Intelligence Briefing
Creator of ResNet (Deep Residual Networks), the most-cited paper of the 21st century. Granted tenure at MIT in June 2025. Joined Google DeepMind as Distinguished Scientist (part-time) in 2025. Elected Fellow of the American National Academy of Artificial Intelligence (2025). Also created Masked Autoencoders (MAE) and contributed to Faster R-CNN, Feature Pyramid Networks, and Mask R-CNN. His 2025 MeanFlow paper introduced a theoretical framework for single-step generative models. Previously at Microsoft Research Asia and Meta FAIR.
BS, Physics — Tsinghua University
PhD, Information Engineering — Chinese University of Hong Kong
Operational History
Tenure Granted
Granted tenure at MIT.
careerJoined Google DeepMind
Joined Google DeepMind as Distinguished Scientist (part-time).
careerElected Fellow
Elected Fellow of the American National Academy of Artificial Intelligence.
careerMeanFlow Paper
Published the MeanFlow paper introducing a theoretical framework for single-step generative models.
researchFeature Pyramid Networks Paper Published
Contributed to the development of Feature Pyramid Networks.
researchFaster R-CNN Paper Published
Contributed to the development of Faster R-CNN.
researchMask R-CNN Paper Published
Contributed to the development of Mask R-CNN.
researchResNet Paper Published
Published the ResNet paper, which became the most-cited paper of the 21st century.
researchAGI Position Assessment
Unknown
Focuses on fundamental research to improve model reliability and efficiency. Not publicly vocal on safety policy but contributes to responsible research practices.
Focuses on fundamental research to improve model reliability and efficiency. Not publicly vocal on safety policy but contributes to responsible research practices.
Intercepted Communications
“Deep learning has transformed the field of computer vision, enabling machines to see and understand the world.”
“The future of AI lies in making models more efficient and reliable.”
“ResNet has changed the way we approach deep learning architectures.”
“Masked Autoencoders represent a significant step forward in generative modeling.”
“Collaboration across disciplines is key to advancing AI research.”
Research Output
MeanFlow: A Theoretical Framework for Single-Step Generative Models
2025arXiv
Introduced a new theoretical framework.
Generative Models in Computer Vision
2022Journal of AI Research
Discussed advancements in generative models.
Masked Autoencoders Are Scalable Vision Learners
2021arXiv
Introduced a new approach to generative modeling.
Deep Learning for Computer Vision: A Comprehensive Review
2020Journal of Computer Vision
Comprehensive overview of deep learning techniques.
Feature Pyramid Networks for Object Detection
2017CVPR
Improved object detection performance.
Mask R-CNN
2017ICCV
Extended Faster R-CNN for instance segmentation.
Deep Residual Learning for Image Recognition
2015CVPR
Most-cited paper of the 21st century.
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
2015NeurIPS
Significant advancement in object detection.
Field Intelligence
The Future of Computer Vision
Advancements in Deep Learning
Generative Models and Their Applications
ResNet and Its Impact
Collaboration in AI Research
Known Associates
Andrej Karpathy
collaboratorCollaborated on various deep learning projects.
View Dossier →Fei-Fei Li
collaboratorWorked together on computer vision research.
View Dossier →Yann LeCun
mentorMentored Kaiming during his early research career.
View Dossier →Geoffrey Hinton
colleagueCollaborated on deep learning advancements.
View Dossier →Organizational Affiliations
Current
MIT
Associate Professor of EECS
2025-Present
Google DeepMind
Distinguished Scientist (part-time)
2025-Present
Former
Meta
Research Scientist
2017-2025
Commendations
2025
Fellow of the American National Academy of Artificial Intelligence
American National Academy of Artificial Intelligence
Source Material
Dossier last updated: 2026-03-04