Browse Articles

Classifying mental stress from eye tracking data: deep learning approaches for out-of-the-lab conditions

Abstract Eye-tracking signals such as pupil diameter and gaze behavior have been widely used for stress detection, yet most approaches rely on task-specific features, controlled laboratory settings, or multimodal sensor combinations, limiting scalability in less controlled environments. This work investigates whether unimodal eye-tracking time-series data can support task-agnostic stress detection beyond static laboratory tasks. We analyze stress classification across two complementary datasets: a virtual reality goalkeeper task with moderate visuomotor activity and stable recording conditions, and a virtual job interview dataset reflecting less controlled settings with uncalibrated signals. The results show that these signals alone contain informative patterns related to stress-associated autonomic and oculomotor responses. Under favorable conditions, performance reaches up to $${95.98}\%$$ macro-averaged F1-score. At the same time, performance varies substantially across datasets, indicating that effective learning depends strongly on data quality, calibration, signal characteristics, and task design. Overall, the findings demonstrate the potential of unimodal eye tracking as a lower-burden alternative to more complex multimodal systems, while highlighting that reliable stress detection is fundamentally conditioned by the interplay of data, signal representation, and modeling approach.

Joint wavelet decomposition of predictors and target variables for drought forecasting

Abstract We propose a leakage-free wavelet-based methodology for drought prediction that enables both competitive forecasting performance and improved interpretability of predictor–target relationships. Unlike standard time-series decomposition approaches, the wavelet transform is applied within moving windows of the past values, preventing the use of future information during model training. The method merges local and large-scale climate predictors across multiple temporal scales and is evaluated using 76 years of drought data from Illinois, USA. Predictive performance is assessed using RMSE, SMAPE, MASE, R 2 , and directional accuracy (DA), including uncertainty calculated with seasonal bootstrap confidence intervals. The proposed model frequently outperforms naïve and univariate wavelet baselines, achieving improvements of up to 26.5% in RMSE and 4.95% in SMAPE and higher directional accuracy in diverse scenarios compared to several benchmark methods. Through decomposing both predictors and target variables, the methodology helps characterize how different frequency components are associated with drought evolution, providing a more interpretable representation of hydroclimatic interactions and additional insights into the scale-dependent dynamics of drought processes.

Learning-based orchestration for low-latency AI deployment in hybrid cloud–edge platforms

Abstract The rapid growth of AI-driven applications in hybrid cloud–edge environments poses substantial challenges to ensuring low latency, high throughput, and effective resource utilization. Conventional deployment models, which are typically fixed or policy-driven, are not sufficiently flexible to respond dynamically to changing workloads and heterogeneous hardware environments. In this work, we introduce and analyze a resource-conscious deep learning-based scheduling system for managing the deployment of AI models on distributed cloud edges. The framework improves inference performance by leveraging real-time system telemetry and model features generated by benchmarks, while maintaining quality of service (QoS) compliance. The proposed system uses a fully connected neural network trained on structured features derived from the MLPerf Inference Benchmark, including compute complexity, memory footprint, and input dimensions. It is guided by real-time data from a hybrid infrastructure (NVIDIA A100/V100 GPUs and Jetson Xavier edge devices) to inform scheduling. Four MLPerf inference workloads – ResNet 50, BERT, SSD ResNet34, and DLRM – were tested and contrasted across various batch sizes and latency thresholds. Generalization experiments with unseen models such as GPT 2 and YOLOv5 yielded > 90% success rates in deployment, with the latency reduction and throughput gain results as presented above. Results of the generalization experiments with unseen models, including GPT 2 and YOLOv5, demonstrated deployment success rates > 90% for the various profiling conditions evaluated, with the latency reduction and throughput improvements as shown above. The results show that learning-based orchestration can be used to deliver space- and resource-aware orchestration solutions that are adaptive for low-latency deployment of AI services in hybrid cloud edge systems, but the effectiveness of the solution will depend on the representativeness of the profiling data and similarity of training and deployment environments.

BlockFedX: a cross-domain federated learning system with explainability, anomaly detection, and tamper-evident logging

Clinical benefit of palbociclib retreatment after abemaciclib exposure in hormone receptor positive, HER2 negative metastatic breast cancer

PM1.0 marker-based diagnostic ratios and PMF-based source apportionment for quantifying oxidative potential in a major point-source region

Lightning initiation on the exoplanet K2-18b

Abstract Recent JWST data provide a basis for investigations into exoplanetary atmospheres at novel high precision. Weather and climate on these planets, along with the implications of the observed atmospheric composition for extraterrestrial life, remain hotly debated. Using photochemical modelling data based on JWST observations as inputs for a thoroughly validated plasma simulation code, we present the first simulations of streamers, precursors of lightning, in an exoplanetary atmosphere. We find the electrical breakdown field magnitude in this atmosphere to be approximately half that of Earth, indicative of an atmosphere more susceptible to lightning than our own. We show that lightning can incept on K2-18b in all three tested atmospheric configurations, and that variations in atmospheric composition, in particular in the presence of water vapour, affect the rate of streamer inception.

