I am Xunjian Yin, a second-year Master’s student at the Wangxuan Institute of Computer Technology at Peking University. My advisor is Prof. Xiaojun Wan. Previously, I obtained my B.S. degree in Computer Science from Peking University. I worked as an intern at Microsoft. Currently, I am a visiting student at UCSB NLP Group, advised by Professor William Wang.

My research focuses on advancing Large Language Models (LLMs), particularly in integrating new information. This includes benchmarking the knowledge boundaries of LLMs to define the extent of their existing expertise (PGDC), and evaluating how LLMs generalize and perform when encountering new knowledge (Alcuna). Additionally, I develop knowledge editing techniques to seamlessly integrate new knowledge while preserving valuable existing information (AToKe). Furthermore, I design unsupervised methods for self-alignment, enabling LLMs to resolve internal contradictions and improve their performance based solely on their current knowledge (ContraSolver). Through these efforts, I strive to push the frontiers of LLM capabilities, ensuring they remain robust and adaptable in an ever-evolving information landscape.

I’m seeking 25 Fall Ph.D. opportunities! Feel free to reach out to me if you’re interested in my research!

Recent News

• 2024-05: One paper is accepted to ACL 2024
• 2023-12: One paper is accepted to AAAI 2024
• 2023-10: Two papers are accepted to EMNLP 2023

Preprints

• ContraSolver: Self-Alignment of Language Models by Resolving Internal Preference Contradictions
  Xu Zhang*, Xunjian Yin*, Xiaojun Wan
  arXiv:2406.08842 [paper]

• Themis: Towards Flexible and Interpretable NLG Evaluation
  Xinyu Hu, Li Lin, Mingqi Gao, Xunjian Yin, Xiaojun Wan
  arXiv:2406.18365 [paper]

• Human-like Summarization Evaluation with ChatGPT
  Mingqi Gao, Jie Ruan, Renliang Sun, Xunjian Yin, Shiping Yang, Xiaojun Wan
  arXiv:2304.02554 [paper]

• Error-Robust Retrieval for Chinese Spelling Check
  Xunjian Yin, Xinyu Hu, Xiaojun Wan
  arXiv:2211.07843 [paper]

Selected Publications

• Benchmarking Knowledge Boundary for Large Language Model: A Different Perspective on Model Evaluation
  Xunjian Yin*, Xu Zhang*, Jie Ruan, Xiaojun Wan
  ACL 2024 [paper] [code]

• History Matters: Temporal Knowledge Editing in Large Language Model
  Xunjian Yin, Jin Jiang, Liming Yang, Xiaojun Wan
  AAAI 2024 [pdf] [code]

• ALCUNA: Large Language Models Meet New Knowledge
  Xunjian Yin*, Baizhou Huang*, Xiaojun Wan
  EMNLP 2023 [pdf] [code]

• Exploring Context-Aware Evaluation Metrics for Machine Translation
  Xinyu Hu, Xunjian Yin, Xiaojun Wan
  EMNLP 2023 findings [pdf] [code]

• Overview of the NLPCC 2023 Shared Task: Chinese Spelling Check
  Xunjian Yin, Xiaojun Wan, Dan Zhang, Linlin Yu, Long Yu
  NLPCC 2023 [pdf] [code]

• How Do Seq2Seq Models Perform on End-to-End Data-to-Text Generation?
  Xunjian Yin, Xiaojun Wan
  ACL 2022 [pdf] [code]

Academic Services

• Reviewer: ACL 2023, ACL ARR 2023, ACL ARR 2024, NeurIPS 2024
• NLPCC Shared Task 8 track chair
• Volunteer: AAAI 2024, ACL 2024