82% accuracy
Built an end-to-end deep learning pipeline for EEG signal classification, handling noisy real-world data and improving model reliability through robust preprocessing and validation, achieving 82% accuracy.
SDN Traffic Analysis • CNN-LSTM • Anomaly Detection
Designed and implemented a deep learning system for detecting DDoS attacks in SDN environments, learning complex temporal traffic patterns using hybrid CNN-LSTM architectures and improving detection performance on imbalanced network data.
BLIP • Image Captioning • Semantic Scene Understanding
Built a vision-language system using BLIP to analyze scene safety by generating contextual captions from images, enabling semantic understanding of environments beyond traditional image classification.