Mahmoud Assran

Scientist, Member of Technical Staff, AMI Labs

Previously: Research Scientist, Meta FAIR

mido dot assran at amilabs dot xyz

I am a Scientist and Member of Technical Staff on the founding team of AMI Labs. Previously, I was a Research Scientist at Meta in Fundamental AI Research (FAIR). I received my PhD from McGill University, advised by Mike Rabbat and supported by a Vanier Scholarship.

I work on self-supervised learning and world models, with a focus on Joint‑Embedding Predictive Architectures (JEPA), which learn abstract representations of the world by predicting in latent space rather than observation space. My work on JEPA (incl. I-JEPA and V-JEPA) has been featured in Quanta Magazine, CNBC, Reuters, Fortune, La Presse, and others. I’ve also spoken about this work at TED.

“We’re all here on this earth to help others; what on earth the others are here for, I have no idea.” — W. H. Auden

Selected Service

Area Chair, NeurIPS
Reviewer, Conferences NeurIPS · ICLR · ICML · CVPR · ICCV · ECCV · AAAI · AISTATS · IFAC World Congress
Reviewer, Journals Science · SIAM Journal on Optimization · TMLR · Automatica · IEEE Transactions on Automatic Control · IEEE Transactions on Pattern Analysis and Machine Intelligence · IEEE Transactions on Control of Network Systems · IEEE Transactions on Signal Processing · IEEE Transactions on Signal and Information Processing over Networks · IEEE Signal Processing Magazine · Journal of Artificial Intelligence Research · ReScience
Reviewer, Books Cambridge University Press · Manning Publications
Organizer, Workshops World Modeling Workshop, ECCV · World Models for the Future of Autonomy, CVPR
Lead, Meta AIM Program The Meta AI Mentorship program offers doctoral students at UC Berkeley, the University of Washington, Mila, and New York University the opportunity to apply for two-year research visits at Meta.

Selected Publications

See Google Scholar for full list of up-to-date publications.

arXiv 2025

V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

M. Assran, A. Bardes, D. Fan, Q. Garrido, …, Y. LeCun, M. Rabbat, N. Ballas

[paper] [code]

TMLR 2024

Revisiting Feature Prediction for Learning Visual Representations from Video

A. Bardes, Q. Garrido, J. Ponce, X. Chen, M. Rabbat, Y. LeCun, *M. Assran, *N. Ballas* Co-last authors

[paper] [code]

ICML 2024

Learning and Leveraging World Models in Visual Representation Learning

Q. Garrido, M. Assran, N. Ballas, A. Bardes, L. Najman, Y. LeCun

[paper]

TMLR 2024

DINOv2: Learning Robust Visual Features without Supervision

M. Oquab, T. Darcet, T. Moutakanni, H. Vo, …, M. Assran, N. Ballas, …, P. Bojanowski

[paper] [code]

CVPR 2023

Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture

M. Assran, Q. Duval, I. Misra, P. Bojanowski, P. Vincent, M. Rabbat, Y. LeCun, N. Ballas

[paper] [code]

ICLR 2023

The Hidden Uniform Cluster Prior in Self-Supervised Learning

M. Assran, R. Balestriero, Q. Duval, F. Bordes, I. Misra, P. Bojanowski, P. Vincent, M. Rabbat, N. Ballas

[paper]

ECCV 2022

Masked Siamese Networks for Label-Efficient Learning

M. Assran, M. Caron, I. Misra, P. Bojanowski, F. Bordes, P. Vincent, A. Joulin, M. Rabbat, N. Ballas

[paper] [code]

ICCV 2021 Oral · Top 3%

Semi-Supervised Learning of Visual Features by Non-Parametrically Predicting View Assignments with Support Samples

M. Assran, M. Caron, I. Misra, P. Bojanowski, A. Joulin, N. Ballas, M. Rabbat

[paper] [code]

ICML 2020

On the Convergence of Nesterov's Accelerated Gradient Method in Stochastic Settings

M. Assran, M. Rabbat

[paper]

NeurIPS 2019

Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning

M. Assran, J. Romoff, N. Ballas, J. Pineau, M. Rabbat

[paper] [code]

ICML 2019

Stochastic Gradient Push for Distributed Deep Learning

M. Assran, N. Loizou, N. Ballas, M. Rabbat

[paper] [code]

Invited Talks

Upcoming

Northeastern University, Multimodal Video Understanding Seminar (2026)
Keynote, Workshop on WORLD: World Representation for Learning and Decision, ICLR (2027)
Keynote, Workshop on Scientific World Models, ICLR (2027)
Keynote, Workshop on New Trends in AI-Generated Media and Security, CVPR (2027)
Keynote, Workshop on World Models and Agents for Physical AI, CVPR (2027)

Selected Previous

TED AI Vienna (2026)
World Modeling Workshop, Mila Quebec AI Institute (2026) [recording]
Keynote, Workshop on World Models: Understanding, Modelling and Scaling, ICLR (2026) [recording]
Carnegie Mellon University, Computer Vision, Language Understanding and Robotics Seminar (2023)
Georgia Institute of Technology, AILA Program (2023)
New York University, CILVR Seminar (2022)
Columbia University, VIGR Seminar (2022)

Podcasts

The TWIML AI Podcast, “V-JEPA: AI Reasoning from a Non-Generative Architecture” (2024) [episode]

Doctoral Students Co-Advised

Co-advised alongside each student's academic advisor.

Gaoyue Zhou, PhD, NYU (2025–2026) Academic advisors: Lerrel Pinto and Yann LeCun
Emily Kaczmarek, PhD, McGill/Mila (2025–2026) Academic advisor: Tal Arbel
Wancong Zhang, PhD, NYU (2025) Academic advisor: Yann LeCun
Artem Zholus, PhD, UdeM/Mila (2024–2026) Academic advisor: Sarath Chandar
Marcel Hussing, PhD, UPenn (2024) Academic advisor: Eric Eaton
Benno Krojer, PhD, McGill/Mila (2024) Academic advisor: Siva Reddy

Open-Source Research Code

vjepa2

Official codebase for V-JEPA 2, a self-supervised video world model for understanding, prediction, and planning.

ijepa

Official codebase for I-JEPA, a non-generative approach for self-supervised learning from images.

jepa

Official codebase for V-JEPA, a method for self-supervised learning of visual representations from video.

msn

Official codebase for Masked Siamese Networks, a self-supervised learning framework for label-efficient image classification.

suncet

Official codebase for PAWS and SuNCEt, semi-supervised methods for learning visual features with limited labels.

sgp

Official codebase for Stochastic Gradient Push, a method for distributed deep learning over peer-to-peer networks.