Adam Paszke

Open-source work

Adam Paszke's GitHub profile describes Paszke as the author of PyTorch and a Research Scientist at Google. PyTorch's license names Paszke in its copyright notice, and the project's maintainers page lists Paszke as an emeritus maintainer of torch.nn, autograd, and multiprocessing.⁵²³

Paszke was the first author of the December 2019 NeurIPS paper describing PyTorch, which explains how the library pairs an imperative, Pythonic programming style with efficient execution on hardware accelerators such as GPUs. Paszke is also named among the authors in the citation entry of JAX, a library of composable transformations of Python and NumPy programs, released under the Apache License 2.0.¹⁴⁶

Research

Paszke's earlier research includes ENet (2016), a deep neural network architecture for real-time semantic segmentation designed to need fewer floating point operations. In 2021 Paszke was first author of "Getting to the Point," which presents a programming language design that treats arrays as memoized functions on typed index sets and describes automatic differentiation that preserves parallelism.⁷⁸

Work and contributions

• PyTorch Open-source code

  First author of "PyTorch: An Imperative Style, High-Performance Deep Learning Library" (2019). The repository's BSD-style LICENSE names Paszke in its copyright notice, and the maintainers page lists Paszke as an emeritus maintainer of torch.nn, torch.autograd, and multiprocessing.¹²³

• JAX Open-source code

  Named among the authors in the "Citing JAX" entry of the JAX README; the repository is released under the Apache License 2.0.⁴⁶

• ENet Published research

  First author of "ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation" (2016), which proposes an efficient network for pixel-wise segmentation in low-latency settings.⁷

Timeline

1. Jun 7, 2016First author of ENet, a network architecture for real-time semantic segmentation.⁷
2. Dec 3, 2019First author of "PyTorch, An Imperative Style, High-Performance Deep Learning Library," presented at NeurIPS 2019.¹
3. Apr 12, 2021First author of "Getting to the Point," on index sets and parallelism-preserving autodiff for array programming.⁸

Where they work

• Google 121 people in the wiki

Related people

• Yangqing JiaAlso at Google
• James BradburyAlso at Google
• Jeff DeanAlso at Google
• Martín AbadiAlso at Google
• Rajat MongaAlso at Google
• Luca AntigaML frameworks and libraries · Training systems
• Brian Van EssenTraining systems · ML frameworks and libraries
• Frédéric BastienML frameworks and libraries · Training systems
• Mu LiML frameworks and libraries · Training systems
• Pieter NoordhuisTraining systems · ML frameworks and libraries