She Yifei is an undergraduate student in Communication Engineering at Beijing University of Posts and Telecommunications (BUPT), expecting to graduate in 2026. He will continue at BUPT to pursue his Ph.D. under the supervision of Prof. Kai Niu. His doctoral research will focus on the interpretability of Large Language Models (LLMs) through the lens of semantic information theory.
His research interest includes semantic information theory, mechanistic interpretability and representation learning.
🔥 News
- 2025.11: I was granted a patent (CN121213699A) for synthesizing training images.
- 2025.09: I was invited to be the reviewer of International Journal of Information Technology & Decision Making (IJITDM).
- 2025.09: I built a pinned chat widget that calls AI APIs for quick questions.
- 2025.09: 🎉🎉 I created my personal academic homepage.
📝 Publications
Semantic Algorithmic Information Theory: From Kolmogorov Complexity to Semantic Equivalence
Jiatong Wu, Sen Wang, Kai Niu, Yifei She, Ping Zhang
- We introduce Semantic Algorithmic Information Theory, formalize the Semantic Turing Machine System, define Semantic Complexity, and propose a model-based estimator of Normalized Semantic Information Distance to measure semantic equivalence beyond syntactic variation.
Learning to Wait: Synchronizing Agents with the Physical World
Yifei She, Ping Zhang, He Liu, Yanmin Jia, Yang Jing, Zijun Liu, Peng Sun, Xiangbin Li, Xiaohe Hu
- We leverage the semantic priors of LLMs to endow them with the ability to actively wait for event completion, thereby reducing the manual effort required for system-side development.
DisLoRA: Task-specific Low-Rank Adaptation via Orthogonal Basis from Singular Value Decomposition
She Yifei, Xinhao Wei, Yulong Wang
- We decompose LoRA’s A and B matrices via SVD to identify task-specific directions for optimized training and use an adaptive soft regularization loss to accelerate convergence.
Enhancing Vehicle Platooning Safety via Control Node Placement and Sizing under State and Input Bounds
Yifei She, Shen Wang, Ahmad Taha, and Xiaofeng Tao
- We utilize a greedy algorithm to select key control nodes and apply convex optimization to constrain over-approximated reachable sets, ensuring vehicle platooning safety.
📖 Educations
- 2022.09 - 2026.06 (Expected), Bachelor of Engineering in Communication Engineering, Beijing University of Posts and Telecommunications (BUPT).
- 2026.09 - Future, Ph.D. Candidate in Artificial Intelligence, Beijing University of Posts and Telecommunications (BUPT). To be supervised by Prof. Kai Niu on the interpretability of LLMs via semantic information theory.
💻 Internships
- 2025.10 - Current, Research Intern at Infrawaves, working on production-grade AIOps and MaaS research.
🎖 Awards and Experiences
- 2026.06, College-level Outstanding Undergraduate Thesis, BUPT.
- 2025.10, Third-Prize University Scholarship, BUPT.
- 2025.08, Merged a Pull Request to PaddleNLP.
- 2025.08, Outstanding Student, Financial Services Project, National University of Singapore (NUS) Summer Workshop.
- 2025.03, Selected for BUPT’s PING Program for advanced studies in Semantic Information Theory.
- 2024.11, Recipient of the BUPT Future Scholars Research Fund.
- 2023.10, Third-Prize University Scholarship, BUPT.