Dumitru Erhan
Senior Research Director at Google DeepMind
San Francisco, California, United States
Summary
👤
Senior🎓
Top SchoolDumitru Erhan is a Senior Research Director at Google DeepMind with 16 years of experience building large-scale deep learning systems that power products across YouTube, Google Photos, Waymo and more. He co-leads Veo, DeepMind’s flagship video-generation effort, and has a long track record of shipping state-of-the-art video and vision models (including Phenaki) from research into production. His work on scalable image annotation, object detection, model-based reinforcement learning and video prediction has been widely adopted across Google and cited in numerous publications. An experienced ML engineer and contributor to the tensorflow/tensor2tensor project, he blends hands-on model engineering with ecosystem-level contributions that accelerate research. Trained at Université de Montréal, he pairs rigorous academic roots with practical product impact, and often focuses on making compute- and data-efficient solutions that scale in real-world systems.
16 years of coding experience
14 years of employment as a software developer
PhD Computer Science, PhD Computer Science at Université de MontréalBSc Electrical Engineering and Computer Science, BSc Electrical Engineering and Computer Science at Jacobs University Bremen
Romanian, French, Russian, German, English
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (5)
machine-learning10
deep-learning10
tensorflow10
python10
reinforcement-learning10
Programming languages (4)
Github contributions (5)
tensorflow/tensor2tensor
Jan 2019 - May 2019
Library of deep learning models and datasets designed to make deep learning more accessible and accelerate ML research.
Role in this project:
ML Engineer
ML EngineerContributions:6 commits in 3 months
Contributions summary:Dumitru primarily contributed to the development and configuration of reinforcement learning models within the tensor2tensor framework. Their commits focused on modifying and adding new configurations for model-based reinforcement learning (RLMB), including adjustments to training steps, and simulation parameters. The user also updated the code to use "true_adam" optimizer across multiple model types. Furthermore, they performed python3 conversions and minor code changes.
pytorchautoencoderdeep-learningmachine-translationreinforcement-learning
doomie/ICML2007
Dec 2019 - Dec 2019
Contributions:1 push, 1 branch in 1 day
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.