Quentin Garrido

Timeline

1. 2021

   Engineering degree at ESIEE Paris

   He completes an engineering degree in computer science at ESIEE Paris in 2021 and receives an excellence prize from the jury.

2. 2022

   Student paper award at ISMB

   At ISMB 2022, he wins the Ian Lawson Van Toch Memorial Award for Outstanding Student Paper for "Visualizing hierarchies in scRNA-seq data using a density tree-biased autoencoder". The paper is published in Bioinformatics.

3. 2023

   ICLR 2023 honorable mention

   His paper "On the duality between contrastive and non-contrastive self-supervised learning", written with Yubei Chen, Adrien Bardes, Laurent Najman and Yann LeCun, receives an Outstanding Paper Honorable Mention at ICLR 2023. His RankMe paper is presented orally at ICML 2023.

4. 2024

   Coauthoring V-JEPA

   In 2024, he coauthors "Revisiting Feature Prediction for Learning Visual Representations from Video", which introduces V-JEPA, a collection of vision models trained solely with a feature prediction objective. The largest model, a ViT-H/16 trained only on videos, obtains 81.9% on Kinetics-400 and 72.2% on Something-Something-v2.

5. 2025

   Intuitive physics and a PhD

   In February 2025, he is first author of "Intuitive physics understanding emerges from self-supervised pretraining on natural videos", which finds that video models predicting in a learned representation space show an understanding of properties such as object permanence and shape consistency. He defends his PhD thesis, "Self-supervised learning beyond invariant image representations", at Université Gustave Eiffel in May 2025. The PhD comes from FAIR at Meta and Université Gustave Eiffel and is co-advised by Yann LeCun and Laurent Najman.

6. June 2025

   V-JEPA 2 and robot planning

   In June 2025, he is one of 30 authors of V-JEPA 2, a joint-embedding predictive architecture pre-trained on over 1 million hours of internet video. Its action-conditioned version, V-JEPA 2-AC, is post-trained on less than 62 hours of unlabeled robot videos and, deployed zero-shot on Franka arms in two labs, picks and places objects using planning with image goals.

7. 2026

   Latent action world models

   In 2026, he is first author of "Learning Latent Action World Models in the Wild", which studies latent actions learned from in-the-wild videos and finds that continuous, constrained latent actions capture the complexity of actions in such videos. The paper is accepted at ICML 2026, as is a second paper of his, "Interpreting Physics in Video World Models".

8. Since June 2026

   Joining AMI Labs in Paris

   In June 2026, he joins AMI Labs in Paris as a Member of Technical Staff, after working as a Research Scientist at FAIR at Meta. His aim there is to build AI systems that understand, plan and reason in the physical world, drawing on self-supervised learning and world models learned from natural data such as raw videos.