Kaiming He
First author of ResNet, Mask R-CNN and MAE; MIT associate professor and part-time scientist at Google DeepMind
- Associate Professor with tenure, Department of EECS, Massachusetts Institute of Technology1
- Distinguished Scientist (part-time), Google DeepMind1
Kaiming He is an Associate Professor with tenure in the Department of EECS at MIT and also works part-time as a Distinguished Scientist at Google DeepMind, with research in computer vision and deep learning. Kaiming He is first author of the 2015 ResNet paper and of the Mask R-CNN, MoCo, and Masked Autoencoders (MAE) papers, and Kaiming He's site also lists Faster R-CNN among this work on visual perception. Before joining MIT in 2024, Kaiming He was a research scientist at Facebook AI Research (2016–2024) and a researcher at Microsoft Research Asia (2011–2016). Courses taught at MIT include deep learning, computer vision, and deep generative models.12345
Work
ResNet (external site: arxiv.org)Published research
First author, with Xiangyu Zhang, Shaoqing Ren, and Jian Sun, of the 2015 paper that introduced ResNets.21
Masked Autoencoders (MAE) (external site: arxiv.org)Published research
First author of "Masked Autoencoders Are Scalable Vision Learners" (2021) with Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick; the facebookresearch/mae repository, a PyTorch re-implementation of the paper, is released under the CC-BY-NC 4.0 license.56
Mask R-CNN and Detectron (external site: github.com)Open-source code
First author of the Mask R-CNN paper (2017). Detectron, FAIR's Apache-2.0 object detection system that implements Mask R-CNN, lists Kaiming He among the authors in its citation entry.37
Other documented roles
Roles the sources document but that are not confirmed as current.
- Research scientist, FAIR (2016–2024), Meta (Facebook AI Research)1
- Researcher (2011–2016), Microsoft Research Asia1
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.