About me
Hi! I am a second-year PhD student at the University of Pennsylvania, advised by Surbhi Goel (Computer Science) and Enric Boix-Adsera (Wharton Statistics & Data Science). I received my Bachelor’s degree in Computer Science & Mathematics from HKUST in 2025. My research is supported by the AWS Asset Fellowship at Penn.
My research interests lie in the theory and empirical science of deep learning and LLMs: running experiments that probe how these models work, then building theory to explain what those experiments reveal. Recently, I have been focusing on the following questions:
- Scaling Science. How can we pre-train, post-train and scale models in a principled way, and how can we understand their training dynamics, both in theory and in practice?
- Alignment Science. How can we keep a model aligned as its capabilities improve? I am interested in misalignment phenomena that remain poorly understood, and in safety monitoring.
Education
- PhD in Computer & Information Science, University of Pennsylvania, 2025–present
- BSc in Computer Science & Mathematics, The Hong Kong University of Science and Technology, 2025
- Exchange semester, EPFL, Spring 2024
Review Experience
- Conference / Journal: NeurIPS (2024, 2025, 2026), ICML 2025, L4DC 2025, TMLR
- Workshops: ICML 2026 Mech Interp Workshop, ICML 2026 CoLoRAI Workshop, ICLR 2026 Workshop on Scientific Methods for Understanding Deep Learning (Sci4DL), ICLR 2024 Workshop on Bridging the Gap Between Practice and Theory in Deep Learning (BGPT), ICML 2024 Workshop on Theoretical Foundations of Foundation Models (TF2M)
Selected Awards
- AWS Asset Fellowship, 2026
- HKUST Epsilon Fund Award, 2024 (for top math students at HKUST, fewer than 5 undergraduates each year)
- Hong Kong Government Scholarship, 2022–2025 (the highest undergraduate academic award)
Other Academic Activities
- Princeton ML Theory Summer School, August 2026
- Workshop on Theoretical Perspectives on LLMs, UCSD, San Diego, March 2025
- Heidelberg Laureate Forum, Heidelberg, Germany, September 2024