AI People Wiki

Builders & open source

Saining Xie

Co-founder and Chief Science Officer (CSO), Advanced Machine Intelligence

  • Co-founder and Chief Science Officer (CSO), AMI Labs1
  • Assistant Professor, Courant Institute, New York University1

Saining Xie is the co-founder and Chief Science Officer of AMI Labs and an Assistant Professor at NYU's Courant Institute. Xie was previously a research scientist at Google DeepMind on the GenAI team and, before that, spent four years as a research scientist at Facebook AI Research (FAIR). Xie co-created diffusion transformers (DiT) with William Peebles and is a co-author of ConvNeXt, MoCo, and Masked Autoencoders; recent work from Xie's group includes Cambrian, a project on multimodal models across images and video.12345

At a glance

Current roles
Co-founder and Chief Science Officer (CSO) +1 more
Sector
Company
Education
NYU
Work listed
2
Honors
4
Sources cited
10
Profile reviewed

Career

Saining Xie spent four years as a research scientist at Facebook AI Research and later worked as a research scientist on the GenAI team at Google DeepMind. Xie is a faculty member at NYU's Courant Institute, where Xie leads a research group of postdoctoral fellows and PhD students, and is co-founder and Chief Science Officer of AMI Labs.1

Research

Xie's architecture work includes the ResNeXt network family, introduced in a 2016 paper, and ConvNeXt, from A ConvNet for the 2020s (2022), which modernized convolutional networks. In representation learning, Xie co-authored Momentum Contrast (MoCo) in 2019 and Masked Autoencoders in 2021.6345

For generation, Xie and William Peebles introduced diffusion transformers (DiT) in Scalable Diffusion Models with Transformers (2022). Projects from Xie's NYU group include Cambrian-1, a fully open, vision-centric exploration of multimodal language models, and Solaris (2026), a multiplayer video world model built in Minecraft.2178

Work and contributions

  • Diffusion Transformers (DiT) Published research

    Co-author with William Peebles of "Scalable Diffusion Models with Transformers" (2022); the official repository credits both authors.29

  • ConvNeXt Open-source code

    Co-author of "A ConvNet for the 2020s" (2022); the ConvNeXt repository is released under the MIT license.310

Timeline

  1. Released the paper introducing the ResNeXt network architecture.6
  2. Co-authored Momentum Contrast for Unsupervised Visual Representation Learning.4
  3. Co-authored Masked Autoencoders Are Scalable Vision Learners.5
  4. Co-authored A ConvNet for the 2020s, which introduced ConvNeXt.3
  5. Released Scalable Diffusion Models with Transformers (DiT) with William Peebles.2
  6. Released Cambrian-1, an open, vision-centric study of multimodal language models.7
  7. Released Solaris, a multiplayer video world model in Minecraft.8

Other documented roles

Roles the sources document that are not confirmed as current.

Honors and recognition

  • NSF CAREER Award · U.S. National Science Foundation1
  • PAMI Young Researcher Award · IEEE PAMI Technical Committee1
  • Marr Prize Honorable Mention · International Conference on Computer Vision (ICCV)1
  • AISTATS Test-of-Time Award · International Conference on Artificial Intelligence and Statistics (AISTATS)1

Where they work

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

10 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. Saining Xie — homepageSaining Xie · Official page · accessed Oct 21, 2026
  2. Scalable Diffusion Models with Transformers (arXiv 2212.09748)arXiv · Paper · published Dec 19, 2022 · accessed Oct 21, 2026
  3. A ConvNet for the 2020s (arXiv 2201.03545)arXiv · Paper · published Jan 10, 2022 · accessed Oct 21, 2026
  4. Momentum Contrast for Unsupervised Visual Representation Learning (arXiv 1911.05722)arXiv · Paper · published Nov 13, 2019 · accessed Oct 21, 2026
  5. Masked Autoencoders Are Scalable Vision Learners (arXiv 2111.06377)arXiv · Paper · published Nov 11, 2021 · accessed Oct 21, 2026
  6. ResNeXt paper (arXiv 1611.05431)arXiv · Paper · published Nov 16, 2016 · accessed Oct 21, 2026
  7. Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs (arXiv 2406.16860)arXiv · Paper · published Jun 24, 2024 · accessed Oct 21, 2026
  8. Solaris: Building a Multiplayer Video World Model in Minecraft (arXiv 2602.22208)arXiv · Paper · published Feb 25, 2026 · accessed Oct 21, 2026
  9. facebookresearch/DiT READMEMeta (GitHub repository) · Repository · accessed Oct 21, 2026
  10. facebookresearch/ConvNeXt README (authors and MIT license)Meta (GitHub repository) · Repository · accessed Oct 21, 2026