Barret Zoph
Barret Zoph is a researcher whose shared work concerns neural network design, training methods, and sparse expert models. With Quoc Le, he introduced neural architecture search, which uses learning to propose model structures. Later papers studied transferring searched designs, sharing parameters to reduce search cost, and self-training for visual tasks. With William Fedus and Noam Shazeer, he introduced Switch Transformers, then led the ST-MoE author list. OpenAI credited him with a training role in reinforcement learning and alignment for GPT-4. Thinking Machines Lab named him chief technology officer in 2025.
- Now at
- Thinking Machines Lab
- Works
- 58
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.
Timeline
-
2016
Searching for a network design
Zoph and Le introduced neural architecture search in 2016.
-
2017
Transferring a learned design
Zoph co-authored NASNet’s transferable architecture research in 2017.
-
2018
Reducing the search cost
Zoph co-authored efficient architecture search in 2018.
-
2021
Switch Transformers
Zoph, Fedus, and Shazeer introduced Switch Transformers in 2021.
-
2022
Stable expert models
Zoph was first author of ST-MoE in 2022.
-
2025
Thinking Machines Lab CTO
Thinking Machines Lab named Zoph chief technology officer in 2025.
Sources and credits
Photo: Barret Zoph; image supplied by Stanford HAI. License: Not stated by source. Source: hai.stanford.edu. Cropped to a square and resized. Ask us to remove or replace this photo.
- Neural Architecture Search with Reinforcement Learning
- Learning Transferable Architectures for Scalable Image Recognition
- Efficient Neural Architecture Search via Parameter Sharing
- Rethinking Pre-training and Self-training
- Switch Transformers
- ST-MoE: Designing Stable and Transferable Sparse Expert Models
- GPT-4 contributions
- Thinking Machines Lab: 2025 indexed team listing