Search Results (582)

The increasing deployment of Internet of Things (IoT) devices introduces significant security challenges, while privacy concerns limit centralized data aggregation for intrusion detection. Federated learning (FL) offers a decentralized alternative, yet the interaction between feature representation, model architecture, and data heterogeneity remains insufficiently understood in IoT malware detection. This study provides a controlled comparative analysis of centralized and federated learning, optionally using amino acid encoding, under IID and Non-IID conditions using a 10,000-sample subset of the CTU–IoT–Malware–Capture dataset. First, we evaluate raw tabular features versus amino acid-based feature encoding, followed by a lightweight multi-layer perceptron (2882 parameters) versus a deeper residual network (70,532 parameters), across binary and multi-class classification tasks. In the binary setting, centralized training achieved up to 98.6% accuracy, while federated IID training reached 98.6%, with differences within statistical variance. Under Non-IID conditions, performance decreased modestly (0.1–0.5 percentage points), and accuracy was consistently lower when using encoded features compared with raw features. The degradation is smaller in deeper architectures and may offer improved stability under highly skewed federated conditions. In the four-class setting, the complex network achieved up to 97.8% accuracy with raw features, while amino acid encoding achieves up to 93.3%. The results show that federated learning can achieve performance comparable to centralized training under moderate heterogeneity, that lightweight architectures are sufficient for low-dimensional IoT traffic features, and that feature compression via amino acid encoding does not inherently mitigate Non-IID effects. These findings clarify the relative impact of representation, heterogeneity, and architectural capacity in practical FL-based IoT intrusion detection systems. Full article

This exploratory study examines the perception and production of the aspiration contrast in Mandarin voiceless retroflex affricates zh [tʂ] and ch [tʂ^h] by ten adult Spanish speakers (three Peruvian, seven Chilean) at Nanjing University. Participants completed a perception identification task and a production reading task using the same set of 128 syllables. Voice Onset Time (VOT) measurements from the production task were converted to binary classifications for cross-modality comparison. Perception accuracy was moderately high (zh [tʂ]: 84.43%; ch [tʂ^h]: 82.39%), whilst production accuracy was substantially lower (zh [tʂ]: 32.61%; ch [tʂ^h]: 19.15% within native VOT ranges). Participants maintained the aspiration contrast (zh [tʂ] = 58 ms, ch [tʂ^h] = 125 ms) but consistently underproduced VOT compared to native speakers (zh [tʂ] = 67 ms, ch [tʂ^h] = 164 ms). Perception patterns align with Category Goodness (CG) assimilation within PAM-L2: both Mandarin sounds map to Spanish [tʃ] but with different goodness-of-fit, enabling moderate discrimination. Production follows SLM-r predictions, with learners developing a Composite L1–L2 Category that maintains the aspiration contrast but fails to establish new phonetic categories. The small sample size (n = 10) precluded robust statistical testing of individual differences. The perception–production asymmetry supports independent modality development in L2 phonetic acquisition. Full article

Cybercriminals use advanced techniques to launch an attack against organizations, which causes disruption of normal business activities. The traditional signature-based malware detection methods are not effective in the detection of ransomware. Therefore, the use of machine learning and deep learning for malware detection is becoming a major area of research. There are two types of malware detection strategies, namely, static and dynamic. This work investigates the task-dependent effectiveness of static PE header-based detection by systematically evaluating three binary classification problems of increasing difficulty: ransomware vs. benign, malware vs. benign, and ransomware vs. other malware families. An end-to-end machine learning pipeline is implemented, including dataset-specific preprocessing, class imbalance handling, model training, and evaluation using imbalance-aware metrics. Random Forest, Support Vector Machine, and XGBoost models are assessed across all tasks, with SHAP used to analyze feature contribution and explain performance degradation. The experimental results demonstrate that tree-based ensemble models, particularly XGBoost, achieve strong detection performance when class boundaries are structurally distinct, but they struggle when ransomware must be distinguished from structurally similar malware. The results indicate that static analysis based on PE header features can be a viable approach for pre-execution triage, but they exhibit clear limitations for fine-grained ransomware discrimination. Full article

