Anurag Singh

The use of Internet of Things (IoT) devices has grown rapidly in recent years. This has increased the need for intrusion detection systems that are accurate, efficient, and easy to understand. Traditional signature-based detection methods are not well suited for modern IoT networks, as attack patterns change quickly. At the same time, many deep learning approaches require high computation and are difficult to interpret, which limits their use on resource-constrained devices.This work presents an intrusion detection framework that combines autoencoder-based feature learning with classical machine learning classifiers. An unsupervised autoencoder is used to learn compact nonlinear representations of network traffic. This reduces the number of features while keeping information that is important for detecting attacks. Interpretability is handled using an encoder weight aggregation method, which helps explain how input features affect the learned representation. To deal with class imbalance, Borderline-SMOTE is applied before splitting the data. In addition to experiments, theoretical analysis is used to study generalization, feature efficiency, and computational cost.The framework is evaluated on the BoT-IoT and UNSW-NB15 datasets. Among all tested models, the Autoencoder–Random Forest combination performs the best. Using only 16 latent features, the model achieves an F1-score of 0.999 and an accuracy of 99.95% on the BoT-IoT dataset. On the more challenging UNSW-NB15 dataset, it reaches an F1-score of 0.809 with an accuracy of 80.95%. These results show that effective and interpretable intrusion detection is possible without using deep or computationally heavy models, making the proposed approach suitable for IoT and edge computing environments.

Many videos over the Internet contain violent and explicit content. As the Internet becomes more accessible, these videos are now available to everyone. It is important to moderate such content as it can have detrimental effects on the mental wellbeing of people, especially children and teenagers. An automated system can prove useful for the efficient analysis of video content. Inspired by the success of Multi-Headed Self-Attention (MHSA) in 2D CNNs, we propose a novel deep-learning architecture by extending MHSA to 3D networks. We develop a framework that utilizes the Res3D network and attention-augmented 3D convolution layers. The proposed model achieves state-of-theart performance; experiments on the Hockey Fights, the Movie Fights, and the Real Life Violence Situations datasets show accuracies of 98.3 ± 0.6%, 99± 1.2% and 96.85 ± 0.66%. An ablation study was conducted to prove the usefulness of attention augmentation in convolutional neural networks for violence detection.

Accurate forecasting of the Indian Ocean Dipole (IOD) and El-Niño-Southern Oscillation (NINO3.4) is crucial for understanding regional weather patterns in the Indian subcontinent. Addressing the challenges associated with IOD and NINO3.4 prediction, a robust multi-task autoregressive deep learning model is introduced for precise forecasting of these indices and key grid projections sea surface temperature (SST), surface-level pressure gradient (SLG), and horizontal wind velocity (U-Comp) over a short to mid-term window (20 months). Utilizing spatiotemporal (SST, SLG, U-Comp) and temporal (IOD and NINO3.4) modalities, the proposed model predicts future IOD and NINO3.4, as well as SST, SLG, and U-Comp, in an autoregressive scheme. The multi-task learning component regularizes the model, effectively capturing the evolving dynamics of global patterns conditioned on IOD and NINO3.4. The comprehensive evaluation explores various task settings, including a duo-setting that predicts IOD or NINO3.4 with spatiotemporal information, showcasing notable proficiency. In a multi-task environment, where both temporal IOD, NINO3.4, and spatiotemporal SST, SLG, U-Comp are predicted, the model successfully forecasts IOD and NINO3.4 indices alongside grid projections with modest accuracy in root mean square error (RMSE). To enhance the model’s interpretability regarding spatiotemporal dynamics, a tailored version of Grad-CAM is employed, providing critical insights for climate prediction. This research advances climate prediction models, offering a comprehensive framework with significant implications for informed decision-making in the Indian subcontinent’s climatic context.

Complex networks and network reconstruction have now become powerful tools for exploring relationships and interactions within systems across various fields to identify patterns and increase predictive accuracy. It provides a structured framework to investigate the relations or connections among individual components or entities within a system. Recently, the importance of the healthcare prediction model in life‐saving efforts has increased. A model is proposed for health prediction that uses network reconstruction methods to build a network from available information, representing features as nodes and the relationships between them as edges. Subsequently, a method is introduced to calculate the value of the decision parameter () for predicting an individual's health status. The proposed model shows substantial improvement over the current state of the prediction model. The first aim of the proposed model is to classify an individual into their appropriate class properly. Another contribution of the proposed model is to measure the factors that are responsible for classifying an individual into a class that shows its significance and impact over the existing state‐of‐the‐art. It provides a new dimension to the prediction mode, emphasizing the importance of identifying critical features and their interdependencies for personalized health diagnostics.