PeopleofSuperintelligence

Researcher

Heming Cui

Also known as 崔鶴鳴

Heming Cui is a computer scientist whose work builds parallel and distributed software systems for training and serving large AI models. After earning computer science degrees at Tsinghua University and a PhD at Columbia University under Junfeng Yang, he joined the University of Hong Kong in January 2015 and received a 2016 Croucher Innovation Award for the RepBox replication project. His research projects with Huawei and HKU included a blockchain-powered Internet layer from 2018 to 2020, the ParaNAS multi-GPU training project from 2021 to 2023, and the MindPipe four-dimensional parallel training system from 2023 to 2026. In 2023, his Fold3D pipeline-parallel training system became a thousands-GPU subsystem in Huawei MindSpore, and in July 2025 he coauthored the FoldMoE system for Mixture-of-Experts training. As of October 2026, he is an Associate Professor in HKU's Division of Computer Science and leads the HKU Systems Software Lab.

Timeline

  1. By 2015

    Computer science degrees and HKU

    He earns bachelor's and master's degrees in computer science at Tsinghua University and a computer science PhD at Columbia University under supervisor Junfeng Yang. In January 2015, he joins the University of Hong Kong right after receiving his PhD, and in 2015 he coauthors the SOSP paper Paxos Made Transparent with Yang and others.

  2. 2016–2021

    RepBox and a Croucher award

    In 2016, as an Assistant Professor in HKU's Department of Computer Science, he receives a HK$5 million Croucher Innovation Award as principal investigator of RepBox, a transparent state-machine replication project at HKU that runs through 2021. In 2017, he and his coauthors receive a best paper award at ACSAC.

  3. 2018–2020

    Huawei blockchain project and AI workshop

    From 2018 to 2020, he serves as principal investigator of a Huawei Innovation Research Program flagship project on a blockchain-powered trustworthy Internet layer and its decentralized applications. In December 2018, he presents in the artificial intelligence and big data session at the HKU-Huawei Joint Research Workshop.

  4. 2021–2023

    ParaNAS for multi-GPU training

    From 2021 to 2023, he leads ParaNAS, a Huawei Theory Lab flagship project on scalable, reliable, high-precision multi-GPU pipeline-parallel training systems for deep neural networks. He receives the Huawei Theory Lab's Best Collaborating Scientist Medal in 2021 and the HKU Faculty Knowledge Exchange Award in 2022, and his group develops the NASPipe pipeline-parallel training system for neural architecture search.

  5. 2023

    Fold3D in Huawei MindSpore

    In 2023, he coauthors Fold3D, a pipeline-parallel training system for large AI models including Transformers and GPT, which is commercialized as a thousands-GPU parallel-training subsystem in Huawei MindSpore. He also serves as program chair of ACM ChinaSys 2023.

  6. 2023–2026

    MindPipe for large AI models

    From 2023 to 2026, he coordinates MindPipe, an HKU project backed by a HK$4.3 million RGC Research Impact Fund grant in 2023 for high-performance, carbon-efficient four-dimensional parallel training of large AI models. During this period, he also serves as principal investigator on a Next Generation AI China National Key Research and Development Program project on four-dimensional-parallel distributed training infrastructure.

  7. 2025

    FoldMoE and an ICSE award

    In July 2025, he coauthors FoldMoE, an ACL paper on long-sequence Mixture-of-Experts training that overlaps computation across Transformer blocks through attention-MoE pipelining. In 2025, he and his coauthors also receive a best paper award at ICSE.

  8. As of October 2026

    Associate Professor at HKU

    In 2026, he coauthors K2+, a VLDB paper on a multi-region OLTP database with disaggregated storage. As of October 2026, he is an Associate Professor in HKU's Division of Computer Science and leads the HKU Systems Software Lab, where his work includes distributed AI training and serving systems and blockchain systems.

Career

  1. NowUniversity of Hong KongAssociate Professor

Education

  1. 2015Tsinghua UniversityBachelor's and master's degrees in computer science
  2. 2015Columbia UniversityPhD

From the biography.

  1. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionIncludes University of Hong Kong. We estimate where the people shaping AI studied and where their careers led. Massachusetts Institute of Technology leads with 467 former students, including 291 recorded doctorates.
  2. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes University of Hong Kong. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  3. Google has the highest combined h-index among AI organizationsIncludes University of Hong Kong. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 5 of the 25 people with the most-cited recent papers also lead on career citationsIncludes University of Hong Kong. We compare citations to 2023–2025 papers with career citations and h-index, after correcting reviewed publication-date errors: who leads, when they started and where they work.

Sources and credits

Photo: We have no published photo of Heming Cui yet. Send us a free-licensed photo.

  1. Cui, H. cs.hku.hk. Checked 22 October 2026.
  2. Heming Cui i.cs.hku.hk. Checked 22 October 2026.
  3. Croucher Innovation Awards | Croucher Foundation scholars.croucher.org.hk. Checked 22 October 2026.
  4. Five CS academic staff presenting at HKU-Huawei Joint Research Workshop cs.hku.hk. Checked 22 October 2026.
  5. MindPipe: High-performance and Carbon-efficient Four-dimensional Parallel Training System for Large AI Models innowings.engg.hku.hk. Checked 22 October 2026.
  6. FoldMoE: Efficient Long Sequence MoE Training via Attention-MoE Pipelining aclanthology.org. Checked 22 October 2026.