Search Results (96)

Symmetry and asymmetry are foundational design principles in artificial intelligence, defining the balance between invariance and adaptability in multimodal learning systems. In audio-visual speaker verification, where speech and lip-motion features are jointly modeled to determine whether two utterances belong to the same individual, these principles govern both fairness and discriminative power. In this work, we analyze how symmetry and asymmetry emerge within a gated-fusion architecture that integrates Time-Delay Neural Networks and Bidirectional Long Short-Term Memory encoders for speech, ResNet-based visual lip encoders, and a shared Conformer-based temporal backbone. Structural symmetry is preserved through weight-sharing across paired utterances and symmetric cosine-based scoring, ensuring verification consistency regardless of input order. In contrast, asymmetry is intentionally introduced through modality-dependent temporal encoding, multi-head attention pooling, and a learnable gating mechanism that dynamically re-weights the contribution of audio and visual streams at each timestep. This controlled asymmetry allows the model to rely on visual cues when speech is noisy, and conversely on speech when lip visibility is degraded, yielding adaptive robustness under cross-modal degradation. Experimental results demonstrate that combining symmetric embedding space design with adaptive asymmetric fusion significantly improves generalization, reducing Equal Error Rate (EER) to 3.419% on VoxCeleb-2 test dataset without sacrificing interpretability. The findings show that symmetry ensures stable and fair decision-making, while learnable asymmetry enables modality awareness together forming a principled foundation for next-generation audio-visual speaker verification systems. Full article

The rapid evolution of modern vehicles into intelligent and interconnected systems presents new complexities in both functional safety and cybersecurity. In this context, ensuring the reliability and integrity of critical sensor data, such as wheel speed inputs for anti-lock brake systems (ABS), is essential. Effective detection of wheel speed sensor faults not only improves functional safety, but also plays a vital role in keeping system resilience against potential cyber–physical threats. Although data-driven approaches have gained popularity for system development due to their ability to extract meaningful patterns from historical data, a major limitation is the lack of diverse and representative faulty datasets. This study proposes a novel dual learning model, based on Temporal Convolutional Networks (TCN), designed to accurately distinguish between normal and faulty wheel speed sensor behavior within a hardware-in-the-loop (HIL) simulation platform implemented on an FPGA. To address dataset limitations, a TruckSim–MATLAB/Simulink co-simulation environment is used to generate realistic datasets under normal operation and eight representative fault scenarios, yielding up to 5000 labeled sequences (balanced between normal and faulty behaviors) at a sampling rate of 60 Hz. Two TCN models are trained independently to learn normal and faulty dynamics, and fault decisions are made by comparing the reconstruction errors (MSE and MAE) of both models, thus avoiding manually tuned thresholds. On a test set of 1000 sequences (500 normal and 500 faulty) from the 5000 sample configuration, the proposed dual TCN framework achieves a detection accuracy of 97.8%, a precision of 96.5%, a recall of 98.2%, and an F1-score of 97.3%, outperforming a single TCN baseline, which achieves 91.4% accuracy and an 88.9% F1-score. The complete dual TCN architecture is implemented on a Xilinx ZCU102 FPGA evaluation kit (AMD, Santa Clara, CA, USA), while supporting real-time inference in the HIL loop. These results demonstrate that the proposed approach provides accurate, low-latency fault detection suitable for safety-critical ABS applications and contributes to improving both functional safety and cyber-resilience of braking systems. Full article

