PeopleofSuperintelligence

Builder

Adam Paszke

Adam Paszke is a research scientist at Google DeepMind who works on Pallas and JAX for GPU and TPU performance engineering. As a computer science and mathematics student, he served on PyTorch's core team and coauthored the 2019 paper describing the library. His later work spans Dex, automatic differentiation, parallel effect handlers, and PartIR, a system for machine-learning partitioning. In October 2026, he presents a talk tracing ten years of machine-learning systems from PyTorch through JAX to Pallas.

Works
26

Publication numbers from OpenAlex, as of 15 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2017

    A student on PyTorch's core team

    By 2017, he studies computer science and mathematics while serving on PyTorch's core team.

  2. 2019

    Describing the PyTorch library

    In 2019, he coauthors the PyTorch paper, which describes an imperative and Pythonic machine-learning library that supports hardware accelerators such as GPUs.

  3. February 2022

    Presenting the Dex language

    In February 2022, he presents Dex, a language for parallel scientific computing, while serving as a senior research scientist at Google. Before Google, he works with Facebook and authors PyTorch.

  4. 2023

    Automatic differentiation for sum types

    His 2023 paper with Gordon Plotkin sets out source-code transformations for automatic differentiation in a language with product and sum types.

  5. September 2024

    Parallel effect handlers

    At ICFP in September 2024, he coauthors a study of parallel algebraic effect handlers, with a formal language model and a Haskell implementation.

  6. 2025

    PartIR partitions machine learning

    In 2025, he coauthors PartIR, a system for composing SPMD partitioning strategies for machine learning, as a Google DeepMind researcher.

  7. As of October 2026

    Ten years of ML systems

    As of October 2026, he is a research scientist at Google DeepMind, where he works on Pallas and JAX to make GPU and TPU performance engineering productive. His October 2026 talk traces ten years of machine-learning systems through PyTorch, JAX, and Pallas.

Sources and credits

Photo: We have no published photo of Adam Paszke yet. Send us a free-licensed photo.

  1. PyData Warsaw 2017 - Adam Paszke pydata.org. Checked 15 October 2026.
  2. PyTorch: An Imperative Style, High-Performance Deep Learning Library proceedings.neurips.cc. Checked 15 October 2026.
  3. Getting to the Point with Dex: Safe Parallel Programming for Scientific Applications cs.ox.ac.uk. Checked 15 October 2026.
  4. Contextual source code AD transformations for sum types (LAFI 2023) popl23.sigplan.org. Checked 15 October 2026.
  5. Parallel Algebraic Effect Handlers icfp24.sigplan.org. Checked 15 October 2026.
  6. PartIR: Composing SPMD Partitioning Strategies for Machine Learning arxiv.org. Checked 15 October 2026.
  7. Program – ASPLOS 2025 asplos-conference.org. Checked 15 October 2026.
  8. Program of ML in PL Conference 2026 conference.mlinpl.org. Checked 15 October 2026.