compared with atomic execution. Full article

The design of IT services—including the architectural design—is considered a core activity to obtain a cost-effective IT service. Consequently, the main extant IT service frameworks and standards—rigorous and lightweight-agile types—aim to provide guidance for this purpose. However, some of them are reported at a coarse-grain conceptual level, and others are practically null. Consequently, their utilization—in the best case for large businesses—demands additional ad hoc organizational efforts. In the worst case, small and medium businesses (SMBs) are practically blocked from performing these relevant IT service activities by their limited resources. In this research, we are interested in supporting IT service design activities for SMBs, and thus introduce the AHP-AgileITS-ArchDesign model. This AHP-AgileITS-ArchDesign model was elaborated with a Design Science Research Methodology (DSRM), and aims to provide theoretically valid, systematic, agile and usable fine-grain guidance to support the design and evaluation of IT service architectural agile designs. The AHP-AgileITS-ArchDesign model uses a three-Layer AHP model derived from the main lightweight-agile IT service design literature, and its utilization is illustrated with an Academic Data Science Analytics Platform IT Service case. Then, its conceptual validity and its usability are evaluated, respectively, by a Panel of 16 Experts and an Exploratory Pilot Sample of 62 ITSM academics and practitioners. The evaluation results indicate that the AHP-AgileITS-ArchDesign model has a valid theoretical conceptualization and satisfactory usability, and thus can be used by practitioners and academics in SMBs. Full article

Social Q&A platforms produce short, noisy, and highly diverse user questions, making coarse topic labels insufficient for accurate information organization. This study proposes a dynamic seed topic construction and multi-demand topic identification framework for health science popularization questions on the Zhihu platform. Using 3529 cleaned questions derived from 2011 to 2024, the framework combines BERTopic clustering, LLM-based topic naming, and a TOP-K cross-filtering update strategy that integrates semantic similarity and frequency. The LLM then performs topic identification, optimization, and assignment, and a co-occurrence network analyzes topic associations. The method generates 26 initial seed topics, which expand to 51 topic instances and are subsequently optimized to 36 topics. Compared with LDA, BERTopic, and LLM-based baselines, our framework achieved the best topic coherence and the lowest topic similarity while maintaining good topic diversity. Human evaluation by 17 volunteers on 30 questions showed that the model performed at least as well as user-assigned tags across relevance, comprehensiveness, clarity, accuracy, and readability, with clearer advantages in relevance and accuracy. These results suggest that the proposed framework can identify fine-grained, interpretable, and strongly associated information-demand topics for social Q&A platforms. Full article

Clayey sand is a typical transitional coastal alluvial soil controlled by both coarse-grain friction and fine-grain cementation. Current studies focus mostly on remolded samples, lacking systematic understanding of water-induced structural degradation and quantitative physico-mechanical correlations for natural undisturbed clayey sand. In this work, 74 intact undisturbed specimens (0.5–23.0 m depth) were tested via basic physical tests, one-dimensional consolidation and consolidated-undrained triaxial shear tests. Pearson correlation analysis was performed to establish prediction relationships between routine physical indices and mechanical parameters. Results show the soil is classified as SC clayey sand with 39.70% fines and an average natural water content of 23.17%. Natural water content dominates soil engineering performance, presenting strong linear correlations with dry density and void ratio (|r| = 0.90). Higher water content and void ratio increase compressibility and reduce shear strength. The compression coefficient and compression modulus exhibited a consistent nonlinear relationship, reflecting the inherent linkage between these two compression parameters. Burial depth has little influence on soil properties, and plasticity index only serves for soil classification. Mechanistically, increasing moisture may thicken adsorbed water films, weaken interparticle contact and matric suction, and the fine particle-filled skeleton may further enhance the water sensitivity of the soil. The established prediction models support fast evaluation of soil mechanical behaviors, offering theoretical and practical support for geotechnical design of similar coastal clayey sand strata. Full article

Monocular depth estimation plays a crucial role in applications such as autonomous driving and mobile 3D reconstruction. However, existing lightweight methods are often constrained by limited computational resources and rely on shallow feature representations for direct depth regression. As a result, cross-scale residual information is insufficiently modeled, which limits their ability to preserve structural consistency and recover fine-grained details in complex scenes. To address these challenges, we propose LapCR-Net, a lightweight monocular depth estimation network based on Laplacian residual reconstruction. Specifically, we formulate a progressive Laplacian residual framework that decomposes depth prediction into a coarse-to-fine multi-scale refinement process. To enhance feature representation in the decoder, we introduce a Structure-aware Feature Recalibration (SFR) module and a Depth-guided Convolution Module (DCM), which strengthen spatial semantic correlations and improve residual prediction across scales. Furthermore, we design an uncertainty-driven collaborative refinement strategy to adaptively adjust residual correction strength. By estimating prediction uncertainty, the proposed strategy sharpens object boundaries while suppressing texture artifacts. Extensive experiments on the NYU-Depth V2 and KITTI benchmarks demonstrate that LapCR-Net achieves competitive performance with only 5.4 M parameters. In particular, it shows clear advantages in structural preservation and detail reconstruction, achieving a favorable trade-off between accuracy and computational efficiency. Full article

Zentropy theory has emerged as a multiscale thermodynamic framework that bridges quantum mechanics, statistical mechanics, and macroscopic materials behavior by embedding internal degrees of freedom within configurational ensembles. This review summarizes its theoretical foundations, representative applications, current limitations, and future directions. By incorporating intrinsic configurational entropy and free-energy-based statistical weighting, zentropy theory enables improved descriptions of phase stability, thermal expansion, and phase transitions in materials such as ferroelectrics, magnetic systems, high-entropy materials, and superconductors. Recent extensions also connect zentropy with artificial intelligence through data-driven thermodynamic modeling. Despite these advances, several challenges remain, including the ambiguity of configurational coarse-graining, strong cross-degree-of-freedom coupling, propagation of density functional theory errors, and limited applicability to delocalized or non-crystalline states. Future progress will require theoretical advances, including non-ergodic extensions, rigorous mathematical treatment of recursive multiscale entropy, and improved descriptions of low-temperature quantum effects. These efforts should be complemented by standardized software workflows, machine learning integration, and robust uncertainty quantification. Addressing these bottlenecks will help to further develop zentropy theory as a critically assessed framework for multiscale thermodynamic modeling and materials design. Full article