Portrait of Nan Duan

Nan Duan

Principal Research Manager

About

Dr. Nan DUAN (段楠) is a Principal Research Manager at Microsoft Research Asia. He is working on fundamental NLP tasks, especially on question answering, semantic parsing, pre-trained models for language+knowledge/image/video/code, and machine reasoning.

Before joining MSRA, he received his Ph.D. on Statistical Machine Translation from Tianjin University in 2011, under supervision of Dr. Ming ZHOU and Dr. Mu LI.

We are hiring researchers and interns! If you have strong publications and experiences in above areas and are willing to work in MSRA, hit me up. 

Highlights:

  • 2020-06-05: M^3P (paper, repo) is a multilingual multimodal pre-trained model that supports 100 languages for both multilingual NLP tasks and multilingual image-language tasks at the same time.
  • 2020-05-28: XGLUE Leaderboard (link, blog, Twitter, Instagram, Facebook) is online. Have a try 🙂
  • 2020-04-03: XGLUE (paper) is a new benchmark dataset for cross-lingual pre-training, understanding and generation.
  • 2020-02-22: ProphetNet (paper, code) is a generative pre-trained model, which achieves SOTA results on Question Generation and News Summarization tasks.
  • 2020-02-19: CodeBERT (paper) is an code-language pre-trained model, which achieves SOTA results on Code Retrieval and Code Generation tasks.
  • 2020-02-15: Unicoder-VL for Videos (paper) is a video-language pre-trained model, which achieves SOTA results on Video Retrieval and Video Captioning tasks.
  • 2019-11-20: Unicoder (paper) and Unicoder-VL for Images (paper) was reviewed by Bill Gates.
  • 2019-10-26: Our tutorialMachine Reasoning for Knowledgeable, Explainable and Inferable Models” is accepted by EMNLP 2020.
  • 2019-10-14: CCF NLPCC Distinguished Young Scientist, 2019.
  • 2019-09-05: Unicoder-VL for Images (paper) is an image-language pre-trained model, which achieves SOTA results on Image-Text Retrieval and Visual QA (GQA) tasks.
  • 2019-08-31: DREAM (paper) achieves SOTA result on the CommonsenseQA leaderboard.
  • 2019-08-07: DREAM (paper) achieves SOTA result on the FEVER leaderboard.
  • 2019-07-19: Unicoder (paper) is a cross-lingual pre-trained model, which achieves SOTA results on XNLI and Cross-lingual QA tasks.
  • 2019-07-13: DREAM achieves SOTA result on the GQA leaderboard.
  • 2019-06-25: DREAM (Dynamic REAsoning Machine) achieves SOTA on the NQ leaderboard.
  • 2019-04-02: MSParS (download) is a Multi-perspective Semantic ParSing dataset for knowledge-based QA.
  • 2018-12-07: Natural Language Understanding (joint work with Bing) was reviewed by Bill Gates.
  • 2018-11-08: Video-based Conversational AI was reviewed by Bill Gates on MSRA’s 20th Anniversary.
  • 2018-09-10: 《智能问答》was published by the Higher Education Press.
  • 2017-12-31: CCF Distinguished Speaker.