Lab leader

William Fedus

William Fedus, also known as Liam, is a Periodic Labs co-founder. His research has examined generative model training, experience replay in reinforcement learning, and sparse expert networks. With Barret Zoph and Noam Shazeer, he introduced Switch Transformers, which route inputs to selected model components. He later co-authored work on stable expert models and the SimpleQA factuality benchmark. His GPT-4 credit concerns data work within reinforcement learning and alignment. Periodic was founded in 2025 to build AI scientists and laboratories for physical experiments. This connects model training and evaluation with an effort to produce and study measurements from the physical world.

Now at
Periodic Labs
Works
34

Publication numbers from OpenAlex, as of 14 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Photo of William Fedus

Timeline

  1. 2020

    Experience replay

    Fedus co-authored experience replay experiments in 2020.

  2. 2021

    Switch Transformers

    Fedus, Zoph, and Shazeer introduced Switch Transformers in 2021.

  3. 2022

    Making sparse models stable

    Fedus co-authored ST-MoE in 2022.

  4. 2024

    Testing factual answers

    Fedus co-authored SimpleQA in 2024.

  5. 2025

    Periodic Labs

    Periodic Labs was founded in 2025 with Fedus as a co-founder.

  6. As of 2026

    Physical science experiments

    Periodic is developing AI scientists and physical laboratories.

Sources and credits

Photo: William Fedus GitHub profile. License: Not stated by source. Source: github.com. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. Many Paths to Equilibrium arxiv.org. Checked 11 October 2026.
  2. Revisiting Fundamentals of Experience Replay arxiv.org. Checked 11 October 2026.
  3. Switch Transformers arxiv.org. Checked 11 October 2026.
  4. ST-MoE: Designing Stable and Transferable Sparse Expert Models arxiv.org. Checked 11 October 2026.
  5. GPT-4 contributions openai.com. Checked 11 October 2026.
  6. Measuring Short-Form Factuality in Large Language Models arxiv.org. Checked 11 October 2026.
  7. Periodic Labs periodic.com. Checked 11 October 2026.
  8. Periodic Labs company profile linkedin.com. Checked 11 October 2026.