PeopleofSuperintelligence

Researcher

Dumitru Erhan

Dumitru Erhan is a machine-learning researcher whose work spans visual recognition, neural-network behavior and video generation. He completes a PhD with Yoshua Bengio at the University of Montreal, then works on search ranking at Yahoo Labs before joining Google Brain and the Google Photos team. His papers include Going Deeper with Convolutions, which proposes Inception, and Phenaki, a model for video generation from text prompts. Intriguing properties of neural networks becomes ICLR's 2024 Test of Time runner-up, and Going Deeper with Convolutions receives the 2025 CVPR Longuet-Higgins Prize. As of October 2026, he is a Senior Research Director at Google DeepMind and co-leads Gemini Omni.

Works
54
Citations
105,240
h-index
36

Publication numbers from OpenAlex, as of 14 October 2026.

Photo of Dumitru Erhan

Timeline

  1. 2011

    Studying deep networks

    He completes a PhD at the University of Montreal with Yoshua Bengio, where he works on understanding deep networks.

  2. 2011–2012

    Ranking Yahoo search

    At Yahoo Labs, he works on search ranking in 2011–2012, then moves to Google.

  3. 2014

    Finding fragile predictions

    With coauthors, he reports that small, almost imperceptible input changes can make neural networks misclassify images. The paper becomes the runner-up for ICLR's 2024 Test of Time Award.

  4. 2015

    Visual search at Google

    He works on Google Brain's visual-understanding effort and spends several years building visual-search models for the early Google Photos team ahead of its 2015 launch.

  5. 2015

    Introducing Inception

    He coauthors Going Deeper with Convolutions, which proposes the Inception architecture and evaluates it on ImageNet 2014 classification and detection. CVPR awards the paper the 2025 Longuet-Higgins Prize.

  6. 2021–2023

    Learning from video

    His Google Brain research covers video prediction and model-based reinforcement learning. With coauthors, he develops Phenaki, which generates arbitrary-length videos from sequences of text prompts.

  7. As of October 2026

    Leading generative video research

    At Google DeepMind, he is a Senior Research Director and co-leads Gemini Omni. He has also led Veo and Phenaki, with work on generative video foundation models and world models.

Sources and credits

Photo: dumitru.ca; Dumitru Erhan — Google DeepMind; creator not stated. License: Not stated by source. Source: dumitru.ca. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. APPM Department Colloquium - Dumitru Erhan | Applied Mathematics | University of Colorado Boulder colorado.edu. Checked 14 October 2026.
  2. Dumitru Erhan — Google DeepMind dumitru.ca. Checked 14 October 2026.
  3. ICLR 2024 Test of Time Award blog.iclr.cc. Checked 14 October 2026.
  4. Intriguing properties of neural networks arxiv.org. Checked 14 October 2026.
  5. Dumitru Erhan research.google. Checked 14 October 2026.
  6. Going Deeper With Convolutions openaccess.thecvf.com. Checked 14 October 2026.
  7. IEEE Computer Society Celebrates 2025 Technical Community on Pattern Analysis and Machine Intelligence (TCPAMI) Awardees cvpr.thecvf.com. Checked 14 October 2026.
  8. Phenaki: Variable length video generation from open domain textual descriptions research.google. Checked 14 October 2026.