PeopleofSuperintelligence

Researcher

Zhenda Xie

Also known as 解振达

Zhenda Xie is an artificial intelligence researcher at DeepSeek AI in Beijing. He was a CS PhD student at Tsinghua University from 2018 to 2023, specializing in self-supervised visual representation learning. In 2022 he and coauthors presented SimMIM, a simple framework for masked image modeling, at CVPR, while he was a long-term intern at Microsoft Research Asia. He coauthored the December 2024 DeepSeek-V3 technical report and the September 2025 Nature paper DeepSeek-R1. In January 2026 he was a core contributor and corresponding author of mHC at DeepSeek-AI.

Now at
DeepSeek AI
Works
31
Citations
7,610
h-index
19

Publication numbers from OpenAlex, as of 17 October 2026.

Timeline

  1. 2018–2023

    PhD student at Tsinghua

    He is a CS PhD student at Tsinghua University from 2018 to 2023. His work specializes in self-supervised visual representation learning.

  2. 2022

    SimMIM at CVPR

    In 2022 he and coauthors present SimMIM, a simple framework for masked image modeling, at CVPR. Using ViT-B, the approach reaches 83.8% top-1 fine-tuning accuracy on ImageNet-1K. He is affiliated with Tsinghua University and is a long-term intern at Microsoft Research Asia.

  3. 2023

    Scaling masked image models

    In 2023 he and coauthors present On Data Scaling in Masked Image Modeling at CVPR. The experiments use data from 10% of ImageNet-1K to full ImageNet-22K and models from 49 million to 1 billion parameters.

  4. December 2024

    The DeepSeek-V3 report

    In December 2024 he is a coauthor of the DeepSeek-V3 technical report. The report presents a Mixture-of-Experts language model with 671B total parameters and 37B activated for each token.

  5. September 2025

    DeepSeek-R1 in Nature

    In September 2025 Nature publishes DeepSeek-R1, and he is a coauthor. The paper shows that the reasoning abilities of LLMs can be incentivized through pure reinforcement learning.

  6. January 2026

    Manifold-constrained hyper-connections

    In January 2026 he is a core contributor and corresponding author of mHC: Manifold-Constrained Hyper-Connections at DeepSeek-AI. The framework projects residual connections onto a manifold to restore the identity mapping property.

  7. As of October 2026

    Researcher at DeepSeek

    He is an artificial intelligence researcher at DeepSeek AI in Beijing. He joins the company in July 2023.

Sources and credits

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

  1. Zhenda Xie（解振达） | alphaXiv alphaxiv.org. Checked 16 October 2026.
  2. SimMIM: a Simple Framework for Masked Image Modeling arxiv.org. Checked 16 October 2026.
  3. CVPR 2022 Open Access Repository openaccess.thecvf.com. Checked 16 October 2026.
  4. SimMIM: A Simple Framework for Masked Image Modeling - Microsoft Research microsoft.com. Checked 16 October 2026.
  5. CVPR 2023 Open Access Repository openaccess.thecvf.com. Checked 16 October 2026.
  6. [2206.04664] On Data Scaling in Masked Image Modeling arxiv.org. Checked 16 October 2026.
  7. [2412.19437] DeepSeek-V3 Technical Report arxiv.org. Checked 16 October 2026.
  8. DeepSeek-V3 Technical Report arxiv.org. Checked 16 October 2026.
  9. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning | Nature nature.com. Checked 16 October 2026.
  10. [2501.12948] DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning arxiv.org. Checked 16 October 2026.
  11. mHC: Manifold-Constrained Hyper-Connections arxiv.org. Checked 16 October 2026.
  12. [2512.24880] mHC: Manifold-Constrained Hyper-Connections arxiv.org. Checked 16 October 2026.
  13. zdaxie (Zhenda Xie) · GitHub github.com. Checked 16 October 2026.
  14. Zhenda Xie - Wikidata wikidata.org. Checked 16 October 2026.