PeopleofSuperintelligence

Researcher

Mostafa Dehghani

Mostafa Dehghani is an AI researcher at Google DeepMind whose papers cover Transformer variants, language-model pre-training and large vision models. He earned a Ph.D. at the Universiteit van Amsterdam in 2020, after coauthoring Universal Transformers in 2018 as a doctoral student and Google Research intern. In October 2020 he was one of the authors of the Vision Transformer paper, and in 2022 he coauthored UL2. In 2023, he was first author of ViT-22B, a 22-billion-parameter vision model, and of NaViT, which processes images at native resolution. In April 2026, he talked with Matt Turck on The MAD Podcast about AI automating its own research and development.

Badges

Timeline

  1. 2018

    Universal Transformers

    As a University of Amsterdam PhD student and Google Research intern, he coauthors Universal Transformers with Stephan Gouws, Oriol Vinyals, Jakob Uszkoreit and Łukasz Kaiser in 2018. The model generalizes the Transformer with recurrence and sets a new state of the art on the LAMBADA language modeling task.

  2. 2020

    A PhD from Amsterdam

    He earns a Ph.D. at the Universiteit van Amsterdam in 2020, with Jaap Kamps as a doctoral advisor. His doctoral research focuses on improving learning with imperfect supervision.

  3. October 2020

    The Vision Transformer paper

    In October 2020, he is one of eleven authors of "An Image is Worth 16x16 Words", alongside Alexey Dosovitskiy, Lucas Beyer and Alexander Kolesnikov. The paper introduces the Vision Transformer (ViT), a pure transformer applied directly to sequences of image patches that performs very well on image classification.

  4. 2022

    UL2 and Mixture-of-Denoisers

    In May 2022, he coauthors UL2: Unifying Language Learning Paradigms with Yi Tay, Jason Wei, Denny Zhou and others. The paper proposes Mixture-of-Denoisers, a pre-training objective that combines diverse pre-training paradigms.

  5. 2023

    ViT-22B and NaViT

    In February 2023, he is first author of "Scaling Vision Transformers to 22 Billion Parameters", which presents a recipe for training a 22B-parameter ViT (ViT-22B). In July 2023, he is first author of NaViT, which uses sequence packing to process images of arbitrary resolutions and aspect ratios.

  6. As of October 2026

    AI researcher at Google DeepMind

    He is an AI researcher at Google DeepMind. In April 2026, he talks with Matt Turck on The MAD Podcast about AI automating its own research and development.

Career

  1. NowGoogle DeepMindAI researcher
  2. 2018Google ResearchIntern

Education

  1. 2020University of AmsterdamPhD

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Mostafa Dehghani. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  3. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. [1807.03819] Universal Transformers arxiv.org. Checked 21 October 2026.
  2. Moving Beyond Translation with the Universal Transformer research.google. Checked 21 October 2026.
  3. Mostafa Dehghani - Wikidata wikidata.org. Checked 21 October 2026.
  4. Mostafa Dehghani research.google. Checked 21 October 2026.
  5. [2010.11929] An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale arxiv.org. Checked 21 October 2026.
  6. [2205.05131] UL2: Unifying Language Learning Paradigms arxiv.org. Checked 21 October 2026.
  7. [2302.05442] Scaling Vision Transformers to 22 Billion Parameters arxiv.org. Checked 21 October 2026.
  8. [2307.06304] Patch n' Pack: NaViT, a Vision Transformer for any Aspect Ratio and Resolution arxiv.org. Checked 21 October 2026.
  9. AI is Already Building AI | Google DeepMind’s Mostafa Dehghani | The MAD Podcast with Matt Turck | CastFox castfox.net. Checked 21 October 2026.