Hongyu Zhao

About Me

I am Hongyu Zhao (赵鸿宇), a Computer Science Ph.D student at University of Maryland. My advisor is Tianyi Zhou. I got my M.S. degree in Computational and Applied Mathematics at University of Chicago, supervised by Hongyuan Mei.
Before joining UChicago, I received B.S. in Mathematics and Applied Mathematics from School of the Gifted Young, University of Science and Technology of China.
My research goal is to develop an NLP solution to Explainable AI. Ultimately I want to develop a systematic approach to make any large pretrained model explain its bahaviour in natural language, with human-understandable concepts and reasoning. I believe that to reach this goal, some prior works in parameter-efficient transfer learning, logical inference and few-shot/semi-supervised learning need to be done.

Contact

E-mail: hongyuz@umd.edu

Google Scholar

Publications

Explicit Planning Helps Language Models in Logical Reasoning
EMNLP 2023 Oral
Hongyu Zhao, Kangrui Wang, Mo Yu and Hongyuan Mei
Arxiv
Robustness of Learning from Task Instructions
Findings of ACL 2023
Jiasheng Gu, Hongyu Zhao, Hanzi Xu, Liangyu Nie, Hongyuan Mei, Wenpeng Yin
Arxiv
Tiny-Attention Adapter: Contexts Are More Important Than the Number of Parameters
EMNLP 2022
Hongyu Zhao, Hao Tan and Hongyuan Mei
Arxiv

Besides Research

I'm a very good go player. I've been playing this game for more than a decade when DeepMind's AlphaGo gave the community a real shock in 2016. The event not only aroused my interest in AI, but also played a role in shaping my research goal: Human go masters have published so many books talking about their own understanding of the game, sometimes with their self-invented concepts. If a machine can play much better, why can't it develop its own theory and explain it to us?

I play all kinds of card games, including traditional games like contract bridge and Texas hold'em, popular TCGs like MTG and YuGiOh, and modern cyber games like DQ and STS. I could feel sanity in the unification of these different axiomatic systems, each is simple yet bears so many possibilities.

I enjoyed jogging at late nights when I was in China. I'm now trying to develop some new habits here in America -- I really don't like gyms that much.