Aleksander HolynskiResearch Scientist · Google DeepMindResearcher↗
Researcher · Academic
Robert Geirhos
Google DeepMindStaff Research Scientist
Robert Geirhos showed that image recognition models often rely on texture more than shape.[1] He helped explain why AI systems learn shortcuts that work in training but fail in new settings.[2] He later co-led a study showing that Veo 3 could solve some visual puzzles without being trained specifically for them.[3]
Career & education
Current roleGoogle DeepMind2026-presentStaff Research Scientist
Google DeepMind2026-presentPreviouslyGoogle DeepMind2025-2026Senior Research Scientist
Google Brain2022-2023Research Scientist
University of Tübingen2021-2022Postdoc
Google DeepMind2025-2026Google Brain2022-2023University of Tübingen2021-2022Education
Work and education history is primarily focused on AI-relevant roles and may not be comprehensive.
Citation Trend
Total citations to scholarly work, including research papers, technical reports, books and theses, as recorded by Google Scholar. Each point is a dated snapshot.
citations23,830As of Nov 13, '26
citations+5,480Since last snapshot, Jun 15, '26
References
- ImageNet-trained CNNs are biased towards texturearxiv.org
First-authored study comparing reliance on shape and texture in humans and image recognition models.
- Shortcut learning in deep neural networksdoi.org
Co-led paper describing rules that work under familiar conditions but fail in new settings.
- Video models are zero-shot learners and reasonersvideo-zero-shot.github.io
The authors credit Geirhos as a joint lead and show Veo 3 solving some visual tasks without task-specific training.
Based on public sources. If accuracy is important, verify details using the links or references.
For corrections, updates, or requests to remove information, contact us.
