Rosie Zhao

About

Rosie Zhao is a Computer Science Ph.D. student with the ML Foundations Group advised by Boaz Barak and Sham Kakade. Her main research interests include the science of deep learning, understanding language model training, and reinforcement learning. Zhao obtained her B.Sc. and M.S. from McGill University, where she worked on representation learning and state abstractions in reinforcement learning. Previously, Zhao has interned at DeepMind and ServiceNow Research. In her free time, Zhao enjoys rock climbing, hiking, and playing board games.

Research Focus

Zhao’s research focuses on the scientific study of deep learning. In particular, she is interested in understanding the training dynamics of deep models with respect to factors such as implicit bias, random variation, and scale. Her current research aims to improve and deepen our understanding of training language models, both in pre-training and in fine-tuning using preference learning algorithms.