• Investigated quantization and bit-depth reduction for optimizing KNN and SVM on medical datasets, achieving reduced computational cost with minimal performance degradation.

• Demonstrated that optimization success is highly context-dependent, influenced by dataset complexity, model architecture, and the specific transformation techniques applied.

• Developed and evaluated KAN-based baseline, robust, and ensemble models to enhance indoor Wi-Fi positioning accuracy under adversarial attack conditions.

• Adversarial training and ensemble strategies significantly reduced positioning error, demonstrating improved system resilience against spoofing and signal manipulation threats.

• Proposed Hyb-KAN ViT, replacing MLPs with Efficient-KAN and CUDA-optimized Wavelet-KAN for interpretable, GPU-efficient spectral feature learning.

• Achieved 84.5% ImageNet Top-1, 52.3 mIoU on ADE20K, and strong COCO detection—surpassing state-of-the-art with fewer parameters.

• Proposed a hybrid KAN-GCN architecture with attention and composite loss, achieving 4.2–6.8% accuracy gain over baselines on CIFAR-10/100 and 5.7% boost on low-data ImageNet-1K. 

• Outperformed ResNet-152, ViT-B/16, and graph models with 1.8× faster inference and interpretable spline-based layers, enabling real-time, robust vision understanding via spatially-aware deep learning.

• Applied U-Net for semantic segmentation of satellite imagery, achieving high-resolution landform detection with strong feature extraction and regularized training.

• Demonstrated U-Net’s practical relevance in autonomous driving, disaster response, and land use planning through accurate pixel-wise topographical segmentation.

• Introduced NaNs in cholesterol values of heart disease data, then used ANN with interpolation to predict missing values accurately.

• Compared predicted vs. original values via visualizations, demonstrating ANN's effectiveness in medical data imputation and preserving feature relationships.

• Developed a Res-GAN model integrating residual learning and GANs to detect subtle anomalies in Wi-Fi traffic with high precision.

• Achieved robust anomaly detection, showing resilience to adversarial evasion, enabling scalable, real-time Wi-Fi security for wireless network infrastructures.

• Designed an AI-based device to identify medicines using QR and camera, delivering usage info via multilingual audio and display.

• Integrated weight sensors and an NLP system to alert on low stock and support safe medication in elderly and nursing care.

View Certificate