Privacy attack contributions: Context-Aware MIAs against Pre-trained LLMs (EMNLP'2025), Trap weights (Euro S&P'2023), etc.
Secure and privacy-preserving AI: A Loss for Differentially Private Deep Learning (PETS'2023), Differentially Private Graph Neural Networks (USENIX'2022), etc.
Work featured in press outlets such as Thurrott and TuringTop10.
Led development of Nebula: a practical, differentially private, auditable, and efficient product analytics system.