I am an assistant professor in the Department of Computer Science at The University of Hong Kong and The HKU Musketeers Foundation Institute of Data Science. I am generally interested in machine learning, stochastic optimization, and graph learning, with a special focus on the theoretical/empirical understanding of the deep learning algorithms. I am also open to any challenging research topic in artifical intelligence and data science.
Previously, I obtained my Ph.D. in the Computer Science department at the University of California, Los Angeles (UCLA), supervised by Prof. Quanquan Gu. I obtained my master degree in electrical engineering and bachlor degree in applied physics, both in Unversity of Science and Technology of China (USTC).
Multiple openings for PhD, Postdoc, interns, and RA. Please drop me an email with your CV and transcript (optional) if you are interesed in joining my research group.
[2024-01] A new manuscript on faster non-log-concave sampling is online, where we show that non-exponential convergence can be achieved for sampling non-log-concave distributions (or distribution without isoperimetry) via diffusion-based Monte Carlo. Any comments are welcomed!
[2023-10] A new manuscript on the Oscillation Effect of SGD is online, where we theoretically investigate the benefit of the oscillation of SGD with large learning rate for genralization. Any comments are welcomed!
[2023-10] A new manuscript on Learning XOR data via ReLU networks is online, where we provide a precise analysis on the generalization performance of ReLU CNN trained by GD on (non-orthogonal) XOR-type data. Any comments are welcomed!
[2023-06] We are looking for PostDocs in Mathematical Data Science. Candidates with a strong background in mathematics, proven publication record and interest in theoretical aspects of machine learning are encouraged to apply. Strong candidates can be referred to HKU IDS PostDoc Research Fellowship (up to HKD 514800/year plus additional research funding support). Please find the details in the advertisement.
[2023-05] Our paper on the implicit bias of batch normalization is accepted to COLT 2023, manustripts will be released soon!
[2023-03] I will serve as the Area Chair in NeurIPS 2023.
[2023-01] Our paper on the generalization separation between Adam and GD has been accepted by ICLR 2023.
[2022-09] Two papers accepted by NeurIPS 2022. The first paper studies the generalization of multi-pass SGD for over-parameterized least squres; the second paper demonstrates the power and limitation of pretraining-finetunning for linear regression with distribution shift.
[2022-08] Dr Difan Zou just joined HKU CS as an assistant professor.