About Me

Hi! I am Ansong Ni (倪安松), a final-year Ph.D. student in the Computer Science Department at Yale University working with Prof. Arman Cohan, and (the late) Prof. Dragomir Radev. Prior to Yale, I obtained my M.S. in CS degree from School of Computer Science at Carnegie Mellon University and B.Eng. from Nanjing University in China.

I worked as a research intern at Google DeepMind (Summer 2023), Meta AI (Summer 2022), MSR Redmond (Summer 2021), AI2 (Summer 2020), MSR Asia (Summer, Fall 2017).

I will be on the job market starting fall 2023 for industrial research labs & startups.

Research Interest

My current research focus is code generation using large language models (LLMs). More specifically, I am intrigued by the abilities / applications enabled by LLMs trained on code, such as:

  • Procedure understanding and reasoning (ICLR’23)
  • AI-assisted software development (ICSE’21, MSR’17)
  • Tool use and robot control

I am also excited by the various learning problems posted by code LLMs, such as:

  • Learning from execution feedback (ICLR’23, ICML’23)
  • Neuro-symbolic learning (ICLR’23)
  • (inter-)active learning (AAAI’20), Transfer Learning (DL4C’23, ICML’23), Multi-task Learning (EMNLP’22)*

Previously, I have also done research in more traditional NLP tasks, such as question answering (EMNLP’21) and text summarization (EMNLP’21, ACL’22, EMNLP’22). I also led the development of SummerTime (230+ stars) at Yale.

For more details about my research, please refer to my publication list below.

Selected Publications

For a full list, please refer to my Google Scholar or Semantic Scholar.
(* denotes equal contribution)

  • Ansong Ni, Srini Iyer, Dragomir Radev, Ves Stoyanov, Wen-tau Yih, Sida I. Wang, and Xi Victoria Lin
    LEVER: Learning to Verify Language-to-Code Generation with Execution
    The 2023 International Conference on Machine Learning (ICML’23)
    [arxiv] [code]

  • Ansong Ni, Jeevana Priya Inala, Chenglong Wang, Oleksandr Polozov, Christopher Meek, Dragomir Radev, and Jianfeng Gao
    Learning Math Reasoning from Self-Sampled Correct and Partially-Correct Solutions
    The 2023 International Conference on Learning Representations (ICLR’23)
    [arxiv] [code]

  • Zhangir Azerbayev, Ansong Ni, Hailey Schoelkopf, and Dragomir Radev
    Explicit Knowledge Transfer for Weakly-Supervised Code Generation
    Deep Learning For Code (DL4C) Workshop @ ICLR’23
    [arxiv]

  • Tianbao Xie*, Chen Henry Wu*,…, Ansong Ni,…, Rui Zhang, Noah A. Smith, Luke Zettlemoyer, and Tao Yu
    UnifiedSKG: Unifying and Multi-Tasking Structured Knowledge Grounding with Text-to-Text Language Models
    The 2022 Conference on Empirical Methods in Natural Language Processing (EMNLP’22)
    [arxiv] [website] [code]

  • Ansong Ni, Matt Gardner, and Pradeep Dasigi
    Mitigating False-Negative Contexts in Multi-document Question Answering with Retrieval Marginalization
    The 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP’21)
    [arxiv] [code]

  • Ansong Ni, Zhangir Azerbayev, Mutethia Mutuma, Troy Feng, Yusen Zhang, Tao Yu, Ahmed Hassan Awadallah, and Dragomir Radev*
    SummerTime: Text Summarization Toolkit for Non-experts
    The 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP’21) Demo Track [arxiv] [code] (230+ stars)

  • Ansong Ni*, Daniel Ramos*, Aidan Yang, Ines Lynce, Vasco Manquinho, Ruben Martins, and Claire Le Goues
    SOAR: A Synthesis Approach for Data Science API Refactoring
    The 43th International Conference on Software Engineering (ICSE’21)
    [arxiv] [code] [talk]

  • Ansong Ni, Pengcheng Yin, and Graham Neubig
    Merging Weak and Active Supervision for Semantic Parsing
    The 34th AAAI Conference on Artificial Intelligence (AAAI’20)
    [arxiv] [code]

Talks and Presentations

  • Enhancing Language Models for Program Synthesis using Execution, Apr 2023, Invited Talk @ UT Austin TAUR Lab
  • Enhancing Language Models for Program Synthesis using Execution, Mar 2023, Invited Talk @ HKUST CSE [recordings]
  • Enhancing Language Models for Program Synthesis using Execution, Mar 2023, Invited Talk @ MIT CSAIL
    [slides] [recordings]
  • Learning from Self-Sampled Correct and Partially-Correct Programs, June 2022, Paper Presentation @ Meta AI Reading Group
  • Merging Weak and Active Supervision for Semantic Parsing, Feb 2020, Oral Paper Presentation @ AAAI Conference

Professional Services

  • Program Committee/Reviewer
    • ICLR 2024
    • ICML 2023
    • ACL 2023
    • NeurIPS 2022, 2023
    • EMNLP 2022
    • ACL Rolling Reviews (ARR) 2021-2022
    • DL4C Workshop @ ICLR 2023
    • SUKI Workshop @ NAACL 2022
    • IntEx-SemPar Workshop @ EMNLP 2020

Work Experience

Education

Miscellaneous

  • My first name can be pronounce as-is: [An-Song], last name is like [Nee].
  • A tribute to my late advisor Drago.
  • I know how to say “I don’t speak [this language]” in 9 different languages (Mandarin, English, Spanish, Cantonese, Italian, Greek, Korean, Hindi, Hebrew). Very handy, maybe you should know them too.
  • I love soccer and I am a Barcelona and Messi fan. I don’t remember how Barca did in UCL 2018-2022 and don’t remind me. My favorite games are:
    • 2015 UCL final: Barca 3-1 Juventus
    • 2017 UCL round of 16: Barca 6-1 PSG (6-5 on agg.)
    • 2022 WC final: Argentina 3-3 France (4-2 in pen.)
  • I am from Nanchang, Jiangxi Province in China. I’ve also lived in Nanjing, Beijing, Berkeley, Pittsburgh, Houston, New Haven, and Menlo Park.