 |
Hongyang Ryan Zhang Address: 177 Huntington Ave 2211, Boston, MA 02115 Email: ho.zhang@northeastern.edu |
I am an assistant professor at Northeastern University in the Khoury College of Computer Sciences, working on machine learning and computer science, focusing on developing statistical foundations and designing optimization algorithms, with applications in networks, data mining, and language modeling. Here are several topics I have recently been working on: 1) Adaptation and transfer of neural networks, including multitask learning, fine-tuning, and in-context learning. I am particularly interested in understanding how neural networks learn to extract information from data, and how this knowledge transfers to downstream tasks. 2) Learning from graph-structured data and geometric data, such as their sampling complexities. 3) Nonconvex optimization including low-rank matrix factorization and matrix completion. 4) Evaluation of AI and LLMs such as their mathematical and algorithmic reasoning capabilities.
I received my Ph.D. in computer science from Stanford University and my B.Eng. in computer science from Shanghai Jiao Tong University. Subsequently, I spent a year as a postdoc within the statistics and data science department at the University of Pennsylvania.
I enjoy working on technically challenging problems, while striving for broader impacts by creating new knowledge that will benefit the society, as well as fostering the next generation of engineers and researchers. I support accessible and reproducible research. If you are a student at Northeastern interested in my research, please feel free to email me.
Recent updates
Talk slides about our work on a Hessian view of learning.
New paper on a linear-time data selection algorithm for in-context learning.
An updated paper on transfer learning random matrices accepted at JMLR! We analyze a (classical) hard parameter sharing estimator and find that simply rebalancing the data leads to minimax optimal rates.
New manucript on an ensemble method for fine-tuning LLMs, built on top of LoRA.
A list of older logs.