Focus on privacy‑preserving federated learning and robust, data‑efficient models for vision, speech and medical imaging. Develops layer‑wise DP defenses, adaptive backdoor detection, and synthetic data pipelines to secure AI deployment in distributed, real‑world settings.