Filtering constitutes a critical step in the post-processing of airborne Light Detection And Ranging (LiDAR) data. Over the past decade, machine learning has emerged as a prominent methodological paradigm across numerous disciplines, attracting significant research interest in its application to LiDAR filtering. From a machine learning perspective, filtering is essentially a binary classification task that aims to discriminate between ground and non-ground points. However, the limited information inherent in point clouds often leads to the generation of highly correlated features, particularly those derived from height data, which can compromise filtering accuracy. To address this issue, feature selection becomes imperative. In this study, we employed height-based mutual information as a criterion to identify and eliminate less discriminative features for filtering. The AdaBoost (Adaptive Boosting) algorithm was adopted as the classifier for point cloud filtering. For each point, nineteen features were derived from the raw LiDAR point cloud based on height and other geometric attributes within a defined neighborhood. The performance of the proposed feature selection approach was evaluated using benchmark datasets provided by the International Society for Photogrammetry and Remote Sensing (ISPRS). Experimental results demonstrate that the method is effective and reliable. After removing three selected features, the average kappa coefficient improved, along with a reduction in three categories of error, although a slight increase in Type II error (0.15%) was observed. Full article

Fire detection is a critical task for early warning systems, particularly in environments where visual sensing is unreliable. While most existing approaches rely on image-based or smoke-based detection, acoustic signals provide complementary information capable of capturing early combustion-related events. This study investigates deep learning models for sound-based fire detection, focusing on convolutional and Transformer-based architectures. VGG16 and VGG19 convolutional neural networks are adapted to process time-frequency audio representations for binary classification into Fire and No-Fire classes. An Audio Spectrogram Transformer (AST) is further employed to model long-range temporal dependencies in acoustic data. Finally, a hybrid VGG19-AST architecture is proposed, in which convolutional layers extract local spectral–temporal features, and Transformer-based self-attention performs global sequence modeling. The models are evaluated on a curated dataset containing fire sounds and diverse environmental background noises under multiple noise conditions. Experimental results demonstrate competitive performance across convolutional and Transformer-based models, while the proposed hybrid VGG19-AST architecture achieves the most consistent overall results. The findings suggest that integrating convolutional feature extraction with self-attention-based global modeling enhances robustness under complex acoustic variability. The proposed hybrid framework provides a scalable and cost-effective solution for sound-based fire detection, particularly in scenarios where visual monitoring may be obstructed or ineffective. Full article

Fish identification and classification tasks allow scientists greater data on the sustainability and diversity of fisheries. Advances in computer vision, particularly convolutional neural networks (CNNs), have enabled automated fish detection at scale; however, increased model depth may not justify the additional time and computational cost. This study examines the ResNet family of models for binary fish identification to assess whether deeper networks provide meaningful performance advantages over simpler configurations. We compare ResNet-18, ResNet-50, ResNet-101, and ResNet-152 using both non-pretrained and pretrained initializations. We introduce an Energy-Weighted Score (EWS) to enable comparison of computational resource usage using cost-based weighting. A reliable fish versus no-fish classification can be achieved with as few as 18 layers, yielding 99.975% accuracy which improved to 100% accuracy following threshold optimization. For binary fish identification, increasing ResNet depth provides increased EWS scores with little impact on accuracy returns over shallower models. Overall, models with fewer layers outperformed deeper models with more parameters, and additional depth and tuning techniques were unable to outperform simpler configurations while delivering higher EWS scores. Full article

