Hi, I am an senior undergraduate student at Renmin University of China, majoring Artificial Intelligence in Gaoling School of Artificial Intelligence. Currently, I am visiting in University of Illinois Urbana Champaign advised by Professor Heng Ji, working on Knowledge Editing. Last summer, I was fortunate to do my summer internship under Professor Jiantao Jiao and Banghua Zhu at University of California, Berkeley. My research interest mainly focus on natural language processing. Currently I am interested in the following topics:

1.Language Models’ Explainability and Interpretability: Recent advancements in NLP have been significantly driven by pre-trained language models. Yet, the understanding of their internal workings remains limited, posing risks in practical applications. Currently, I am particularly interested in how Large Language Models (LLMs) encode knowledge, as they can be conceived as repositories of knowledge. Having LLMs retrieve in their own parametric implicitly and explicitly can have totally different behaviors. Furthermore, I want to explore more methods for LLMs to organize knowledge like disentangling information storage.
2.Language Models’ Reasoning and Planning Ability: In my previous PlanBench project, I discovered that LLMs fine-tuned with Reinforcement Learning (RL) exhibited a superior generalized planning capabilities even in complex tasks where traditional algorithms like Proximal Policy Optimization (PPO) faltered. This experience has led me to believe that advanced LLMs could be ideal for planning in real-world scenarios. I also theorize that effective planning is closely linked to commonsense reasoning. Moving forward, my goal is to augment the capabilities of LLMs by delving into their underlying reasoning and planning mechanisms.
3.Information Retrieval Algorithms and Retrieval-Augmented Language Models (RaLMs): Information retrieval (IR) plays a crucial role in creating transparent and dependable AI systems. Presently, LLMs grapple with issues of data fabrication or 'hallucination.' Employing retrieval methods enables LLMs to source evidence for their outputs, thereby enhancing their reliability.

Publication

2023

• H-InDex: Visual Reinforcement Learning with Hand-Informed Representations for Dexterous Manipulation.

  Yanjie Ze, Yuyao Liu†, Ruizhe Shi†, Jiaxin Qin, Zhecheng Yuan, Jiashun Wang, Huazhe Xu

  NeurIPS 2023 [paper] [bib]

  
  

  @article{ze2023h,
    title={H-InDex: Visual Reinforcement Learning with Hand-Informed Representations for Dexterous Manipulation},
    author={Ze, Yanjie and Liu, Yuyao and Shi, Ruizhe and Qin, Jiaxin and Yuan, Zhecheng and Wang, Jiashun and Xu, Huazhe},
    journal={arXiv preprint arXiv:2310.01404},
    year={2023}
  }
                      

• Towards Effective Ancient Chinese Translation: Dataset, Model, and Evaluation.

  Geyang Guo, Jiarong Yang, Fengyuan Lu, Jiaxin Qin, Tianyi Tang, Wayne Xin Zhao

  NLPCC 2023 [paper] [bib]

  
  

  @inproceedings{guo2023towards,
    title={Towards Effective Ancient Chinese Translation: Dataset, Model, and Evaluation},
    author={Guo, Geyang and Yang, Jiarong and Lu, Fengyuan and Qin, Jiaxin and Tang, Tianyi and Zhao, Wayne Xin},
    booktitle={CCF International Conference on Natural Language Processing and Chinese Computing},
    pages={416--427},
    year={2023},
    organization={Springer}
  }