Kaiming He
Associate Professor
Cambridge, Massachusetts, United States
Summary
🤩
Rockstar🎓
Top SchoolKaiming He is an associate professor at MIT with a decade of experience at the intersection of academic research and industry innovation, following an eight-year research tenure at Facebook and earlier work at Microsoft Research Asia. He specializes in computer vision and machine learning, contributing practical engineering improvements—such as Group Normalization support and training-layer freezing—to the widely used Detectron object-detection platform. Based in Cambridge, MA, he blends deep research rigor with hands-on implementation skills, routinely modifying core modeling and configuration code to bridge experiments and deployable systems. His background reflects a pattern of shipping influential research into production-grade open-source tooling, making him comfortable navigating both theoretical advances and engineering constraints. Colleagues describe him as someone who pairs careful experimentation with pragmatic code changes that accelerate adoption of new ideas.
10 years of coding experience
12 years of employment as a software developer
Tsinghua University
Github Skills (10)
mask-rcnn10
configuration-management10
faster-rcnn10
pytorch10
computer-vision10
deep-learning10
resnet10
python10
machine-learning9
object-detection9
Programming languages (4)
Github contributions (5)
facebookresearch/Detectron
Jan 2018 - Jul 2018
FAIR's research platform for object detection research, implementing popular algorithms like Mask R-CNN and RetinaNet.
Role in this project:
ML Engineer
ML EngineerContributions:9 commits, 12 comments, 3 issues in 5 months
Contributions summary:Kaiming contributed to the Detectron research platform by adding support for Group Normalization, modifying the network printing functionality, and fixing an aspect ratio grouping bug. They also implemented changes related to configuration, allowing for the copying of weights and freezing specific layers during training. The user made modifications to the core modeling and configuration files, including implementing new layers and adjusting existing training parameters.
pytorchmaskdetectron2deep-learningretinanet
KaimingHe/deep-residual-networks
Apr 2016 - Jul 2016
Deep Residual Learning for Image Recognition
Contributions:4 commits, 5 PRs, 19 pushes in 3 months
imagenetimage-recognitionefficientnetdeep-learningrecognition
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.