PeopleofSuperintelligence

Researcher

Shunfeng Zhou

Shunfeng Zhou is an AI researcher whose work spans computer-vision training and language-model research. He coauthored studies on knowledge distillation, example weighting and neural architecture search from 2019 to 2021. DeepSeek-AI's 2024 reports named him among the authors of DeepSeek LLM, DeepSeek-V2 and DeepSeek-V3. In 2025, he coauthored a Nature study that shows pure reinforcement learning can incentivize language-model reasoning without human-labelled reasoning trajectories. As of October 2026, he coauthors DeepSeek-AI's DeepSeek-V4.1-Flash report on key-value cache compression.

Now at
DeepSeek
Works
26
Citations
3,110
h-index
9

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2019

    Correlation congruence for distillation

    In 2019, he coauthors “Correlation Congruence for Knowledge Distillation” at ICCV. The framework transfers correlations between examples alongside instance-level information for distillation.

  2. 2020

    Learning to auto weight

    He coauthors Learning to Auto Weight (LAW) in 2020, a framework that learns example-weighting policies from data and finds step-dependent weights during training.

  3. 2021

    Neural architecture search

    At IEEE Transactions on Image Processing, he coauthors Block Proposal Neural Architecture Search in 2021. The method combines an evolutionary block-structure search with a two-stage neural architecture search framework.

  4. January 2024

    Scaling DeepSeek LLM

    In January 2024, he coauthors DeepSeek LLM, a report on scaling laws for open-source models in 7B and 67B configurations.

  5. May 2024

    DeepSeek-V2 mixture-of-experts model

    In May 2024, he coauthors DeepSeek-V2, a 236B-parameter mixture-of-experts model that activates 21B parameters per token.

  6. December 2024

    DeepSeek-V3 technical report

    In December 2024, he coauthors DeepSeek-V3, which has 671B total parameters, activates 37B per token and is pretrained on 14.8 trillion tokens.

  7. September 2025

    DeepSeek-R1 reasoning research

    In September 2025, he coauthors a Nature paper on DeepSeek-R1; it finds that pure reinforcement learning can incentivize language-model reasoning without human-labelled trajectories.

  8. As of October 2026

    V4.1-Flash KV-cache compression

    As of October 2026, he coauthors DeepSeek-AI's September 2026 V4.1-Flash report. It uses cross-layer KV-cache reuse and FP4 caching to reduce the global KV-cache footprint to 890 bytes per token.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Shunfeng Zhou. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes DeepSeek. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  3. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staffIncludes DeepSeek. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staff. We test the estimate against 3,913 named model contributors and dated all-staff records across Google DeepMind, OpenAI and Anthropic, plus six additional frontier-grade labs.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes DeepSeek. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Sources and credits

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

  1. Correlation Congruence for Knowledge Distillation openaccess.thecvf.com. Checked 21 October 2026.
  2. Learning to Auto Weight: Entirely Data-Driven and Highly Efficient Weighting Framework ojs.aaai.org. Checked 21 October 2026.
  3. Block Proposal Neural Architecture Search | IEEE Signal Processing Society signalprocessingsociety.org. Checked 21 October 2026.
  4. DeepSeek LLM: Scaling Open-Source Language Models with Longtermism arxiv.org. Checked 21 October 2026.
  5. DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model arxiv.org. Checked 21 October 2026.
  6. DeepSeek-V3 Technical Report arxiv.org. Checked 21 October 2026.
  7. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning nature.com. Checked 21 October 2026.
  8. DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression arxiv.org. Checked 21 October 2026.