To ensure reliable, efficient and sustainable operation of modern power networks, accurate load forecasting is an important task in system planning and control. It is also a crucial task for the efficient operation of smart grids to maintain a balance between load shifting, load management and power dispatch. In this regard, this research study aims to investigate the efficiency of various machine learning models for whole-house energy consumption prediction and appliance-level load disaggregation using Non-Intrusive Load Monitoring (NILM). The primary objective is to determine which model offers the most accurate forecasts for both individual appliance consumption patterns and the total amount of energy used by the household. The empirical study presents comparative performance analysis of machine learning models, i.e., Random Forest, Decision Tree, K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Gradient Boosting and Support Vector Regressor (SVR) for load forecasting and load disaggregation. This research is conducted on PRECON: Pakistan Residential Electricity Dataset consisting of 42 Pakistani households. The dataset was recorded originally as one minute per sample, but the proposed study aggregated it to hourly samples to evaluate models’ alignment with the typical sampling rate of smart meters in Pakistan. It enables the models to more accurately depict implementation scenarios in real-world settings. The statistical measures MAE, MSE, RMSE and R² have been employed for performance evaluation. The proposed Random Forest algorithm out-performs all other employed models, with the lowest error values (MAE: 0.1316, MSE: 0.0367, RMSE: 0.1916) and the highest R² score of 0.9865. Furthermore, for detecting appliance events from aggregate power data, ensemble models such as Random Forest performed better than other models for ON/OFF prediction. To evaluate the suitability of machine learning models for real-time, appliance-level energy forecasting using Non-Intrusive Load Monitoring (NILM), this study presents a novel evaluation framework that combines learning speed and edge adaptability with conventional performance metrics (e.g., R², MAE). This paper introduces a NILM-based approach for load forecasting and appliance-level ON/OFF prediction, representing its capacity to improve residential energy efficiency and encourage sustainable energy consumption, while emphasizing operational metrics for implementation in embedded smart grid systems—an area mainly neglected in prior NILM-based research articles. The results provide useful information for improving demand-side energy management, facilitating more effective load disaggregation, and maximizing the energy efficiency and responsiveness of smart grids. Full article

Pooling strategies are fundamental to convolutional neural networks, shaping the trade-off between accuracy, robustness to spatial variations, and computational efficiency in modern visual recognition systems. In this paper, we present and validate ECA110-Pooling, a novel rule-based pooling operator inspired by elementary cellular automata. We conduct a systematic comparative study, benchmarking ECA110-Pooling against conventional pooling methods (MaxPooling, AveragePooling, MedianPooling, MinPooling, KernelPooling) as well as state-of-the-art (SOTA) architectures. Experiments on three benchmark datasets—ImageNet (subset), CIFAR-10, and Fashion-MNIST—across training horizons ranging from 20 to 50,000 epochs show that ECA110-Pooling consistently achieves higher Top-1 accuracy, lower error rates, and stronger F1-scores than traditional pooling operators, while maintaining computational efficiency comparable to MaxPooling. Moreover, when compared with SOTA models, ECA110-Pooling delivers competitive accuracy with substantially fewer parameters and reduced training time. These results establish ECA110-Pooling as a principled and validated approach to image classification, bridging the gap between fixed pooling schemes and complex deep architectures. Its interpretable, rule-based design highlights both theoretical significance and practical applicability in contexts that demand a balance of accuracy, efficiency, and scalability. Full article

Based on the WRF-3DVar system, this study investigates the impacts of assimilating radar and automatic weather station (AWS) observations, both independently and jointly, for a squall line case that occurred over complex terrain in China on 30 May 2024. It is found that radar data assimilation with spatial truncation significantly enhances the representation of convective structures while reducing false echoes by about 40%. However, when the variance and correlation length scales are enlarged, reflectivity intensity is increased by 5–10 dBZ with false signals and positional errors also introduced, while a balanced scheme is observed to yield the highest skill scores. Assimilation of AWS alone provides limited improvements, whereas radar assimilation introduces localized structures that rapidly decay within 1–2 h due to the absence of boundary-layer constraints. The benefits of joint assimilation are clearly demonstrated in terms of spatial continuity and vertical consistency, with the assimilation order being identified as a decisive factor. When AWS is assimilated prior to radar, low-level thermodynamic and dynamic conditions are optimized, leading to strengthened cold pool structures by around 2 K, enhanced updrafts by over 20%, and improved wind distribution. The critical role of AWS-radar joint assimilation in capturing the dynamical characteristics of squall lines is thus highlighted. Detailed examination of the forecast and analysis indicates that assimilating AWS before radar not only optimizes boundary-layer conditions but also enhances the coupling between cold pools and updrafts, resulting in improved simulation accuracy in both horizontal and vertical structures. These findings provide valuable insights for advancing the prediction of severe convective systems. Full article

