Portrait of Shuxin Zheng

Shuxin Zheng

Principal Researcher


Dr. Shuxin Zheng (郑书新), a Principal Researcher at Microsoft Research, serves as an Industrial Adjunct Lecturer at Chinese Academy of Sciences and is the Lead for Microsoft’s scientific foundation models. He has won multiple world championships in artificial intelligence and has trained the largest scientific foundation model to date. He has published over 20 papers in top-tier scientific journals and AI conferences like Nature Computational Science, and Nature Machine Intelligence, accumulating over 3000 citations. Dr. Zheng regularly serves as a reviewer for premier AI conferences and journals, and as a guest lecturer at Tsinghua University, the Chinese Academy of Sciences, and the Microsoft AI Academy, teaching general courses like “Foundations of Machine Learning Methods and Applications” and “Advanced Machine Learning.”

His representative works include:

  • Graphormer [arxiv (opens in new tab)] [github (opens in new tab)] [blog (opens in new tab)]: the first general-purpose Transformer for graph data, which won the 1st place of KDD Cup 2021 OGB-LSC challenge [kdd (opens in new tab)] [technical report (opens in new tab)] and the 1st Open Catalyst Challenge [link (opens in new tab)], outperforming teams from Google DeepMind, Meta AI Research, and others.
  • General Graphormer (GeG): the world’s largest foundation model (with 22 billion parameters) for molecular science, which can perform multiple scientific tasks across different domains and scales, and serve as a powerful scientific assistant for accelerating scientific discovery.
  • Distributional Graphormer (DiG) [demo (opens in new tab)] [arxiv (opens in new tab)] [blog (opens in new tab)]: a breakthrough model that goes beyond AlphaFold2 by predicting the equilibrium distribution of protein structures, rather than a single structure. DiG is also a disruptive innovation in statistical mechanics, where it uses generative AI technology to revolutionize traditional molecular dynamics simulation or sampling methods.
  • DeepRSM [paper (opens in new tab)]: a joint work with the School of Environment of Tsinghua University, which developed an AI numerical model for regional air quality and climate modeling, and was adopted by China’s “14th Five-Year Plan” as the core technology for controlling air pollution and carbon emission in China.


I’m looking for highly-motivated collaborators all the time. Please contact me if you’re interested.