I am an applied scientist at Amazon working on foundational AI models. I received my Ph.D. in Computer Engineering from Northwestern University, where I was advised by Prof. Qi Zhu. I obtained my B.S. in Electrical Engineering from Zhejiang University in 2019.
My long-term research goal is to build trustworthy AI agents, with an emphasis on:
My work sits at the intersection of machine learning, autonomy, and safety, focusing on building reliable decision-making systems in interactive environments. Recently, I have been working on knowledge distillation and reinforcement learning to develop state-of-the-art LLMs and agents for real-world applications at scale.
Demonstrates that LLM-based embodied agents can be vulnerable to backdoor attacks and provides evaluation and defenses toward safer LLM-driven decision-making.
We revisit the trade-off between general and domain-specific capabilities in supervised fine-tuning and introduce a new loss design that more effectively balances the two.
Introduces a kinematics-aware latent SDE model that generates physically consistent and diverse future trajectories for autonomous driving.
Uses semantics-guided adversarial training to improve trajectory prediction robustness under noisy and adversarial agent behaviors.
Bridges safe control and RL by enforcing hard safety constraints through soft barrier functions under stochastic dynamics.
NeurIPS, ICLR, ICML, ECCV, ICCV, CVPR, IROS, ICRA, AAAI, AISTATS, IV
TMLR, RA-L, TMM, TNNLS, TCAD, TCPS, TIV, IEEE JSAC