Background and Objectives: Stratifying thyroid nodules according to malignancy risk is a crucial step in early diagnosis and patient care. Recently, deep learning techniques have emerged as powerful tools for medical diagnostics, particularly with convolutional neural networks (CNNs) applied to medical image classification. This study aimed to develop a new hybrid CNN model for classifying thyroid nodules using the TN5000 ultrasound image dataset. Materials and Methods: The TN5000 dataset includes 5000 ultrasound images, with 3572 malignant and 1428 benign nodules. To address the issue of class imbalance, the researchers applied an R-based anomaly data augmentation method and a GAN-based technique (G-RAN) to generate synthetic benign images, resulting in a balanced dataset for training. The model architecture was built on a pre-trained EfficientNet-B3 backbone, further enhanced with squeeze-and-excitation (SE) blocks and residual refinement modules to improve feature extraction. The task was to classify malignant nodules (labeled 1) and benign nodules (labeled 0). Results: The proposed hybrid CNN achieved strong performance, with an accuracy of 89.73%, sensitivity of 90.01%, precision of 88.23%, and an F1-score of 88.85%. The total training time was 42 min. Conclusions: The findings demonstrate that the proposed hybrid CNN model is a promising tool for thyroid nodule classification on ultrasound images. Its high diagnostic accuracy suggests that it could serve as a reliable decision-support system for clinicians, improving consistency in diagnosis and reducing human error. Future work will focus on clinical validation, explainability of the model’s decision-making process, and strategies for integration into routine hospital workflows. Full article

Background and Objective: Accurate classification of standard fetal ultrasound planes is a critical step in prenatal diagnostics, enabling reliable biometric measurements and anomaly detection. Conventional deep learning approaches, particularly convolutional neural networks (CNNs) and transformers, often face challenges such as domain variability, noise artifacts, class imbalance, and poor calibration, which limit their clinical utility. This study proposes a hybrid state–space and vision transformer framework designed to address these limitations by integrating sequential dynamics and global contextual reasoning. Methods: The proposed framework comprises five stages: (i) preprocessing for ultrasound harmonization using intensity normalization, anisotropic diffusion filtering, and affine alignment; (ii) hybrid feature encoding with a state–space model (SSM) for sequential dependency modeling and a vision transformer (ViT) for global self-attention; (iii) multi-task learning (MTL) with anatomical regularization leveraging classification, segmentation, and biometric regression objectives; (iv) gated decision fusion for balancing local sequential and global contextual features; and (v) calibration strategies using temperature scaling and entropy regularization to ensure reliable confidence estimation. The framework was comprehensively evaluated on three publicly available datasets: FETAL_PLANES_DB, HC18, and a large-scale fetal head dataset. Results: The hybrid framework consistently outperformed baseline CNN, SSM-only, and ViT-only models across all tasks. On FETAL_PLANES_DB, it achieved an accuracy of 95.8%, a macro-F1 of 94.9%, and an ECE of 1.5%. On the Fetal Head dataset, the model achieved 94.1% accuracy and a macro-F1 score of 92.8%, along with superior calibration metrics. For HC18, it achieved a Dice score of 95.7%, an IoU of 91.7%, and a mean absolute error of 2.30 mm for head circumference estimation. Cross-dataset evaluations confirmed the model’s robustness and generalization capability. Ablation studies further demonstrated the critical role of SSM, ViT, fusion gating, and anatomical regularization in achieving optimal performance. Conclusions: By combining state–space dynamics and transformer-based global reasoning, the proposed framework delivers accurate, calibrated, and clinically meaningful predictions for fetal ultrasound plane classification and biometric estimation. The results highlight its potential for deployment in real-time prenatal screening and diagnostic systems. Full article

