PeopleofSuperintelligence

Researcher

Feryal Behbahani

Feryal Behbahani is an AI researcher whose work spans reinforcement learning, learning from demonstrations, generative world models and language-model self-correction. After earning degrees at Heriot-Watt University and Imperial College London, she led the learning team at Latent Logic in Oxford and joined DeepMind as a Research Scientist. Her research includes the ViBe and Acme frameworks, work using vision-language models as reward sources, the ICML 2024 Best Paper Genie, and the SCoRe method for language-model self-correction. As of October 2026, she is a Google DeepMind researcher working on Gemini agent post-training.

Badges

Timeline

  1. 2010–2016

    Degrees at Heriot-Watt and Imperial

    She earns a BSc in Information Technology with First Class Honours at Heriot-Watt University in 2010 and an MSc in Artificial Intelligence with distinction at Imperial College London in 2012. In 2016, she completes a PhD in Computing in Imperial's Department of Computing, studying computational neuroscience and machine learning at the Brain and Behaviour Lab. Her doctoral work on Haptic SLAM is a finalist for best paper at EuroHaptics 2016.

  2. 2018

    Learning from traffic video

    Behbahani coauthors Learning from Demonstration in the Wild in 2018, which introduces Video to Behaviour (ViBe) to learn behavior models from unlabelled video of a traffic scene. In June 2018, she co-organises a Workshop on Learning from Demonstrations for high-level robotic tasks at RSS 2018 and serves as a mentor for the WiML Workshop 2018.

  3. 2020

    Acme for distributed reinforcement learning

    In June 2020, she coauthors Acme, a framework for building reinforcement-learning algorithms from modular components and scaling them to larger, more complex environments.

  4. 2020

    DeepMind and Latent Logic roles

    In October 2020, she is a Research Scientist in DeepMind's Machine Learning team led by Nando de Freitas, with research focused on reinforcement learning. Before joining DeepMind, she leads the learning team at Latent Logic in Oxford as a Lead Research Scientist in deep reinforcement learning and learning from demonstrations.

  5. 2023

    Vision-language models as rewards

    With Kate Baumli and Satinder Baveja in December 2023, she coauthors Vision-Language Models as a Source of Rewards, studying how CLIP-family vision-language models can serve as reward sources for reinforcement-learning agents.

  6. 2024

    Genie wins an ICML Best Paper

    She coauthors Genie: Generative Interactive Environments in February 2024, a model trained from unlabelled Internet videos to generate action-controllable virtual worlds; ICML names the paper a Best Paper in 2024. In June 2024, she coauthors a position paper arguing that open-endedness is essential to artificial superhuman intelligence.

  7. 2025

    Language-model self-correction

    At ICLR 2025, she coauthors SCoRe, a multi-turn online reinforcement-learning method that improves language-model self-correction using self-generated data.

  8. As of October 2026

    Gemini agent post-training

    As of October 2026, she is a Google DeepMind researcher working on Gemini agent post-training. In July 2026, she announces a hiring opening for Agent Post-Training Research on the Gemini team.

Career

  1. Since 2020Google DeepMindResearch Scientist
  2. Latent LogicLearning Team Lead

Education

  1. 2016Imperial College LondonPhD
  2. 2012Imperial College LondonMSc
  3. 2010Heriot-Watt UniversityBSc

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Feryal Behbahani. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. 16% of AI career moves in the 2020s went from a university to a companyIncludes Feryal Behbahani. 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.
  3. 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.
  4. 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.

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 Feryal Behbahani yet. Send us a free-licensed photo.

  1. Feryal Behbahani feryal.github.io. Checked 22 October 2026.
  2. Learning from Demonstration in the Wild arxiv.org. Checked 22 October 2026.
  3. Acme: A Research Framework for Distributed Reinforcement Learning arxiv.org. Checked 22 October 2026.
  4. Women in AI: Careers in Research cdt-art-ai.ac.uk. Checked 22 October 2026.
  5. [2312.09187] Vision-Language Models as a Source of Rewards arxiv.org. Checked 22 October 2026.
  6. Genie: Generative Interactive Environments arxiv.org. Checked 22 October 2026.
  7. ICML 2024 Awards icml.cc. Checked 22 October 2026.
  8. Open-Endedness is Essential for Artificial Superhuman Intelligence arxiv.org. Checked 22 October 2026.
  9. Training Language Models to Self-Correct via Reinforcement Learning proceedings.iclr.cc. Checked 22 October 2026.
  10. Google DeepMind Expands Gemini AI Team as Feryal Behbahani Announces Hiring for Agent Post-Training Research meyka.com. Checked 22 October 2026.