Vincent Berges
Researcher At FAIR at Meta
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolVincent Berges is a researcher at Meta FAIR with eight years of experience building and scaling reinforcement learning systems and embodied AI environments. He helped lead development of Unity ML-Agents at Unity Technologies and contributed backend RL components to high-profile open-source projects like facebookresearch/habitat-lab, focusing on reward shaping, sensor configuration, and training metrics. With dual master’s degrees from Ecole Polytechnique and Stanford, his background blends applied mathematics, physics, and management science to tackle both theoretical and production ML problems. He has a track record of improving tooling for reproducible RL training (tensorboard integration, type fixes) and translating research into practical simulation platforms. Based in San Francisco, he combines rigorous academic training with hands-on engineering across simulation, data pipelines, and model evaluation.
8 years of coding experience
5 years of employment as a software developer
Master's Degree, Physics, 3.97/4, Master's Degree, Physics, 3.97/4 at Ecole polytechniqueMaster’s Degree, Management Science, 3.79, Master’s Degree, Management Science, 3.79 at Stanford University
English, French
Github Skills (9)
python10
ai10
reinforcement-learning10
deeplearning-ai9
deep-learning9
pytorch9
deep-reinforcement-learning9
tensorboard8
computer-vision7
Programming languages (3)
Github contributions (5)
facebookresearch/habitat-lab
Dec 2021 - Jan 2023
A modular high-level library to train embodied AI agents across a variety of tasks and environments.
Role in this project:
Back-end Developer / ML Engineer (Likely working on RL components)
Back-end Developer / ML Engineer (Likely working on RL components)Contributions:2 releases, 381 reviews, 228 commits in 1 year 1 month
Contributions summary:Vincent's commits primarily focus on modifying and extending the `habitat-lab` library, particularly concerning reward functions and sensor configurations related to embodied AI agents. They made changes to the `RearrangeReward` measure to ensure positive penalties, implying adjustments to reward shaping within a reinforcement learning context. Additional contributions include incorporating tensorboard logging for metrics and losses, and debugging and fixing typing annotations for gym environments. This indicates experience with RL training procedures, model metrics tracking, and potentially, working with the broader software architecture.
pythonembodied-aideep-reinforcement-learningsimulatorai-agents
Unity-Technologies/ml-agents-dots
Feb 2019 - Oct 2021
Contributions:199 commits in 2 years 8 months
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.