Due to the highly individualised presentation of sport-related concussion (SRC), multi-domain assessments examining cognitive, migraine, vestibular, ocular, mood, sleep, and neck-related function have been suggested to assist clinicians with diagnosis, management, and rehabilitation. Normative data on such assessments for female, community players from countries outside the U.S. are needed. This study aimed to (i) describe normative data from community-level Ladies Gaelic Football players using a multi-domain assessment, and (ii) compare findings between adolescent and adult players. A total of 138 LGF players without SRC (101 adults, 37 adolescents) completed a multi-domain SRC assessment including Sport Concussion Assessment Tool 5th Edition, Concussion Clinical Profiles Screening, Vestibular/Ocular Motor Screening (VOMS), Immediate Post-Concussion Assessment and Cognitive Testing (ImPACT^®), Patient Health Questionnaire-9, Generalised Anxiety Disorder-7, Pittsburgh Sleep Quality Index, Migraine Disability Assessment, and Neck Bournemouth Questionnaire, and neck dynamometry. Normative data were summarised using descriptive statistics, while differences in test scores between adolescents and adults were examined using parametric or non-parametric tests. While adolescents and adults scored similarly on most measures, adolescents scored worse on ImPACT^® visual–motor speed (d = 0.09) and reaction time (r = 0.52), SCAT5 concentration (V = 0.38), total modified Balance Error Scoring System (r = 0.42), and CP Screen vestibular profile (r = 0.38) (p < 0.05). This is the first study to describe and compare normative data for multidomain SRC assessments in adolescent and adult female, community athletes. Differences in some tests between adolescents and adults highlight the need for demographic-specific normative data when interpreting post-SRC assessment results. Full article

This study compared the effects of core (CTG) or mobility training (MTG) on basketball-specific skills in youth players, focusing on dynamic balance. Both training modalities have a recognized role in enhancing performance, but few studies have examined their impact on this population. Thirty-one young (age 14.71 ± 2.27 years) males were assigned to an 8-week CTG or MTG. Overhead Squat, Y-Balance Test, Agility T-Test, Sit-and-Reach, Functional Hop Tests, and the Balance Error Scoring System were assessed before (pre) and after (post) the intervention for both dominant (D) and non-dominant (ND) limbs. Both groups improved the postero-lateral direction of the Y-Balance Test for the D (CTG, MD [95% CIs] = −8.108 [−15.620, −0.595], p = 0.035; MTG, MD [95% CIs] = −15.234 [−23.512, −6.956], p = 0.024) and ND (CTG, MD [95% CIs] = −9.110 [−16.150, −2.070], p = 0.013; MTG MD [95% CIs] = −13.899 [−21.657, −6.141], p = 0.001) limb and the medial reach for D (CTG, MD [95% CIs] = −17.279 [−26.364, −8.194], p = 0.001; MTG, MD [95% CIs] = −22.050 [−32.061, −12.039], p = 0.03) and ND (CTG, MD [95% CIs] = −9.309 [−17.093, −1.526], p = 0.021; MTG, MD [95% CIs] = −13.614 [−22.190, −5.037], p = 0.003), the Overhead Squat Test (CTG, MD [95% CIs] = −3.059 [−3.797, −2.321], p = 0.001; MTG, MD [95% CIs] = −3.643 [−4.456, −2.830], p = 0.001), and Agility T-Test (CTG, MD [95% CIs] = 0.572 [0.072, 1.073], p = 0.026; MTG, MD [95% CIs] = 0.696 [0.145, 1.248], p = 0.024) skills. Only CTG showed a significant improvement (MD [95% CIs] = −8.294 [−16.162, −0.426], p = 0.04) in single-leg hop performance for the ND limb. No significant improvements were observed in static balance or flexibility. No time × group effect was found. Both interventions improved key basketball-specific motor abilities and could be added to the basketball training session without adverse effect. Full article

