Nicolas Hug

Open-source work

Hug describes the current work as part of the PyTorch team at Meta, focused on preprocessing and performance, and as the primary developer of TorchVision and TorchCodec. PyTorch's governance documentation lists Hug as a maintainer of both libraries: TorchVision provides datasets, transforms and models for computer vision, and TorchCodec handles media decoding and encoding for PyTorch.⁶²³⁴

Before joining Meta, Hug was a postdoc and part of scikit-learn's core development team for several years, and is now listed among its emeritus maintainers. In scikit-learn 0.21 Hug added the histogram-based gradient boosting estimators HistGradientBoostingClassifier and HistGradientBoostingRegressor, an implementation inspired by LightGBM.⁶⁵⁷

During a PhD, Hug developed Surprise, a Python library for building and analyzing recommender systems, which Hug described in a 2020 paper in the Journal of Open Source Software.⁶⁸

Teaching and writing

Hug's blog explains technical topics in machine learning, including a four-part 2017 series on matrix factorization for recommendation, 2019 posts on gradient boosting and on Cython, and a 2022 deep dive into the internals of the FFCV data-loading library.⁹

Work and contributions

• TorchVision Open-source code

  Listed as a TorchVision maintainer on the PyTorch Governance "Maintainers" page; TorchVision provides datasets, model architectures and image transformations for computer vision.²¹³¹⁰

• TorchCodec Open-source code

  Listed as a TorchCodec maintainer on the PyTorch Governance "Maintainers" page; TorchCodec provides media decoding and encoding for PyTorch.²¹⁴¹¹

• scikit-learn Open-source code

  Listed among the emeritus maintainers on the scikit-learn About page.⁵¹²

Timeline

1. Jun 14, 2017Began a four-part blog series on matrix factorization for recommendation.⁹
2. May 2019Added histogram-based gradient boosting estimators to scikit-learn in version 0.21.⁷
3. Jun 1, 2019Published a blog post explaining gradient boosting as gradient descent.⁹
4. Aug 5, 2020Published a Journal of Open Source Software paper on the Surprise recommender systems library.⁸
5. Sep 30, 2022Published a deep dive into the internals of the FFCV data-loading library.⁹

Where they work

• Meta 106 people in the wiki

Related people

• Yuxin WuAlso at Meta
• Yangqing JiaAlso at Meta
• Francisco MassaAlso at Meta
• Ross GirshickAlso at Meta
• Andrea VedaldiAlso at Meta
• Sayak PaulComputer vision · ML frameworks and libraries
• Tong HeML frameworks and libraries · Computer vision
• François CholletML frameworks and libraries
• Łukasz KaiserML frameworks and libraries
• Xiaodong HeComputer vision