Download CV

Education

• Ph.D. in Computer Science, University of Illinois at Urbana-Champaign, Urbana, IL, USA, 2024-2028 (expected).
• B.Eng. in Computer Science and Technology, Tsinghua University, Beijing, China, 2020-2024

Academic

• Undergrad GPA: Undergrad 3.90 / 4.00.
• Undergrad Selected Courses of A & A+: Linear Algebra, Discrete Mathematics, Foundation of Object-Oriented Programming, Software Engineering, Computer Architecture, Introduction to Artificial Intelligence, Artificial Neural Networks, Writing and Communication.
• Ph.D.: A member of Blender Lab, advised by Prof. Heng Ji.
• B.Eng.: A member of THUNLP, advised by Prof. Zhiyuan Liu.

Research Interests

• Tool learning and tool creation.
• LLM-driven AI agent, embodied AI.
• Agent for Science.

Publications

(*indicates equal contribution)

• Cheng Qian, Xinran Zhao, Sherry Tongshuang Wu. “Merge Conflicts!” Exploring the Impacts of External Distractors to Parametric Knowledge Graphs. ArXiv 2023.9. (Paper / Code)

• Cheng Qian, Chi Han, Yi R. Fung, Yujia Qin, Zhiyuan Liu, Heng Ji. CREATOR: Disentangling Abstract and Concrete Reasonings of Large Language Models through Tool Creation. EMNLP 2023 Findings. (Paper / Code)

• Cheng Qian, Chenyan Xiong, Zhenghao Liu, Zhiyuan Liu. Toolink: Linking Toolkit Creation and Using through Chain-of-Solving on Open-Source Model. ArXiv 2023.10. (Paper / Code)

• Yujia Qin, Shengding Hu, Yankai Lin, Weize Chen, Ning Ding, Ganqu Cui, Zheni Zeng, Yufei Huang, Chaojun Xiao, Chi Han, Yi Ren Fung, Yusheng Su, Huadong Wang, Cheng Qian, Runchu Tian, Kunlun Zhu, Shihao Liang, Xingyu Shen, Bokai Xu, Zhen Zhang, Yining Ye, Bowen Li, Ziwei Tang, Jing Yi, Yuzhang Zhu, Zhenning Dai, Lan Yan, Xin Cong, Yaxi Lu, Weilin Zhao, Yuxiang Huang, Junxi Yan, Xu Han, Xian Sun, Dahai Li, Jason Phang, Cheng Yang, Tongshuang Wu, Heng Ji, Zhiyuan Liu, Maosong Sun. Tool Learning with Foundation Models. Nature Machine Intelligence, under review.(Paper / Code)

• Cheng Qian*, Yujia Qin*, Yankai Lin, Xu Han, Zhiyuan Liu, Maosong Sun and Jie Zhou. Recyclable Tuning for Continual Pre-training. ACL 2023 findings. (Paper / Code)

• Cheng Qian*, Yujia Qin*, Jing Yi*, Weize Chen, Yankai Lin, Xu Han, Zhiyuan Liu, Maosong Sun and Jie Zhou. Exploring Mode Connectivity for Pre-trained Language Models. EMNLP 2022. (Paper / Code)

