Vincent Dumoulin
Research Scientist at LISA, University of Montreal
Montreal, Quebec, Canada
Summary
🤩
Rockstar🎓
Top SchoolVincent Dumoulin is a research scientist based in Montreal with 14 years of hands-on experience building and optimizing machine learning systems and data pipelines. He holds a Ph.D. in Computer Science from Université de Montréal and combines deep academic training with practical contributions at Google and MILA/LISA. His open-source work spans core ML tooling—contributing GPU-optimized kernels to Theano, optimization step rules to Blocks, and dataset handling in Fuel and Meta-Dataset—demonstrating expertise in performance-critical and infrastructure-level code. Vincent also bridges research and engineering through reproducible tutorials and browser-side ML tooling (Magenta.js), showing a knack for making complex models accessible. Notably, he has implemented specialized layers and optimization algorithms (e.g., CumulativeProbabilitiesLayer, AdaDelta and RMSProp support), reflecting both theoretical insight and careful implementation. He’s driven by making ML research usable at scale, from low-level GPU ops to dataset and training orchestration.
13 years of coding experience
1 year of employment as a software developer
Doctor of Philosophy (Ph.D.), Computer Science, Doctor of Philosophy (Ph.D.), Computer Science at Université de Montréal
English, French
Github Skills (44)
algorithm10
unit-testing10
hdf10
subsampling10
gpgpu10
operation10
optimizers10
python10
image-processing10
classification10
testing10
tensorrt10
meta-learning10
tensorflowjs10
machine-learning10
Programming languages (8)
Github contributions (5)
mila-iqia/welcome_tutorials
Oct 2017 - Nov 2017
Various tutorials given for welcoming new students at MILA.
Role in this project:
ML Engineer
ML EngineerContributions:11 commits, 1 PR, 12 comments in 1 month
Contributions summary:Vincent contributed to a TensorFlow tutorial notebook within a repository designed to welcome new students to MILA. The commits show the implementation of core TensorFlow concepts like constants, variables, placeholders, and control flow. The user added sections on gradients, scopes, control flow, and device placement, demonstrating a focus on expanding the tutorial's coverage of TensorFlow's capabilities.
mila-iqia/fuel
Feb 2015 - Jan 2017
A data pipeline framework for machine learning
Role in this project:ML Engineer
ML EngineerContributions:457 commits, 112 PRs, 54 pushes in 1 year 10 months
Contributions summary:Vincent primarily contributed to the development of the `fuel` library, a data pipeline framework for machine learning, with a focus on data loading and processing capabilities for image datasets. Their commits involved adding support for the binarized MNIST and CelebA datasets, improving the HDF5 dataset interface, and adding support for variable length data. They also implemented improvements to the download and conversion scripts for various datasets, which indicates their involvement in data pipeline infrastructure.
pipelinedata-sciencepipeline-frameworkmachine-learningmlops
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.