![Minjia Zhang](https://ppopp.org/getProfileImage/minjiazhang2/829a2d06-baa5-4686-a0b1-1bb4b766a2bc/small.jpg?1723500065000)
Registered user since Mon 12 Aug 2024
I am an assistant professor (tenure-track) at the Grainger College of Engineering Computer Science of the University of Illinois Urbana-Champaign. I am affiliated with the Department of Electrical and Computer Engineering and NCSA at UIUC. Prior to my appointment at UIUC, I had wonderful seven years at Microsoft Research Redmond and WebXT division as a Principal Researcher and technical lead. I have had all sorts of fun of developing highly efficient and cost-effective systems and algorithms, including but not limited to: enabling and accelerating large-scale deep learning training on parallel/distributed/heterogeneous hardware, building ultra-fast inference engine, different types of model compression, large-scale data management. My research works have been published in major venues, including system and high-performance computing conferences (e.g., ASPLOS, NSDI, USENIX ATC, SC), and top-tier machine learning conferences (e.g., ICML, NeurIPS, ICLR). Several of my work has been applied to Microsoft systems and products, such as Bing, Ads, Azure SQL, Windows, etc., leading to significant latency improvement and cost reduction.
At Microsoft, I was an early member of DeepSpeed, an open-source deep learning optimization library that makes training and inference DL models easy, efficient, and effective. DeepSpeed has enabled the training of some of the largest language models in the world, such as Megatron-Turing 530B. It has been widely adopted by both the industry and academia and has become a common backend for various popular DL frameworks such as HuggingFace, PyTorch Lightning, Fairscale, etc. I was also the co-chair of the engineering/scaling group of the BigScience project, contributing to the training of the BLOOM 176B model, which was the world’s largest open multilingual language model. Before DeepSpeed, I drove the DeepCPU project at Microsoft, a DL inference optimization library that brought order-of-magnitude latency and cost reduction to mission-critical production DL models.
Before joining Microsoft, I finished my Ph.D. from the Computer Science Department at Ohio State University in May 2016, where I was a member of the PLaSS group working on building efficient and scalable systems with strong semantics for parallel programs and advised by Prof. Michael D. Bond. Along the way, I spent the summer/fall of 2015, the summer of 2016 at Microsoft Research Redmond, working with Kathryn McKinley, Sameh Elnikety, and Yuxiong He.
I have been serving as area chair of NeurIPS, program committee member for ASPLOS, USENIX ATC, MLSys, IPDPS, AAAI, and reviewers for ICLR, ICML, CVPR, ICCV, ECCV, ECAI, VLDB, etc. I co-organize tutorials and workshops, such as “Mixture-of-Experts in the Era of LLMs: A New Odyssey”, at ICML 2024. I have received several awards including the Distinguished Paper Award and Distinguished Artifact Award in OOPSLA 2015, Microsoft Excellence Awards, and the Honorable Mention of the ICLR 2024 Outstanding Paper Award.
Contributions