Matthew Le
AI Research Engineer at Meta
New York, New York, United States
Summary
🤩
Rockstar🎓
Top SchoolMatthew Le is an AI research engineer with a decade of experience building and scaling production-grade machine learning systems at Meta (FAIR) and prior research roles. He led an 11-person engineering team developing generative audio models and infrastructure—authoring a NeurIPS 2023 paper on Flow Matching and shipping a GPU- and Hive-scale feature backfilling pipeline. His work spans core ML research and engineering, from custom CUDA kernels for scalable Hawkes process likelihoods to advancing low-resource MT and maintaining widely used open-source toolkits like fairseq and Poincaré embeddings. Comfortable moving ideas from papers to large-scale training and evaluation, he combines deep research chops with pragmatic systems engineering and a track record of publishing and shipping ML at scale.
11 years of coding experience
6 years of employment as a software developer
Master of Science (M.S) Computer Science, Master of Science (M.S) Computer Science at Rochester Institute of TechnologyBachelor of Science (BS) Computer Science, Bachelor of Science (BS) Computer Science at University of Minnesota
English
Github Skills (12)
transformer-models10
pytorch10
machine-learning10
machine-translation10
artificial-intelligence10
nlp10
python10
natural-language-processing10
evaluation9
eval9
data-loading9
testing6
Programming languages (13)
Github contributions (5)
facebookresearch/poincare-embeddings
Feb 2019 - Sep 2021
PyTorch implementation of the NIPS-17 paper "Poincaré Embeddings for Learning Hierarchical Representations"
Role in this project:
ML Engineer
ML EngineerContributions:1 release, 13 commits, 12 PRs in 2 years 7 months
Contributions summary:Matthew primarily contributed to the project by modifying core training and evaluation scripts, particularly focusing on the `embed.py` file. These changes involved adapting the code to newer versions of PyTorch (1.0 fixes), optimizing data loading, and refactoring the code to use new manifolds. The user also made improvements to the evaluation pipeline. These changes suggest an active role in refining the training process and ensuring model performance.
pytorch
facebookresearch/fairseq
Mar 2019 - Apr 2021
Facebook AI Research Sequence-to-Sequence Toolkit written in Python.
Role in this project:ML Engineer
ML EngineerContributions:22 commits, 9 PRs, 5 pushes in 2 years 1 month
Contributions summary:Matthew primarily contributed to fixing and improving the functionality of the Fairseq toolkit, specifically focusing on machine translation tasks. Their work involved debugging semi-supervised translation processes, addressing loading issues with XLM pretraining, and adjusting configurations to enhance the performance of masked language models. The user also addressed linting errors and made minor adjustments to improve the codebase.
aipythonsequence-to-sequencepytorchartificial-intelligence
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.