RSIHub
A research framework for principled agent self-improvement under frozen evaluators and declared mutation boundaries, with verifiable lineage for reproducibility and auditing.
About Me
I am a third-year PhD student at the University of Science and Technology of China, currently pursuing research as an intern at the Beijing Institute for General Artificial Intelligence (BIGAI), under the guidance of Professor Qi Liu and Researcher Zilong Zheng. I hold a bachelor's degree from Beihang University.
Open to collaboration. My research centers on reward modeling, LLM alignment, and continual learning. Feel free to reach out if you are interested in these topics.
Research Interests
- Reward Modeling
- LLM Alignment
- Continual Learning
- Agent Memory
Research Overview
“Toward language models that keep improving from their own experience — through better reward signals, robust alignment, and continual adaptation.”
Reward Modeling & Alignment
Designing reliable reward models and preference-optimization methods that stay calibrated and aligned as the policy evolves beyond static training data (e.g., UAPO, SAVE).
Continual Learning & Memory
Enabling models to acquire, retain, and update knowledge over time through knowledge editing and ever-improving memory systems (e.g., ICE, RAM).
Publications
View Full List →* indicates equal contribution.
Preprints
Latest News
View All →Projects
Belief Context Graph
A confidence-aware belief context graph that drives more deterministic reasoning in long-horizon agents.
Education
Ph.D. Student
2024.09 - PresentUniversity of Science and Technology of China
B.Eng.
2020.09 - 2024.06Beihang University
Experience
Research Intern
2025.07 - PresentBeijing Institute for General Artificial Intelligence (BIGAI), Beijing, China
Algorithm Intern
2024.01 - 2024.08Beijing Institute for General Artificial Intelligence (BIGAI), Beijing, China
Commercial Product Intern
2023.10 - 2024.01ByteDance · Commercialization Product & Technology, Beijing, China
Academic Service
Conference Reviewer
Program Committee Member
ACL, EMNLP, ICLR, NeurIPS
Contact
Location
Beijing, China