Incorporating game-based language learning into vocational education: a quasi-experimental study

A scalable solution for multi-symptom disease prediction and lifestyle recommendations using machine learning

Abstract Accurate prediction of diseases from multiple co-occurring symptoms remains an important challenge in intelligent healthcare systems. Machine-learning models can support early symptom-based risk screening; however, their clinical use is limited by dataset quality, lack of external validation, interpretability concerns, and the risk of overstated diagnostic claims. This study presents a mobile decision-support prototype for multi-symptom disease prediction and rule-based lifestyle recommendation. The experiments were conducted using the publicly available Kaggle Medicine Recommendation System Dataset, which contains 4,920 symptom-based records spanning 41 disease classes, along with supporting files for disease descriptions, precautions, medications, dietary suggestions, and workout recommendations. Seven supervised machine-learning models were evaluated, including Decision Tree, Random Forest, Naive Bayes, Logistic Regression, XGBoost, Support Vector Machine, and K-Nearest Neighbors. The best-performing models achieved high classification accuracy under the adopted evaluation protocol. To avoid overinterpretation, these results are reported as benchmark performance on a secondary public dataset rather than evidence of clinical diagnostic validity. The Android-Flask prototype links the predicted disease class to a lookup-based recommendation layer that retrieves disease-associated information from supporting datasets. The system should therefore be interpreted as a decision-support and educational prototype, not as a clinically validated diagnostic or prescribing tool. Future work should include external validation in independent clinical cohorts, clinician assessment of recommendations, robustness testing under missing or noisy symptom data, and broader evaluation across real-world healthcare settings.

Characterization and comparative analysis of the complete chloroplast genomes of twelve Allium species from Kazakhstan

Dual engagement of Spike and ACE2 by annexin A5 contributes to pleiotropic SARS-CoV-2 inhibition

Abstract COVID-19 is primarily a respiratory tract infection caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), which can enter host airway epithelial cells through angiotensin-converting enzyme 2 (ACE2) receptors. Emerging SARS-CoV-2 variants are a roadblock to irradicating the disease. Given recombinant human annexin A5 (Anx5) inhibits proinflammatory responses, improves survival in sepsis models and binds to several receptors and lipids, we hypothesized that Anx5 impedes SARS-CoV-2 viral entry and lessens disease severity. To assess SARS-CoV-2 Spike-receptor binding domain (RBD) and ACE2 interactions with Anx5, the recombinant proteins were expressed and isolated, and interactions were evaluated using solution NMR, microscale thermophoresis, steady-state fluorescence and size-exclusion chromatography coupled to multi-angle light scattering. Remarkably, we found that Anx5 binds to both the Spike-RBD and ACE2. Further, Anx5 induced extensive 15 N-Spike-RBD amide resonance broadening, most concentrated in the region that engages with ACE2 and least enriched on the opposite face, highlighting the ACE2 interface as the primary site of Anx5 binding. Finally, we demonstrate that Anx5 inhibits SARS-CoV-2 pseudoviral entry and reduces viral burden while enhancing survival in SARS-CoV-2-infected mammalian cells. Collectively, our findings demonstrate that Anx5 directly engages both Spike-RBD and ACE2, inhibits SARS-CoV-2 pseudoviral entry and reduces viral burden while enhancing survival of infected mammalian cells, supporting further investigation of Anx5 as a multi-functional anti-SARS-CoV-2 therapeutic.

Ecological and methodological insights from genetic and coprological profiling of gastrointestinal communities in wild howler monkeys

FL-TWIN: a unified federated learning system for intrusion detection with digital twins modelling

How internet use and policy awareness affect village cadres’ pro-environmental behaviors in China: a social cognitive theory perspective

Optimized doubly-curved ARH-based metastructure to augment specific vibrational energy conversion using a hybrid constraint multi-objective genetic algorithm

Cross-sectional and prospective associations between multidimensional psychological distress and urogenital disorders: findings from the UK biobank

Advanced glycation end products and the CALLY index reflect inflammatory burden in familial Mediterranean fever

Dynamic monitoring of railway bridges via coupling digital twins with deep reinforcement learning

Genome-wide DNA methylation profiling during metabolic dysfunction-associated steatohepatitis-related hepatocarcinogenesis in patients in Japan and the United States

A novel hybrid clustering approach for robust ramp event characterization

Abstract Large power fluctuations in a brief amount of time, or ramp events, are an increasing concern for grid operators due to the rise in renewable energy generation and the unreliable hour-ahead predictions. To balance these ramp events, grid operators need to be aware of their anticipated occurrence intervals and range. Prior studies used binary ramp event categorization, whereas other studies employed non-causative classification techniques. Existing clustering methods, Z-score and k-means, have strengths but distinct limitations. To address these, this paper introduces the ZK-means hybrid approach, integrating Z-score normalization with k-means partitioning, forming a centroid-based clustering algorithm to enhance adaptability, noise resistance, and interpretability in ramp classification. Its need arises from the growing demand for accurate and efficient ramp analysis to support reliable grid operation and forecasting. Two comparison phases for the ZK-means approach were conducted: First, it was evaluated against its constituent methods to assess the benefits of their combination; second, it was compared to the density-based spatial clustering of applications with noise (DBSCAN) algorithm to verify its robustness and general applicability. Although DBSCAN can capture local variations in data, it produced inconsistent cluster numbers and required frequent parameter tuning across the ten years. In contrast, ZK-means achieved more stable clustering patterns and lower within-cluster variance, demonstrating superior reliability for long-term ramp event characterization. The new categorization method is applied to a real case study, and the results reveal that the new hybrid method offers significant improvements in the quality, robustness, and interpretability of the clustering process and its resulting cluster characteristics, as it combines the stability of normalization with the scalability of k-means, offering a robust and practical solution for large-scale, high-dimensional clustering. While this new method does entail a slight increase in time-speed, computational complexity, and energy consumption compared to its constituent methods, it remains faster than DBSCAN and the enhanced insights it provides offer critical advantages for effective grid management.