Road safety has emerged as a global issue, driven by the rapid rise in vehicle ownership and traffic congestion. Human error, like distraction, drowsiness, and panic, is the leading cause of road accidents. Conventional driver monitoring systems (DMSs) frequently fail to detect these emotional and cognitive states, limiting their potential to prevent accidents. To overcome these challenges, this work proposes a robust deep learning-based DMS framework capable of real-time detection and response to emotion-driven driver behaviors that pose safety risks. The proposed system employs convolutional neural networks (CNNs), specifically the Inception module and a Caffe-based ResNet-10 with a Single Shot Detector (SSD), to achieve efficient, accurate facial detection and classification. The DMS is trained on a comprehensive and diverse dataset from various public and private sources, ensuring robustness across a wide range of emotions and real-world driving scenarios. This approach enables the model to achieve an overall accuracy of 98.6%, an F1 score of 0.979, a precision of 0.980, and a recall of 0.979 across the four emotional states. Compared with existing techniques, the proposed model strikes an effective balance between computational efficiency and complexity, enabling the precise recognition of driving-relevant emotions, making it a practical and high-performing solution for real-world in-car driver monitoring systems. Full article

Concussions, classified as a type of mild traumatic brain injury (mTBI), are frequently underdiagnosed due to the subjective nature of symptoms and limitations in existing diagnostic methodologies. Current clinical evaluations, including tools such as the Sport Concussion Assessment Tool 5 (SCAT5), Balance Error Scoring System (BESS), and Vestibular Ocular Motor Screening (VOMS), demonstrate high sensitivity and specificity but often fail to capture the full complexity of concussive injuries. Emerging diagnostic approaches, such as blood biomarkers (for example, glial fibrillary acidic protein (GFAP), ubiquitin C-terminal hydrolase-L1 (UCH-L1), S100 calcium-binding protein B (S100B), and tau) and advanced neuroimaging techniques (for example, diffusion tensor imaging (DTI) and functional magnetic resonance imaging (fMRI)), show promise but remain impractical for routine clinical use due to accessibility and standardization challenges. This review examines objective markers, including neuroimaging, electrophysiological measures (for example, Electroencephalography (EEG), Magnetoencephalography (MEG)), and real-time diagnostic tools, as complementary strategies to enhance traditional clinical evaluations. Findings indicate that while clinical assessments remain central to concussion diagnosis, integrating them with advanced imaging and electrophysiological tools can provide more accurate diagnostics and recovery tracking. Biomarkers, although not yet ready for widespread use, hold significant potential for future applications. Further research is required to validate these methods and establish standardized protocols to facilitate their integration into clinical practice. Full article

The rising demand for elderly care amid ongoing nursing shortages has highlighted the limitations of conventional alert systems, which frequently generate excessive alerts and contribute to alarm fatigue. The objective of this study is to develop a hybrid, context-aware nurse alerting framework for long-term care (LTC) facilities that minimizes redundant alarms, reduces alarm fatigue, and enhances patient safety and caregiving balance during multi-person care scenarios such as mealtimes. To do so, we aimed to intelligently suppress, delay, and validate alerts by integrating rule-based logic with Large Language Model (LLM)-driven semantic reasoning. We conducted an experimental study in a real-world LTC environment involving 28 elderly residents (6 high, 8 medium, and 14 low care levels) and four nurses across three rooms over seven days. The proposed system utilizes video-derived skeletal motion, care-level annotations, and dynamic nurse–elderly proximity for decision making. Statistical analyses were performed using F1 score, accuracy, false positive rate (FPR), and false negative rate (FNR) to evaluate performance improvements. Compared to the baseline where all nurses were notified (100% alarm load), the proposed method reduced average alarm load to 27.5%, achieving a 72.5% reduction, with suppression rates reaching 100% in some rooms for some nurses. Performance metrics further validate the system’s effectiveness: the macro F1 score improved from 0.18 (baseline) to 0.97, while accuracy rose from 0.21 (baseline) to 0.98. Compared to the baseline error rates (FPR 0.20, FNR 0.79), the proposed method achieved drastically lower values (FPR 0.005, FNR 0.023). Across both spatial (room-level) and temporal (day-level) validations, the proposed approach consistently outperformed baseline and purely rule-based methods. These findings demonstrate that the proposed approach effectively minimizes false alarms while maintaining strong operational efficiency. By integrating rule-based mechanisms with LLM-based contextual reasoning, the framework significantly enhances alert accuracy, mitigates alarm fatigue, and promotes safer, more sustainable, and human-centered care practices, making it suitable for practical deployment within real-world long-term care environments. Full article

