I am a researcher in the field of machine learning and artificial intelligence. I obtained my BASc in Computer Engineering and Mathematics at the University of Toronto, and PhD at Purdue University, supervised by Dr. Stanley H. Chan.

I am interested in both the theoretical and applied sides of deep learning. My research interests primarily revolve around developing a mechanistic understanding of deep neural networks, towards the goal of improving the generalization and interpretability of these models.

In particular, my recent efforts have been focused on studying how large language models (LLMs) reason, and disentangle and compose concepts. We discover intriguingly human-interpretable circuits in LLMs for solving simple propositional logic problems, and concept-composition mechanisms in LLMs for solving multi-hop in-context learning puzzles. The modularity of reasoning processes, and the ability to disentangle and compose concepts in LLMs point to the possibilities of monitoring and steering the LLMs’ reasoning actions in the future.

I have also developed mathematical theories which connect generalization performance of DNNs with their feature-learning process, in the settings of transfer learning, and feature-based distillation.

News

(Jan. 2026) Starting Postdoctoral Researcher position at the University of Oxford, hosted by Dr. Philip Torr.

(June 2024) Started student researcher position at Google Research with Dr. Rina Panigrahy.

(Oct. 2022) Started student researcher position at Google Research with Dr. Yin Cui and Dr. Enming Luo.