Research Experience

• Jul 2023 - Sep 2023: Exploring the Impacts of External and Internal Knowledge Conflicts Jul 2023 – Sep 2023
  • Directed by Prof. Sherry Tongshuang Wu, CMU HCII & LTI.
  • Investigated systematically how the Large Language Models (LLMs) respond to knowledge conflicts; Proposed parametric knowledge graph to model LLM’s internal knowledge and distractors to introduce external knowledge;
  • Discovered LLMs generally experience confidence boost when encountering external knowledge; Revealed LLMs’ tendency to deviate from parametric knowledge particularly when presented direct conflicts or confounding changes of information within detailed contexts.
  • First author. Submitted to ACL 2024, under review.
• Mar 2023 - Jun 2023: CREATOR: Tool Creation for Disentangling Abstract and Concrete Reasonings of Large Language Models
  • Directed by Professor Heng Ji, UIUC Blender Lab, and Associate Professor Zhiyuan Liu, THUNLP.
  • Investigated into the Large Language Model’s ability to create useful tools to solve the queries, and devised the CREATOR framework, which leverages the model’s tool creation ability through four stages: creation, decision, execution and rectification.
  • Disentangled the model’s abstract and concrete reasoning abilities and raised the models performance under MATH and TabMWP benchmarks. Released the Creation Challenge dataset which aims to test the model’s tool creation ability. Proved the effectiveness and generalization ability of our method.
  • First author. Paper accepted to EMNLP 2023 Findings.
• Mar 2023 - Jun 2023: Toolink: Linking Toolkit Creation and Using through Chain-of-Solving on Open-Source Model
  • Directed by Associate Professor Chenyan Xiong, CMU LTI, Associate Professor Zhenghao Liu, Northeastern University, and Associate Professor Zhiyuan Liu, THUNLP.
  • Investigated how the LLM’s tool using ability can be transferred to smaller, open-sourced models to help solve various tasks and raise performance.
  • First Author. Preprint publicized on ArXiv.
• Jan 2023 - Apr 2023: Tool Learning with Foundational Models
  • Explored the Large Language Model’s ability to utilize external tools in various scenarios, and formulated a general tool learning framework, in which the foundational model understands human instructions, adjusts its tool-using plan through reasoning, and effectively conquer the target tasks.
  • Contributed to part of the survey paper writing, conducted experiments and case studies under various scenarios including online shopping, cooking assistant, weather inquiry, and search engine.
  • Co-author. Nature Machine Intelligence under review.
• Aug 2022 – Jan 2023: Recyclable Tuning for Continual Pretraining
  • Directed by Associate Professor Zhiyuan Liu, THUNLP.
  • Formulated the task of compatible tuning as PLM continually acquire fresh knowledge from emerging data, and explored how to make earlier adapted weights compatible with subsequent upgraded PLMs.
  • Explored the parametric connections among continually pre-trained models; Proposed CLoP, which enables compatible tuning in a data-efficient and training efficient way; Experimented on various NLP tasks and demonstrated the superiority of CLoP; Construct the first benchmark regarding to the field of compatible tuning.
  • Co-first author. Accepted by ACL 2022 findings.
  • Project selected to THU Undergraduate Academic Advancement program and won ￥30K support.
• Mar 2022 – Jul 2022: Exploring Mode Connectivity for Pre-trained Language Models
  • Directed by Associate Professor Zhiyuan Liu, THUNLP.
  • Analysed the geometric connections of different minima in loss landscape through the lens of mode connectivity, which measures whether two minima can be connected with a low loss path.
  • Explored how various hyperparameters and training data affect PLMs’ mode connectivity; Discovered the role of pre-training in facilitating mode connectivity and pulling task boundaries closer; Investigated into how PLMs task knowledge change along the connected path quantitatively.
  • Co-first author. Accepted by EMNLP 2022 main conference.
  • Project established in THU Student Research Training Program.
• Mar 2022 – Jun 2022: THUPat: A Convenient Campus Helper
  • Directed by Associate Researcher Chun Yu, Theory and Practice of Human Computer Interaction course project.
  • Proposed “pat” for the first time as the medium in human-phone interaction. Built an open source android software THUPat that can help with various kinds of campus events via simply patting the phone.
  • Collaborator. Software released in THU and benefited the campus community.
• Jul 2022 – Aug 2022: Quantum Automata: Capability and Efficiency
  • Directed by Professor Zhengfeng Ji and Professor Mingsheng Ying, Topics in Quantum Computing course project.
  • Defined the efficiency of quantum automata from 3 different perspectives with respect to acceptance probability, space and time; Proposed an algorithm that can effectively optimize quantum automata’s acceptance probability, applying the knowledge from neural network.
  • First author. Course thesis won high recognition.

Awards

• December-9th Scholarship, highest scholarship in Dept. of CST, 1/180. (2021)
• Volunteering & Social Survey Excellence Scholarship, Dept. of CST, 1/180. (2022)
• Awards of Excellent Student Cadre, Tsinghua University. (2021)
• Second Prize in National Undergraduate Physics Competition, Beijing Physics Society. (2021)
• Third Prize in THU Challenge Cup Academic Competition, Tsinghua University. (2022)

Languages

Mandarin(Native), English(Fluent)

• TOEFL 115/120 (Reading 30, Listening 30, Speaking 26, Writing 29).
• GRE Verbal Reasoning 162/170, Quantitative Reasoning 170/170, Analytical Writing 4/6.

Skills & Expertise

• Highly self-motivated researcher.
• Strong interpersonal skills with a good sense of teamwork.
• Programming Skills: Python, C/C++.
• Rich experience in state-of-the-art deep learning techniques.

Service & Leadership

• Reviewer of COLING 2022.
• Olympic Family Assistant of IBSF President Ivo Ferriano and HRH Prince Jigyel Ugyen Wangchuck (Bhutan), 2022 Beijing Winter Olympics.
• Member of Student Union and Student Association of Science and Technology, Tsinghua University. 2021-2023.
• Member of Tsinghua Orienteering Team (Sport, Professional), Tsinghua University. 2020-2023.
• Student Mentor of Tsinghua Summer School (Beijing), Tsinghua University. 2022.
• Leader of Summer Education-Aiding Program in NingXia, China. The program won Gold Prize in Tsinghua. 2021.