Mao Yang

About

I (Mao Yang, 杨懋) am the Deputy Managing Director of Microsoft Research Asia (MSRA), where I oversee the systems and networking research area. My passion is building practical, large-scale distributed systems that power next-generation AI and information retrieval.

Since joining Microsoft Research in 2006 after completing my Ph.D. at Beijing University, I’ve had the privilege of seeing research move from theory to global-scale production. As an architect, I helped design and build several core components for Microsoft Bing, including the “Tiger” flash-based index serving platform and the “Cougar” semantic ranking system, both of which have powered all Bing web queries for over a decade.

My current focus is on creating the foundational AI infrastructure for the era of Large Language Models (LLMs). My team’s research is centered on solving the most critical challenges in large-scale AI:

Scaling LLMs: We are pushing the boundaries of model scale and context. Our work on LongRoPE enables LLMs to process context windows of over 2 million tokens, and projects like WaferLLM explore inference at the wafer scale.

Efficient AI Systems: We build high-performance, efficient systems for training and inference. This includes foundational libraries like lookup tabel based kernel LUT-NN, T-MAC, as well as novel techniques for model quantization (VPTQ) and acceleration.

Enhancing AI Reasoning: We develop novel approaches, like rStar-Math, that empower even small language models (SLMs) to achieve world-class mathematical reasoning capabilities, rivaling models many times their size.

I also led the creation of several impactful open-source projects, including the OpenPAI AI training platform, NNI (Neural Network Intelligence) AutoML toolkit, and MMdnn for model conversion.

I am a doctoral supervisor at the University of Science and Technology of China (USTC) and our team consistently publishes at top-tier conferences, including OSDI, SOSP, NSDI, ICML, and NeurIPS.

杨懋博士现任微软亚洲研究院常务副院长，领导微软亚洲研究院在计算机系统和网络领域的研究工作。

杨懋博士于2006年加入微软亚洲研究院，主要从事分布式系统、搜索引擎系统和深度学习系统的研究、设计与实现。同时领导团队在计算机系统、计算机安全、计算机网络、异构计算、边缘计算和系统算法等方向进行关键技术研究。团队及个人在OSDI、SOSP、NSDI、SIGCOMM、ATC等计算机系统和网络的顶级会议上持续发表多篇论文。团队在研究的同时还注重与实际计算机和网络系统的演进结合，与Azure云计算、Bing搜索引擎系统、Windows操作系统、SQL Server数据库系统以及多个开源社区密切合作。杨懋博士同时还是中国科学技术大学博士生导师。

杨懋博士拥有北京大学计算机体系结构专业博士学位以及哈尔滨工业大学硕士和学士学位。