饶子扬 / Ziyang Rao

I am a PhD candidate in Artificial Intelligence at HKUST (Guangzhou), advised by Prof. Hui Xiong, and a research intern at the X-Lab of AI² Robotics.

My work centres on embodied AI: I study the geometry of flow-matching policies and ask when their actions can be trusted, so that vision-language-action models fail loudly rather than silently. This grows out of a wider interest in uncertainty quantification and mechanistic interpretability — what representations a model actually forms, and what its internals reveal about the reliability of its outputs. My earlier research also spans information retrieval and multimodal LLMs.

I earned my MPhil in AI at the same lab, co-advised by Prof. Hui Xiong and Dr. Xuming Hu, where alongside research I was actively engaged in AI startups and knowledge transfer.