Final-year Computer Science student at Sathyabama Institute of Science and Technology, Chennai.
I keep ending up at the same question from different angles: how do you build ML systems that are actually secure and trustworthy, not just accurate on a benchmark?
network-intrusion-detection LSTM trained on real network traffic (CICIDS2017) — 99.75% accuracy on DDoS. The interesting part: SHAP showed the model was basically just counting ACK flags. Works great on DDoS. Completely blind to brute force and web attacks. Explainability exposed what accuracy hid.
pqc-federated-learning Federated learning across 5 simulated hospitals, secured with the new NIST post-quantum standards. Faster than RSA, Byzantine-robust, differentially private. Paper under review at ICISS 2026.