Tao Qin

Timeline

1. 2003–2008

   Degrees in electronic engineering

   He earns a bachelor's degree in electronic engineering at Tsinghua University in 2003 and a doctorate from the same department in 2008.

2. 2007–2010

   Learning to rank benchmarks

   In 2007 he coauthors the ICML paper Learning to rank: from pairwise approach to listwise approach. In April 2008 he, Tie-Yan Liu, and Hang Li publish a WWW paper on learning to rank relational objects and its application to web search. In 2010 he, Tie-Yan Liu, Jun Xu, and Hang Li publish LETOR, a benchmark collection for learning to rank in information retrieval.

3. 2008–2022

   Years at Microsoft Research Asia

   From 2008 to 2022 he is a senior principal researcher and manager at Microsoft Research Asia. In March 2018 he speaks at the Hong Kong University of Science and Technology as a senior research manager at Microsoft Research Asia, and he is an adjunct professor and PhD advisor at the University of Science and Technology of China.

4. December 2016

   Dual learning for translation

   In December 2016 he coauthors Dual Learning for Machine Translation with Di He, Yingce Xia, Liwei Wang, Nenghai Yu, Tieyan Liu, and Wei-Ying Ma. The paper develops a dual-learning mechanism that lets a neural machine translation system learn from unlabeled data.

5. 2018–2020

   Translation, speech, and mahjong

   The translation work reaches human parity in Chinese-English translation in 2018 and takes first place in eight tasks at WMT in 2019. FastSpeech uses feed-forward networks to generate mel-spectrograms 270 times faster and to run end-to-end speech synthesis 38 times faster, and Azure text-to-speech integrates it for more than 70 languages and locales and more than 200 voices. Suphx, a 2020 arXiv system, is the first 10-dan mahjong AI based on deep reinforcement learning.

6. 2020

   The Dual Learning monograph

   Springer publishes his monograph Dual Learning in 2020. The book sets out a framework that uses structural duality between AI tasks, with applications in machine translation and speech processing.

7. 2022–2025

   AI for science at Microsoft

   From 2022 to 2025 he is a global research partner at Microsoft Research AI for Science, and as a partner research manager he leads the AI for Science Asia team. In November 2022 he coauthors BioGPT, a domain-specific generative Transformer language model pre-trained on biomedical literature, in Briefings in Bioinformatics. In December 2023 he discusses the protein-ligand method FABind as a senior principal research manager at Microsoft Research.

8. Since 2025

   Vice president in Beijing

   Since 2025 he is vice president of Beijing Zhongguancun Academy. In 2025 a Nature Machine Intelligence paper describes accelerating protein engineering with fitness landscape modeling and RL, and E2Former, an efficient equivariant transformer with linear-scaling tensor products, appears at NeurIPS.