PeopleofSuperintelligence

Researcher

Jeremy Bernstein

As of October 2026, Jeremy Bernstein is a machine learning researcher at Thinking Machines Lab in San Francisco, and his work concerns optimizers for training neural networks. He co-wrote signSGD in 2018, a method that transmits just the sign of each minibatch gradient, and NVIDIA named him a 2019 graduate fellow. He submitted his PhD thesis on optimisation and generalisation to arXiv in October 2022. With Laker Newhouse he posted Modular Duality in Deep Learning in October 2024, and a variant of its methods was used to set speed records for training NanoGPT. In September 2025 he published Modular Manifolds on the Thinking Machines Lab blog.

Works
20
Citations
482
h-index
6

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2018

    The signSGD optimizer

    He co-writes signSGD in 2018, a compressed optimization method that transmits just the sign of each minibatch gradient.

  2. 2019

    An NVIDIA graduate fellowship

    NVIDIA lists him among its 2019 graduate fellows from the California Institute of Technology.

  3. October 2022

    A PhD thesis on optimisation

    He submits his PhD thesis, Optimisation & Generalisation in Networks of Neurons, to arXiv in October 2022.

  4. October 2024

    Modular duality in deep learning

    He and Laker Newhouse post Modular Duality in Deep Learning in October 2024, and a variant of its methods is used to set speed records for training NanoGPT.

  5. September 2025

    A post on modular manifolds

    He publishes Modular Manifolds on the Thinking Machines Lab blog in September 2025.

  6. As of October 2026

    Researcher at Thinking Machines Lab

    As of October 2026, he works at Thinking Machines Lab as a machine learning researcher based in San Francisco.

Sources and credits

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

  1. [1802.04434] signSGD: Compressed Optimisation for Non-Convex Problems arxiv.org. Checked 14 October 2026.
  2. ICML Poster signSGD: Compressed Optimisation for Non-Convex Problems icml.cc. Checked 14 October 2026.
  3. Past Graduate Fellows | Research nvidia.com. Checked 14 October 2026.
  4. [2210.10101] Optimisation & Generalisation in Networks of Neurons arxiv.org. Checked 14 October 2026.
  5. jeremybernste.in jeremybernste.in. Checked 14 October 2026.
  6. [2410.21265] Modular Duality in Deep Learning arxiv.org. Checked 14 October 2026.
  7. Modular Duality in Deep Learning arxiv.org. Checked 14 October 2026.
  8. Modular Manifolds - Thinking Machines Lab thinkingmachines.ai. Checked 14 October 2026.
  9. jeremybernste.in jeremybernste.in. Checked 14 October 2026.