Research Case Studies

A multimodal deep learning system fusing RGB-D vision with GelSight tactile sensor data, enabling robotic arms to grasp and manipulate unseen deformable objects with near-human dexterity — without hard-coded object priors.

A real-time host-based intrusion detection system that constructs OS system-call provenance graphs and deploys a heterogeneous Graph Neural Network to identify novel zero-day exploits without relying on any known attack signatures.

A resilient autonomous multi-drone framework combining Byzantine-fault-tolerant consensus with adversarially trained sensor validation — maintaining coordinated swarm operation under simultaneous GPS spoofing and full-spectrum RF communication jamming.

A self-supervised deep learning framework that classifies brain tumors from MRI scans while providing explainable AI outputs that clinicians can actually trust and interpret.

A knowledge distillation framework that compresses large CNN models into lightweight versions deployable on edge devices — achieving near-identical accuracy at a fraction of the energy cost.

Evaluated adaptive padding strategies for post-quantum TLS handshakes against traffic fingerprinting attacks — measuring the privacy-bandwidth trade-off in real network conditions.

An attention-based CNN for classifying gastric cancer from endoscopic images, with integrated XAI providing visual explanations for each classification decision.