Welcome to Fanghui Liu's Homepage! [中文简介]
I'm currently a tenure-track associate professor at Institute of Natural Sciences and School of Mathematical Sciences, Shanghai Jiao Tong University (SJTU). I serve as a PhD supervisor in School of Mathematical Sciences and School of Artificial Intelligence, Shanghai Jiao Tong University. I have also held a Global Visiting Professorship at the Technical University of Munich, Germany. Previously, I was an assistant professor at University of Warwick, UK (FAM Division and affiliated at DIMAP). Before my faculty position, I was a postdoc researcher at KU Leuven and EPFL from 2019 to 2023. I received my Bachelor’s degree in Automation from Harbin Institute of Technology in 2014 and my PhD degree from Shanghai Jiao Tong University in 2019.
Email: x@y with x=fanghui.liu, y=outlook.com or sjtu.edu.cn
Research Interests
I'm generally interested in foundations of modern machine learning from the lens of learning efficiency, both theoretically and empirically. My research is always contributing to how to handle nonlinearity at a theoretical level and how to precisely and efficiently approximate nonlinearity at a practical level under theoretical guidelines, which is a longstanding question over science, technology and engineering.
My research (the past, ongoing and future) focuses on the following directions:
machine learning theory: what is the largest function space that can be learned by neural networks, both statistically and computationally efficiently (computational-statistical gaps)?
LLM pre-training: the principles of scaling laws, training dynamics
post-training in LLMs: expand the frontiers of empirical and theoretical knowledge on when and where to fine-tuneinference, and how much we can fine-tuneinference, precisely, efficiently, and robustly
AI4MLT: AI for machine learning theory, Lean auto-formalisation