Background: Advanced liver fibrosis (LF) is a major determinant of prognosis across chronic liver diseases. Current biomarkers are often etiology-specific and lack cross-cohort robustness. Shared molecular drivers across etiologies remain incompletely defined, and effective anti-fibrotic therapies are limited. Methods: We developed a multi-algorithm consensus machine-learning framework to derive a robust LF progression signature. In the training non-alcoholic fatty liver disease (NAFLD) cohort GSE213621 (n = 368), samples were formulated as a binary classification task (mild fibrosis, F0–F2; advanced fibrosis, F3–F4). Candidate genes were screened in parallel using Boruta, Least Absolute Shrinkage and Selection Operator (LASSO), random forest, and eXtreme Gradient Boosting (XGBoost). Genes selected by at least two algorithms were defined as a high-consensus pool, and genes consistently selected by all four algorithms were prioritized to construct a core signature. Model performance was evaluated by stratified cross-validation in the training cohort and externally validated in four independent cohorts of different etiologies (GSE49541, GSE84044, GSE130970, and GSE276114). Cellular sources of signature genes were characterized using single-cell RNA sequencing (scRNA-seq) datasets GSE136103 (human) and GSE172492 (mouse). For therapeutic discovery, the high-consensus expression profile was queried against the Connectivity Map (CMap) to prioritize compounds predicted to reverse the fibrotic transcriptional program. Withaferin A (WFA) was selected for experimental validation in a carbon tetrachloride (CCl₄)-induced mouse LF model and in the transforming growth factor-β1 (TGF-β1)-stimulated human hepatic stellate cell line LX-2. Bulk liver RNA-seq profiling was performed to interrogate WFA-associated molecular changes in vivo. Results: We identified a six-gene signature (CLEC4M, COL25A1, ITGBL1, NALCN, PAPPA, and PEG3) that discriminated advanced from mild fibrosis, achieving a mean AUC of 0.890 in internal cross-validation and an average AUC of 0.864 across external validation cohorts. scRNA-seq analysis revealed cell-type-specific expression with prominent enrichment in fibroblast populations. In vivo, WFA markedly attenuated CCl₄-induced fibrosis (p < 0.05) and reversed 1314 fibrosis-associated differentially expressed genes (adjusted p < 0.05), which were enriched in fatty acid metabolism and PPAR signaling, as well as extracellular matrix (ECM)–receptor interaction and focal adhesion (adjusted p < 0.05). In vitro, WFA suppressed TGF-β1-induced LX-2 activation, reducing α-SMA and Fibronectin expression (p < 0.05). Conclusions: We report a six-gene signature that robustly predicts advanced LF across etiologies, define its cellular context using single-cell atlases, and validate the anti-fibrotic activity of WFA in both in vivo and in vitro models. Bulk liver RNA-seq and cellular evidence further suggest that WFA-associated effects are linked to lipid metabolic programs, ECM remodeling, and attenuation of hepatic stellate cell activation. Full article

Early-stage fungal contamination in maize kernels is difficult to identify visually and it can cause severe quality and safety risks during storage and transportation. Short-wave infrared (SWIR) hyperspectral imaging offers a rapid, non-destructive approach by capturing chemical information related to water, proteins, and lipids. This study investigates the early detection and classification of Gibberella zeae contamination in maize kernels using SWIR hyperspectral imaging combined with machine learning. Two maize varieties were artificially inoculated and cultured under controlled conditions, followed by hyperspectral data collection over six contamination stages. Various preprocessing techniques including standard normal variate (SNV), second derivative (SD), multiplicative scatter correction (MSC), and derivatives were evaluated to enhance data quality. Feature wavelength selection was performed using successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), and uninformative variable elimination (UVE), significantly reducing redundancy and improving classification performance. Multiple models, including linear discriminant analysis (LDA), multilayer perceptron (MLP), support vector machine (SVM), a convolutional neural network (CNN), long short-term memory (LSTM) network, and a hybrid architecture Transformer that integrated a CNN, a LSTM network, and a Transformer (abbreviated as CLT), were constructed for both binary (healthy vs. contaminated) and multiclass classification tasks. Specifically, the multiclass task consisted of six contamination stages corresponding to contamination time from Day 0 to Day 5. The best binary classification task accuracy of 100% was achieved using SNV-preprocessed data with the MLP model. For multiclass classification task, the SD-preprocessed LDA model reached a test accuracy of 92.56%. Combined with appropriate preprocessing, feature selection and modeling, these results demonstrate that hyperspectral imaging is a powerful tool for the non-destructive, early-stage identification of fungal contamination in maize kernels, offering strong support for food safety and quality monitoring. Full article

