Paul R. Barham
Also known as Paul Barham, Paul Ronald Barham
Paul Barham is a systems researcher who has worked on large machine learning infrastructure. He submitted a Cambridge PhD dissertation on devices in a multi-service operating system in June 1996. The TensorFlow paper, published at OSDI in 2016, and the Pathways paper, submitted to arXiv in March 2022, both list him as a co-author. In December 2023, the Gemini report listed him as an infrastructure co-lead. As of October 2026, Google Research lists him under machine intelligence.
- Now at
- Google Research
- Works
- 55
- Citations
- 38,363
- h-index
- 31
Publication numbers from OpenAlex, as of 15 October 2026.
Timeline
-
June 1996
A Cambridge doctorate
He submits a Cambridge PhD dissertation on devices in a multi-service operating system in June 1996.
-
2016
The TensorFlow paper
The TensorFlow paper, published at OSDI in 2016, lists him as a co-author.
-
March 2022
The Pathways dataflow paper
In March 2022, Pathways: Asynchronous Distributed Dataflow for ML is submitted to arXiv with him among its authors.
-
April 2022
The PaLM language model
The PaLM paper, titled PaLM: Scaling Language Modeling with Pathways, reaches arXiv in April 2022 with him as a co-author.
-
December 2023
Infrastructure co-lead on Gemini
He co-leads infrastructure work on the Gemini report, which is submitted to arXiv in December 2023.
-
As of October 2026
Machine intelligence at Google Research
Google Research lists him under machine intelligence.
Sources and credits
Photo: We have no published photo of Paul R. Barham yet. Send us a free-licensed photo.
- Department of Computer Science and Technology – Technical reports: UCAM-CL-TR-403
- TensorFlow: A system for large-scale machine learning
- [1605.08695] TensorFlow: A system for large-scale machine learning
- [2203.12533] Pathways: Asynchronous Distributed Dataflow for ML
- Paul Barham
- [2204.02311] PaLM: Scaling Language Modeling with Pathways
- [2312.11805] Gemini: A Family of Highly Capable Multimodal Models
- Gemini: A Family of Highly Capable Multimodal Models