Earthquake-induced landslide susceptibility zonation (EQ-LSZ) mapping commonly relies on the landslide inventory. However, in many seismically active regions, the lack of comprehensive landslide inventories poses challenges for susceptibility mapping. This study focuses on the Indian Himalayan region, specifically Sikkim, which experienced a series of earthquake-induced landslides, including those triggered by the 2011 Sikkim earthquake. The research develops EQ-LSZ maps using both inventory-inclusive methods—statistical models (Frequency Ratio, FR) and machine learning models (Random Forest, RF)—and an inventory-exclusion method (Newmark Displacement, ND). The study also performs a comparative analysis of these models. Input data include landslide inventory, seismic parameters (peak ground acceleration), landslide-controlling factors (topography, lithology, distance to faults, hydrology, distance to roads, land use/land cover, geomorphology, soil properties), factor of safety, and yield acceleration. Model performance was evaluated using success and prediction rate curves. The FR method achieved 85.34
The rapid adoption of the Industrial Internet of Things (IIoT) in smart manufacturing and critical infrastructure has significantly increased the exposure of industrial networks to sophisticated cyber threats. Ensuring secure communication and reliable threat detection in IIoT environments has therefore become a critical challenge. This study proposes an intelligent Cyber Threat Detection and Response System that integrates a Hybrid Deep Neural Network with the Grey Wolf Optimizer to enhance security in IIoT networks. The proposed framework utilizes CyberTec IIoT Malware Dataset (CIMD‑2024) on Kaggle containing network traffic characteristics, device communication patterns, and anomaly indicators. A comprehensive data preprocessing phase is employed, including noise removal, normalization, and missing value handling, to improve data quality and model reliability. The hybrid deep learning architecture combines Convolutional Neural Networks for spatial feature extraction with Long Short-Term Memory networks to capture temporal dependencies in network behavior. Additionally, a dual-attention mechanism is incorporated to emphasize significant spatial and temporal features, thereby improving the accuracy of cyber threat classification. The Grey Wolf Optimizer is applied to optimize key hyperparameters such as learning rate, dropout rate, and batch size, leading to improved model performance. Experimental results demonstrate that the proposed model achieves an accuracy of 96.5
Weather forecasting is vital for sectors such as disaster response, agriculture, and risk management, and even influences daily decision-making. Reliable forecasts reduce potential losses, strengthen preparedness, and help in planning routine activities. In recent years, artificial intelligence techniques—particularly deep learning—have become more prominent for improving prediction accuracy. This paper reviews the application of AI-based models in weather forecasting and compares them across several technical aspects. Models are evaluated based on their accuracy, computational needs, and suitability across different climatic regions. The study also points out key challenges in data balance, resource use, and model transparency. Overall, it provides an updated outlook on AI-assisted forecasting and discusses ways to develop more efficient, reliable, and interpretable prediction systems.
Traditional computational fluid dynamics and physics-informed neural networks (PINNs) often suffer from high computational cost, mesh sensitivity, and reduced accuracy for strongly nonlinear and time-dependent flows. To address these limitations, we propose PhysicsFormer, a simple and efficient Transformer-based physics-informed neural network framework for complex fluid flow simulations. The proposed architecture employs encoder–decoder multi-head attention to capture long-range temporal dependencies and enhance spatio-temporal information propagation. Unlike conventional multilayer perceptron-based PINNs, PhysicsFormer utilizes pseudo-sequential spatio-temporal representations together with a dynamics-weighted loss formulation to improve convergence, stability, and predictive accuracy. Owing to its lightweight architecture and parallel learning strategy, the proposed framework achieves faster training and lower computational cost than existing Transformer-based PINN models. The performance of the proposed framework is demonstrated on the convection equation, Burgers' equation, lid-driven cavity flow at Re=100, and inverse Navier–Stokes and flow reconstruction problems for flow past a circular cylinder at Re=100 and Re=3900. For the inverse Navier–Stokes problem at Re=100, the proposed framework simultaneously reconstructs the flow field and identifies governing equation parameters with nearly 0% absolute error under both clean and noisy data conditions. Furthermore, for the high-Reynolds-number case at Re=3900, PhysicsFormer accurately reconstructs the velocity and pressure fields using only 25 spatial measurements per snapshot over 100 temporal snapshots. The obtained results demonstrate that PhysicsFormer provides an accurate, robust, and computationally efficient framework for complex time-dependent fluid flow problems.
This study investigates the efficacy of transformer-based Bidirectional Encoder Representations from Transformers (BERT) model and Deep Averaging Network version of Universal Sentence Encoder (USE) model for Automatic Short Answer Grading (ASAG). As educational institutions move toward automated Learning Analytics, the challenge remains in aligning high dimensional machine embeddings with nuanced human assigned rubrics. Using a stratified sample from Question Set 6 of the HP:SAS dataset, student responses were transformed into high-dimensional vectors and compared against five model answers using Cosine Similarity. The core technical contribution of this work lies in the comparative evaluation of two scoring frameworks: a traditional linear mapping and non-linear mapping protocol designed to capture the exponential variance in human grading. Experimental results indicate that non-linear mapping significantly outperforms linear methods in predicting scores on a scale of 0 to 3. Comprehensive error analysis reveals specific performance gains in handling outlier responses and varied sentence structures. This work provides a reproducible roadmap for researchers to improve grading accuracy through non-linear score calibration. This offers a scalable solution for the 'public problem' of high-volume assessment in modern digital education.
