Portrait of Fan Yang

Fan Yang

Principal Researcher/Research Manager


I am a researcher and research manager of System Research team in System and Networking Research Group at Microsoft Research Asia. I joined MSR-Asia in 2004, after receiving my Ph.D. in the department of computer science at Nanjing University, 2003.

My recent research focuses more on machine learning system, including new challenges in large-scale compute cluster for AI, Deep Learning compiler, and AutoML toolkit. I am an architect of the machine learning systems OpenPAI, NNFusion, and NNI, which have been used by Microsoft products like Bing and Azure. As open-source projects, these systems also have a growing number of external users across the world.

In the past, I worked on graph systems. The graph systems I co-developed like GraM set a new speed record for trillion-scale graph analytics and have been used by Bing to improve Ads coverage and web mining. I also contributed to SCOPE, Microsoft’s big data engine. I helped re-architect and optimize the SCOPE-based analytic pipeline of Bing Ads team, improving its processing capacity by more than 50%, which generated tens of millions of dollar of additional revenue.

During the research and development of complex computer systems, I found that Richard Gabriel’s “worse-is-better philosophy” quite helpful, especially the following: “It is slightly better to be simple than correct.” I would love to hear more real-world examples following this philosophy. And if you do and happen to looking for a job, maybe you can consider our team.

To prospective interns

If you consider yourself a good system guy and look for an internship at MSR-Asia, please drop me an email.

If you are interested in systems research and don’t know where to begin with, you could start by reading the papers mentioned here.

If you are looking for advice on your research, Richard Hamming’s talk on “You and Your Research” is a must-read.


e-mail: fanyang AT microsoft DOT com

phone: +86-10-59174296

Personal homepage: https://fanyangcs.github.io