Short-term forecasting of solar and wind power generation is critical for smart grid management but challenging due to non-stationarity and extreme generation events. This study addresses a multi-task learning problem: regression-based forecasting of power output and binary detection of extreme events defined by a quantile-based threshold (q = 0.90). A hybrid spatio-temporal model, DP-STH++, is proposed, implementing parallel causal fusion of LSTM, GRU, a causal Conv1D stack, and a lightweight causal transformer. The architecture employs regression and classification heads, while an uncertainty-weighted mechanism stabilizes multitask optimization in the regression tasks; extreme event detection performance is evaluated using AUC. Training and evaluation follow a leakage-safe protocol with chronological data processing, calendar feature integration, time-aware splitting, and training-only estimation of scaling parameters and extreme thresholds. Experimental results obtained with a one-hour forecasting horizon and a 24 h context window demonstrate that DP-STH++ achieves the best regression performance on the hold-out set (RMSE = 257.18, MAE = 174.86–287.90, MASE = 0.2438, R² = 0.9440) and the highest extreme event detection accuracy (AUC = 0.9896), ranking 1st among all compared architectures. In time-series cross-validation, the model retains the leading position with a mean MASE = 0.3883 and AUC = 0.9709. The advantages are particularly pronounced for wind power forecasting, where DP-STH++ simultaneously minimizes regression errors and maximizes AUC = 0.9880–0.9908. Full article

The effective management of municipal road infrastructure requires up-to-date, standardized and reliable condition information to support sustainable maintenance. While visual road-condition assessment methods based on established standards are widely applied to municipal roads, they remain largely manual, time-consuming, costly and subjective. This study presents an end-to-end workflow for the automated visual inspection and condition assessment of municipal road infrastructure using high-resolution, 3D street-level imagery acquired by professional mobile mapping systems. The proposed approach integrates an efficient preprocessing pipeline for precise road-surface extraction with deep learning models trained for the specific task and an advanced postprocessing method for robust results aggregation. For this purpose, a large dataset covering approximately 352 km of municipal roads across eight municipalities was created by combining street-level imagery with expert-annotated road-condition index (RCI) values. Two neural network variants were implemented: a regression model predicting standardized RCI values and a binary classifier distinguishing between roads requiring maintenance and those in good condition. To ensure decision-oriented outputs at the infrastructure-asset level, frame-based predictions are aggregated into homogeneous road segments using outlier detection and change-point analysis along the road axis. The regression model achieved a mean absolute error of 0.48 RCI values at frame level and 0.40 RCI values at road-segment level, outperforming conventional inter-expert variability, while the binary classification model reached an F1-score of 0.85. These findings demonstrate that AI-based visual road-condition assessment using professional mobile mapping data can provide accurate, standardized and scalable condition information for municipal road infrastructure. The proposed workflow supports maintenance prioritization and infrastructure management decisions without requiring explicit detection of individual pavement defects, offering a practical pathway toward automated, cost-effective road-condition monitoring. Full article

Fine-tuning a BERT-Base model for specific tasks, such as sentiment analysis, has become resource-intensive and often requires high computational power and memory. This paper introduces SCALE, a novel resource-efficient fine-tuning method that targets the most critical transformer layers, which reduces computational costs without sacrificing performance. By dynamically profiling transformer layers via activation magnitudes and attention entropy, SCALE selects and adapts only the most influential layers with lightweight adapter modules. The proposed method outperforms traditional fine-tuning techniques, achieving a 2.3% improvement in accuracy on the IMDB dataset and reducing training time by 56.3% compared to full-model fine-tuning. Experiments across various sentiment analysis benchmarks demonstrate SCALE’s effectiveness in optimizing fine-tuning for the BERT-base model in resource-constrained environments, achieving up to 99% of the performance of full-model fine-tuning while using only 40% of the parameters. The empirical validation in this study is restricted to binary and multi-class sentiment classification. The evaluation specifically reflects effectiveness in sentiment analysis text classification tasks. Full article

Arrhythmia is a life-threatening cardiovascular disease if not detected early. While deep learning models have demonstrated strong performance in ECG-based arrhythmia classification, deploying these models on resource-constrained wearable devices remains challenging. In this paper, we present a quantization-compatible neural architecture search (NAS) framework that discovers ultra-compact minority-aware convolutional neural networks (CNN). We formulate NAS as a multi-objective optimization problem, jointly maximizing balanced accuracy and minority-classes recall while minimizing model size and computational complexity. Furthermore, we constrain our search space to INT8-compatible operations. We evaluated our framework on the MIT-BIH Arrhythmia Database. We discovered NanoCNN models for binary and multi-class classification tasks. The models trained without augmentation achieved 98.7% and 98.21% overall accuracies outperforming the state-of-the-art. The discovered models required 38.3 K and 51.5 K multiply-accumulate operations (MAC) per inference, enabling their deployment on ARM Cortex-M4 microcontrollers. With augmentation and other minority-aware interventions, our model attained 91.6% balanced accuracy. Our results validated the effectiveness of the adopted search and training techniques for arrhythmia screening and diagnosis. Full article

