Chi Han 韩迟

Chi Han is a PhD candidate at UIUC CS, advised by Prof. Heng Ji. Previously, he obtained his Bachelor's degree from YaoClass, Tsinghua University. Expected graduation in June 2026.

[Twitter]; [Google Scholar]; [GitHub]; [Curriculum Vitae];
My Research

My long-term research goal is to develop the science of language models (LMs) by asking and answering the following questions: What role does each hidden representation play in LLMs? When will they malfunction or cause counterintuitive phenomena? How can we adapt the mechanisms to transcend their inherent limits and better serve the downstream tasks? To this end, I have been working on a theoretical and empirical analysis of LLM hidden representations LMs, with the aim to provide insights and develop useful tools and cures for LMs.

Awards and Honors
News, Service and Experiences
Publications ( show selected / show all by date )
Computation Mechanism Behind LLM Position Generalization
Chi Han, Heng Ji

Arxiv Preprint

Can Language Models Follow Multiple Turns of Entangled Instructions?
Chi Han,

Homepage, Arxiv Preprint, Github Repository, HuggingFace Dataset

LM-Infinite: Zero-Shot Extreme Length Generalization for Large Language Models
Chi Han, Qifan Wang, Hao Peng, Wenhan Xiong, Hugo Chan, Heng Ji, Sinong Wang

NAACL 2024 (Outstanding Paper Award), Paper, Codes, Video

LM-Steer: Word Embeddings Are Steers for Language Models
Chi Han, Jialiang Xu, Manling Li, Yi Ren Fung, Chenkai Sun, Nan Jiang, Tarek Abdelzaher, Heng Ji

ACL 2024 (Outstanding Paper Award), Paper, Codes, HuggingFace Live Demo, Slides, Poster

Explaining Emergent In-Context Learning as Kernel Regression
Chi Han, Ziqi Wang, Han Zhao, Heng Ji

ArXiv, Paper

Logical Entity Representation in Knowledge-Graphs for Differentiable Rule Learning
Chi Han, Qizheng He, Charles Yu, Xinya Du, Tonghang Hang, Heng Ji

ICLR 2023, Paper, Codes

Learning Shared Semantic Space for Speech-to-Text Translation
Chi Han, Mingxuan Wang, Lei Li, Heng Ji

ACL 2021 Findings, Paper, Codes