Upcoming Event: Center for Autonomy Seminar

Trustworthy and Explainable AI for Real-World Interaction

Hanjiang Hu,

Abstract

Artificial intelligence systems are increasingly deployed in the real world, where they must perceive and interact under uncertainty. Despite strong empirical performance, current AI frameworks often lack guarantees of optimality, robustness, and safety, particularly when exposed to physical uncertainty and long-horizon interaction. In this talk, I present my research on trustworthy and explainable AI for real-world interaction, centered on a neuro-symbolic perspective that incorporates structural and physical priors directly into AI frameworks. By leveraging information-theoretic modeling, low-dimensional manifold structure, and dynamical system principles, my work enables built-in explainability for optimal sensor design for autonomous vehicles, certifiable robustness of robot perception under physical perturbations, and safe multi-turn interaction of large language models. I conclude by outlining future directions toward safe embodied and physical AI, trustworthy AI agents beyond robotics, and next-generation neuro-symbolic structure learning, with the long-term goal of building AI systems that can be reliably deployed in real-world safety-critical environments.

Biography

Hanjiang Hu is a final-year Ph.D. candidate in Electrical and Computer Engineering at Carnegie Mellon University, where he also earned an M.S. in Machine Learning. Prior to that, he received his M.S. and B.Eng. from Shanghai Jiao Tong University. His research focuses on provable guarantees in learning-enabled systems at the intersection of machine learning, control theory and formal verification. He develops neuro-symbolic methods for explainable, robust, and safe AI in the safety-critical applications of autonomous driving, robotics and large language models. His work has been recognized as the DAAD AINet Fellow on Explainable AI, the ASME DSCD Rising Star and IEEE CSS Rising Star on Security and Privacy.