Facial expression recognition (FER) in naturalistic settings is constrained by label ambiguity and variability in stimulus–response alignment. Adopting a data-centric perspective, this study examined whether emotional intelligence (EI)-stratified training data influence FER performance by treating EI as a qualitative factor associated with affective data consistency. Naturally elicited facial expressions were collected in a controlled emotion induction experiment with subjective arousal and valence ratings. Using response-driven labeling, neutral ratings were retained as indicators of ambiguity. Participants were grouped into High and Low EI based on the alignment between subjective evaluations and outputs from a pretrained affect estimator. Identical binary classifiers for arousal and valence recognition were trained while varying only the training data composition and evaluated across baseline, unambiguous, and ambiguous test sets using independent training repetitions with repetition-level statistical aggregation. EI-stratified training was associated with statistically detectable, context-dependent performance differences: group effects were observed primarily under baseline conditions and, to a lesser extent, under ambiguous conditions, whereas no reliable differences emerged under unambiguous conditions. Pooled discrimination differences were modest, but item-level analyses identified significant differences in classification correctness in specific task–condition combinations. Comparable patterns were observed across alternative backbone architectures. These findings indicate that FER performance in naturalistic contexts is influenced not only by model architecture but also by the statistical structure and internal coherence of the training data, supporting EI-informed data selection in ambiguity-prone scenarios. Full article

Lung cancer continues to be one of the leading causes of cancer-related deaths worldwide, as pulmonary nodules are often diagnosed at later stages. Therefore, accurate nodule classification is crucial for enabling early detection and supporting timely clinical decision-making. This study proposes a computer-aided diagnosis (CAD) system for lung nodule classification using computed tomography (CT) images, specifically focused on malignancy prediction and structural morphology analysis. The proposed framework is based on a novel attention-based Convolutional Neural Network (CNN) that incorporates both channel-wise and spatial attention mechanisms. This dual-attention structure enables the model to emphasize diagnostically relevant features while suppressing irrelevant background information, thereby improving interpretability and classification accuracy. For benchmarking purposes, CNN, CNN-SVM, and ResNet101 architectures were implemented for comparison. Experimental results on the LIDC-IDRI dataset for binary classification (benign vs. malignant) and on the IQ-OTH/NCCD dataset for both binary and three-class (normal, benign, malignant) classification tasks demonstrate that the proposed Attention-Based CNN outperforms all baseline models, achieving a maximum classification accuracy of 98% in the binary setting. In addition to accuracy, the proposed model achieves strong performance across multiple evaluation metrics, including precision, recall, F1-score, AUC, and separately reported confusion matrices for both binary and multiclass evaluations, indicating the robustness of the approach. The dual-attention mechanism enhances salient feature localization and discriminative representation learning, thereby contributing to improved performance in both binary and multiclass classification tasks Full article

Haze, as a critical factor affecting regional air quality and human health, necessitates accurate remote sensing identification for pollution monitoring and climate research. This study proposes a two-stage haze mapping algorithm (THMA), based on a backpropagation neural network and a random forest model, which achieves high-precision identification of haze, clouds, and clear air using FY-4A AGRI geostationary satellite data, with small misclassification rates and high F1 scores. Through detailed comparison with CALIOP observations, THMA performs well over most regions over Asia, successfully extending the traditional binary classification task of distinguishing only clouds and clear air. Notably, the model provides good classification capability in vertically overlapping areas of broken clouds and haze, with minimal misclassification even over bright surfaces such as deserts and ice/snow. Statistical analysis for the year 2022 shows that the annual average number of haze days is 51.3 in China. This study confirms the significant complementary value of satellite remote sensing and ground-based observations for haze monitoring. Full article