Muhammad Umair

The growing demand for deploying deep learning (DL) models for cyber-attack detection on consumer devices is constrained by their limited memory and computational resources. High-performing DL models are often too resource-intensive for on-device deployment. Knowledge distillation (KD) offers an alternative by transferring knowledge from a complex teacher model to a smaller, lightweight student model, reducing model size and inference time with minimal accuracy loss. In this paper, a KD-based framework is utilized to design and train both teacher and student models for intrusion detection using the N-BaIoT and CIC-IDS 2023 datasets. The student model achieves comparable performance to the teacher, with accuracies of 99.87% and 98.71% on the utilized datasets. It maintains accurate classification across various attack classes, achieving F1-scores above 0.98, while reducing computational complexity from 263k to 120k trainable parameters, model size from 1.00 MB to 471.67 KB, and inference time from 48 ms to 34 ms. Moreover, SHAP-based explainable AI techniques are employed to interpret both models, demonstrating that the distilled student model retains key feature attributions of the teacher model.

Timely prediction of the State of Health (SoH) of lithium-ion batteries is important for battery management and longevity. Traditional centralized deep learning models have shown promising results, but they raise concerns related to data privacy, as data needed to be collected and trained on a single node. This study addresses this challenge by utilizing both centralized (i.e., deep learning) and decentralized (i.e., federated learning) approaches for SoH prediction. The NASA battery dataset, containing charging and discharging cycles, is used for model training and evaluation. Three deep learning architectures 1D Convolutional Neural Networks (CNN), CNN plus Long Short-Term Memory (LSTM), and CNN plus Gated Recurrent Units (GRU) are used in the centralized approach. The 1DCNNmodel outperforms, demonstrating strong predictive capabilities, thus for decentralized learning (i.e., federated learning), the 1D CNN model is utilized with federated averaging technique across five clients, allowing for local training without sharing raw data. Obtained results shows that the highest testing RMSE (0.666) and MAPE (0.980) are observed during decentralized learning, while the centralized approach shows varying performance across different batteries. The decentralized approach effectively balances performance and privacy, highlighting the reliability of federated learning in SoH prediction for lithium-ion batteries.

Early detection of plant diseases is essential for effective crop disease management to prevent yield loss. In this study, we developed a methodology for classifying diseases in rice leaves using four deep learning models and a dataset with 2658 images of healthy and diseased rice leaves. Four models, namely LeafNet, Modified LeafNet, MobileNetV2, and Xception, were compared. The Modified LeafNet model involved updates to LeafNet’s architectural parameters, whereas transfer learning techniques were applied to the MobileNetV2 and Xception pretrained models. The optimal hyperparameters for training were determined by considering several factors such as batch size, data augmentation, learning rate, and optimizers. The Modified LeafNet model achieved the highest accuracies of 97.44% and 87.76% for the validation and testing datasets, respectively. In comparison, LeafNet obtained 88.92% and 71.84%, Xception obtained 88.64% and 71.95%, and MobileNetV2 obtained 82.10% and 67.68% for the validation and test accuracies on the same datasets, respectively. This study contributes to the development of automated disease classification systems for rice leaves, thereby leading to increased agricultural productivity and sustainability.

Covid-19 is unpredictable evolutionary discipline which requires continuous advancements for its appropriate Detection & Classifications which can be helpful for bio-medical stream. In this research, two dimensions are covered that is detection & classification using self-proposed 2 stage learning detector. Detection of different variants of Covid-19 are performed using images of CT-Scan and X-Rays of effected lungs. Furthermore, classification of different variants is carried out. Dataset of 27000 indigenous images were used for detection & classification purposed. Moreover, in depth survey & comparison is carried out with state-of-the-art Yolo v5 single state detector & Faster R-CNN 2 stage detector. Accuracy analysis of self-proposed 2 stage detector was 91.66% & 87.9% for detection & classification in comparison with YOLOv5 which had accuracy of 92.8% & 87.175% for detection & classification. Moreover, in comparison with Faster R-CNN which had accuracy of 94.8% & 87% The training analysis was performed on Nvidia T4 (16GB GDDR6). Self-proposed MNN-2 superseded Yolov5 & faster R-CNN in real time video analysis with least real time rate at FPS 30 at duration of 72 min video.

The COVID-19 outbreak began in December 2019 and has dreadfully affected our lives since then. More than three million lives have been engulfed by this newest member of the corona virus family. With the emergence of continuously mutating variants of this virus, it is still indispensable to successfully diagnose the virus at early stages. Although the primary technique for the diagnosis is the PCR test, the non-contact methods utilizing the chest radiographs and CT scans are always preferred. Artificial intelligence, in this regard, plays an essential role in the early and accurate detection of COVID-19 using pulmonary images. In this research, a transfer learning technique with fine tuning was utilized for the detection and classification of COVID-19. Four pre-trained models i.e., VGG16, DenseNet-121, ResNet-50, and MobileNet were used. The aforementioned deep neural networks were trained using the dataset (available on Kaggle) of 7232 (COVID-19 and normal) chest X-ray images. An indigenous dataset of 450 chest X-ray images of Pakistani patients was collected and used for testing and prediction purposes. Various important parameters, e.g., recall, specificity, F1-score, precision, loss graphs, and confusion matrices were calculated to validate the accuracy of the models. The achieved accuracies of VGG16, ResNet-50, DenseNet-121, and MobileNet are 83.27%, 92.48%, 96.49%, and 96.48%, respectively. In order to display feature maps that depict the decomposition process of an input image into various filters, a visualization of the intermediate activations is performed. Finally, the Grad-CAM technique was applied to create class-specific heatmap images in order to highlight the features extracted in the X-ray images. Various optimizers were used for error minimization purposes. DenseNet-121 outperformed the other three models in terms of both accuracy and prediction.