- Machine learning theory under the framework of non-convex landscape analysis, aiming to understand under what conditions global or generalizable solutions can be found and their guarantees.
- Efficient machine learning practices, guided by theories developed in the first part, focusing on three aspects: data efficiency (using less data for similar performance), algorithm efficiency (using zeroth order optimization to reduce memory and compute demands for training and fine-tuning), and model efficiency (better quantization strategies for compressing model sizes).
Education
Ph.D. from the EECS department of UC Berkeley, supervised by Somayeh Sojoudi; previously studied Engineering Science at the University of Toronto and conducted research at STARS lab.
Background
Currently an assistant professor in the CS department at City University of Hong Kong. Received the honorary title of presidential assistant professor at CityU. His research aims to develop explainable and efficient machine learning systems, emphasizing the exploration of new theoretical tools and perspectives to demystify various phenomena in modern machine learning.
Miscellany
Enjoys photography, watching anime, and experiencing a city by wandering about (city walk).