Zilei Zoe Shao

I am a Ph.D. student in Computer Science at the University of California, Los Angeles, advised by Prof. Guy Van den Broeck in the StarAI Lab. I go by Zoe.

My research integrates tractable probabilistic models into modern neural architectures to improve the training efficiency, inference quality, and controllability of deep generative models. I also develop optimization techniques for probabilistic and generative models, spanning parameter learning, low-variance gradient estimation, and differentiable routing in large-scale architectures. In a related line of work, I study tokenization in Large Language Models (LLMs) as a neurosymbolic problem, showing that non-canonical tokenizations of the same input can substantially change model behavior, with the broader goal of making these large-scale systems more tractable and reliable.

Before my doctoral studies, I earned my Bachelor of Science degree in a joint major of Computer Science and Mathematics at Harvey Mudd College, where I was deeply involved in bounding-box inference within the context of model-based Reinforcement Learning under the supervision of Prof. Erin Talvitie.