The domain of Internet of Things (IoT) has been rapidly developing, consequently expanding the attack surface for malware that exploits vulnerabilities across various hardware architectures and evades conventional static detection techniques. Past frameworks for IoT malware detection and classification often rely on a single data modality or are designed for a specific device architecture, which limits their ability to detect and classify malware in heterogeneous environments. This paper proposes CAMRA, a cross-architecture malware detection and classification framework that combines multimodal data, Deep Reinforcement Learning (DRL), and Federated Learning (FL). Using the CIC-YNU-IoT-MAL-26 dataset, CAMRA fuses network-level (PCAP), hardware-level (SAR), and system-level (STRACE) data streams through cross-modal attention across the ARM, MIPS, MIPSEL, and x86 architectures. A DRL controller adapts the per-class loss weights during training to address class imbalance, and federated learning adapts the centrally trained model across architecture-specific clients while keeping their raw data local. Over ten random seeds, CAMRA achieves a Macro-F1 of 0.752 ± 0.012 (accuracy 98.4% ), and the federated model achieves 0.741 ± 0.008 (accuracy 98.6% ), outperforming every single-modality model and standard baseline. Ablation and paired statistical tests show that the frequency-based class weights and the loss function are the main drivers of performance, with the DRL controller performing on par with fixed class weighting. In holdout experiments, malware from unseen architectures and from an unseen malware family is detected as malicious, while an unseen ARM architecture leads to false alarms on benign data.
Supply-network disruptions can propagate rapidly across interconnected suppliers, plants, ports, carriers, products, and customers, making timely and reliable recovery decisions difficult. Existing data-driven approaches often focus on either prediction or optimization and may therefore overlook the combined effects of network structure, intervention uncertainty, and operational feasibility. This study proposes an integrated decision-support framework for supply-network reconfiguration under cascading disruptions and demand shocks. The framework combines network-aware learning, counterfactual response evaluation, uncertainty assessment, and robust optimization to compare alternative recovery strategies and identify feasible actions under changing disruption conditions. Experiments involving 9,215 logistics orders and 100,000 disruption events demonstrate a normalized weighted score of 0.948, supplier-change F1 of 0.947, policy value of 0.867, and resilience of 0.762. Compared with the historical policy, the proposed framework reduces expected cost by 33.73
Enhancing academic engagement and adaptive learning systems requires an understanding of students’ emotions in the classroom. Conventional facial expression recognition techniques frequently fall short in educational environments because of issues such inadequate lighting, occlusion, changing head positions, and the requirement to record contextual data outside of the face. With three significant innovations, this study suggests a unique framework for robust academic emotion recognition called Context Attentive Cascaded Preprocessing with Dual Stream Emotion Aware Swin Transformer (CADE-Swin) to overcome these constraints. To preserve both fine-grained facial expressions and broader behavioral context, a Context Attentive Multi-Task Cascaded Convolutional Neural Network (CA-MTCNN) preprocessing is first constructed to jointly extract aligned face crops and face-context regions. Second, by combining emotion-aware attention heads, the Dual Stream Emotion Aware Swin Transformer (DE-Swin) architecture prioritizes salient facial and action unit regions while processing face-only and context crops simultaneously. Third, sequence-level reasoning for continuous engagement analysis is made possible by the extension of the DE-Swin model with cross-frame temporal attention. The DAiSEE KAGGLE dataset’s 6000 video segments were used to extract the original images that capture contextual and face signals. Six augmentation techniques are applied to 12,000 original classroom images of DAiSEE dataset to create an augmented dataset of 84,000 images, which guarantees data diversity and generalization. After that, the data is manually annotated into six academic emotion classes: frustrated, bored, engaged, confused, sleepy, and yawning. The proposed CADE-Swin system outperforms current state-of-the-art algorithms in academic emotion recognition with an impressive 96.13
Social media platforms generate vast amounts of textual data that offer critical insights into public mental health. However, effectively analyzing this data for early depression detection requires robust language models that can handle real-world variations. This study addresses the need for scalable and reliable AI-driven tools by exploring the effectiveness of various Large Language Models (LLMs) in classifying mental health status from social media text, aiming to identify architectures that are resilient to shifts in data volume, preprocessing, and domain. A comprehensive comparative evaluation of several prominent LLMs specifically LLaMA-1b, LLaMA-3b, ModernBert, Bert, GPT-2, and Pythia-1b was conducted. The models were systematically trained on subsets of the Sentiment140 dataset (100k and 50k rows) with and without text preprocessing to assess the impact of data volume and cleaning. Generalization was rigorously tested by evaluating all trained models on an independent, unseen, depression-focused dataset (DD-Kaggle). Performance was measured using accuracy, precision, recall, F1-score, and AUC. On in-distribution validation sets, a clear performance hierarchy emerged, with the larger LLaMA-3B model achieving a peak validation accuracy of 92.33
Efficient generation of test reports during engineering system testing is very important when it comes to diagnosing bugs or defects and determining the overall quality of a product, especially when making decisions regarding such products. With the explosion of test data due to increased complexity, especially when it comes to test data crowdsourcing, generating these reports manually becomes more complex and cumbersome. In order to deal with this problem, an optimized test report generation using machine learning algorithm will be developed in order to assist in generating an optimized test report. This model is called Efficient Monarch Butterfly Optimized Light Gradient Boosting Machine (EMB-LGBM).Machine learning dynamically adjusts crossover and mutation probabilities based on population metrics such as average fitness, best fitness, and diversity. The model adapts quickly during early iterations while maintaining exploration capability in later stages, effectively avoiding local optima. The model is validated using publicly available engineering test datasets, and its performance is benchmarked against various traditional optimization approaches. Results demonstrate that EMB-LGBM significantly improves report diversity, severity relevance, convergence speed, and population uniformity while reducing execution time. The proposed model delivers more informative and well-structured reports with greater efficiency and achieves a 5-fold cross-validation accuracy of 92.14