This study presents a novel approach to financial time series forecasting by introducing asymmetric loss functions. This is specifically designed to enhance directional accuracy across major stock indices (S&P 500, DJI, and NASDAQ Composite) over a 33-year time period. We integrate these loss functions into an attention-based Long Short-Term Memory (LSTM) framework. The proposed loss functions are evaluated against traditional metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and other recent research-based losses. Our approach consistently achieves superior test-time directional accuracy, with gains of 3.4–6.1 percentage points over MSE/MAE and 2.0–4.5 percentage points over prior asymmetric losses, which are either non-differentiable or require extensive hyperparameter tuning. Furthermore, proposed models also achieve an F1 score of up to 0.74, compared to 0.63–0.68 for existing methods, and maintain competitive MAE values within 0.01–0.03 of the baseline. The optimized asymmetric loss functions improve specificity to above 0.62 and ensure a better balance between precision and recall. These results underscore the potential of directionally aware loss design to enhance AI-driven financial forecasting systems. Full article

The use of artificial intelligence (AI) in Translation Quality Assessment (TQA) has emerged as an exciting new line of research hoping to explore the potential of this revolutionary technology within the field of translation studies in general and its effect on translator training ecosystem. The aim of this study is to explore how AI’s evaluation of students’ legal translations aligns with instructors’ evaluations and to look at the potential benefits and challenges of using AI in evaluating legal translations tasks. Ten anonymous copies of instructor-graded English-to-Arabic mid-term exam translations were collected from an undergraduate legal translation course at a Saudi university and evaluated using ChatGPT-4o. The system was prompted to detect the translation errors and score the exam using the same rubric that was used by the instructors. A manual segment-by-segment comparison of ChatGPT-4o and human evaluations was conducted, categorizing errors by type and assessing alignment by comparing the scores statistically to determine if there were significant differences. The results indicated a high level of agreement between ChatGPT-4o and the instructors’ evaluation. In addition, paired sample t-test comparisons of instructor and ChatGPT-4o scores indicated no statistically significant differences (p > 0.05). Feedback provided by ChatGPT-4o was clear and detailed, offering error explanations and suggested corrections. Although such results encourage effective integration of AI tools in TQA in translator training settings, strategic implementation that balances automation with human insight is essential. With proper design, training, and oversight, AI can play a meaningful role in supporting modern translation pedagogy. Full article

It is essential to balance energy supply and demand in residential buildings through accurate forecasting of energy use due to varying daily and seasonal residential building loads. This study demonstrates a data-driven Virtual Energy Replication Framework (VERF) to predict the behavior of residential buildings using weather forecast data. The framework integrates supervised machine learning models and time-ahead weather parameters to estimate photovoltaic (PV) power production, heat pump energy consumption, and indoor thermal comfort. The accuracy of prediction models is validated using TRNSYS simulations of a typical household in Saarbrucken, Germany, a temperate oceanic climate region. The XGBoost model exhibits the highest reliability, achieving a root mean square error (RMSE) of 0.003 kW for PV power generation and 0.025 kW for heat pump energy use, with R² scores of 0.94 and 0.87, respectively. XGBoost and random forest regression models perform well in predicting PV generation and HP electricity load, with mean prediction errors of 5.27–6% and 0–7.7%, respectively. In addition, the thermal comfort index (PPD) is predicted with an RMSE of 1.84 kW and an R² score of 0.80 using the XGBoost model. The mean prediction error remains between 2.4% (XGBoost regression) and −11.5% (lasso regression) throughout the forecasted data. Because the framework requires no real-time instrumentation or detailed energy modelling, it is scalable and adaptable for smart building energy systems, and has particular value for Building-Integrated Photovoltaics (BIPV) demonstration projects on account of its predictive load-matching capabilities. The research findings justify the applicability of VERF for efficient and sustainable energy management using weather-informed prediction models in residential buildings. Full article